# Citeroot: documentation and guides (full text) > Citeroot shows whether ChatGPT, Gemini and Google AI mention and cite your brand, traces each answer to its sources and turns gaps into reviewable work. This file holds the full text of Citeroot's documentation and guides, for AI systems. The index of the whole site is at https://citeroot.app/llms.txt. Prices and plan limits change: https://citeroot.app/pricing is the authoritative source for them. # Documentation ## Introduction Source: https://citeroot.app/docs/getting-started/introduction Updated: 2026-11-16 > What Citeroot does today, what it deliberately doesn't, and how the pieces fit. Citeroot records what AI answer engines say when your buyers ask questions, traces each answer to its sources, and checks your site for what keeps you out of those answers. ## What it does 1. **Collects real answers.** You write (or generate) a panel of buyer questions. Citeroot asks each one on ChatGPT, Gemini and Google AI Mode (on their own websites, one fresh chat per question) and reads Google's AI Overview, and stores the full answer and every source it cited. Each question also carries its search keyword and a topic. Enterprise can add other platforms, such as Claude and Perplexity, on request. 2. **Measures them.** For each answer it records whether your brand was named (and in what order), whether your domain was cited, whether the answer reads as a recommendation, and which competitors were named. 3. **Traces sources.** Source Map shows which domains the answers lean on, and which cite your competitors but never you. 4. **Audits your site.** Site Audit checks AI-crawler access in robots.txt and scores each page for answer-readiness. 5. **Drafts the next step.** Article Studio writes an article for each question where AI doesn't name you, and Insights (beta) writes up your results. **What to do next** on Home lists the jobs to start with. Every draft is for you to review. ## What it doesn't do - It doesn't contact anyone. It puts an article on your site only if you connect the site and ask for it (see [Publishing](https://citeroot.app/docs/connect/publishing)); otherwise it changes nothing there. - On the standard plans it doesn't track Claude, Perplexity, Microsoft Copilot, Meta AI or Grok. Enterprise can add them on request. - It doesn't know what people ask AI apps, because the apps don't publish it. - It isn't a replica of a logged-in consumer chat. See [Engines and models](https://citeroot.app/docs/measure/engines-and-models). - Analytics and Search Console integrations and webhooks are planned, not built. (The MCP server and a read-only API are live: see [Connect AI apps](https://citeroot.app/docs/connect/ai-apps) and [The API](https://citeroot.app/docs/connect/api).) See [Exports and roadmap](https://citeroot.app/docs/connect/exports-and-roadmap). ## How the docs are organised - **Getting started** — this page, the [quickstart](https://citeroot.app/docs/getting-started/quickstart) and the [concepts](https://citeroot.app/docs/getting-started/concepts). - **Measuring AI visibility** — metric definitions, questions and keywords, sources, variance, AI platforms and the monthly answer allowance. - **Diagnosing your site** — Site Audit, crawler access, articles and insights. - **Connect** — what you can export today and what is planned. - **Account** — trial, billing, security and your data. --- ## Quickstart Source: https://citeroot.app/docs/getting-started/quickstart Updated: 2026-10-27 > From your website address to your first answers in a few minutes. This walkthrough takes you from sign-up to a first baseline and a website health check. ## 1. Create an account Type your website address into the box on the [home page](https://citeroot.app/) and press the button next to it. Citeroot keeps the address through sign-up and starts reading your website as soon as your account exists. You can also [sign up for the free plan](https://citeroot.app/register?plan=free) and add your website afterwards. You start on the Free plan with **no card**. Sign-up asks for no plan: right after it you choose one, and you can stay on Free. If a paid plan comes with a free trial (the [pricing page](https://citeroot.app/pricing) says which), start it on that step or later from Plan & billing: you add a card, nothing is charged until the trial ends, and you can cancel before then. Nothing is ever deleted if you go back to the Free plan. If email confirmation is switched on for your deployment, confirm your address from the email we send: your first check runs without it, but later checks need it. ## 2. Add your website Enter your website address. (If you used the box on the home page, this has already started.) Citeroot does most of the rest by itself, usually in about a minute: 1. **Reads your website.** Citeroot reads your home page and a few key pages (the same ones a visitor would see) and respects your `robots.txt`. 2. **Writes your questions.** The AI writes the questions a customer would ask when they need what you sell: up to 20, or your plan's limit (5 on Free). About four in five name no brand, and about one in five name your business. A question with your name in it is mostly answered from your own website, so you can read the two kinds apart with the question type filter. Each has a search keyword and an intent. 3. **Shows you the questions to check.** Change the wording, remove what doesn't fit, or add your own. Nothing is asked until you press **Start my first check**. 4. **Starts your first check.** It asks every question on all four AI platforms, and scans your website for what keeps AI from reading it. You land on Home, where the answers appear as they arrive. Set-up takes the whole window, with three steps at the top: your business, your questions, your first check. ### A local business with no website Choose **No website** on the first screen. Paste your **Google Business Profile link** (in Google Maps, open your business, press **Share**, then **Copy link**), and enter the business name, what you sell, your country and your city or area. Citeroot writes your questions from that, and you check them the same way. Citeroot keeps the link as your business's address. It reads nothing from Google. A project with no website has no Website health scan, and no cited source is counted as your own site. You can add a website later in the project's settings. **If Citeroot can't fetch your website** (the address is mistyped, the site isn't live yet, it's a placeholder page, or it blocks bots), the page says so and why. You can start again with another address, or tell Citeroot about your business by hand: what you sell, what the business does, who it is for, its name and country. It writes your questions from that. If it could read your website but couldn't tell what you sell, it asks you only that. Questions are written by the AI model your deployment uses. If no model is set up, Citeroot uses plain question templates and says so. You can change everything later under **Questions** and **Settings**. ## 3. Competitors come from the answers Citeroot doesn't guess your competitors. When the first check is back, it adds the brands the answers name most (up to five, each named in at least two answers) and recounts every answer with them. Forums, review sites, publishers and the AI assistants themselves are never added. Home tells you which brands it added; edit the list on **Competitors** or add your own there. After that, checks run by themselves: every week on Free, every day on Grow, Pro and Max. Home shows when the next one starts. See [Checks and allowance](https://citeroot.app/docs/measure/runs-and-allowance). ## 4. Read the results - **Home** shows how often you're named, how often your site is cited, how often you're recommended and your share of voice — each with a 95% interval — plus the questions where you're missing. - **What to do next**, on Home, is a short list of jobs found in your own data, each with its evidence. Examples: let AI crawlers read your site; draft an article for a question where others are named and you are not; join the conversation on Reddit, make a YouTube video or get reviewed on a review site when AI quoted those in answers that named your competitors and not you. Mark a job Done, Snooze it for 7 days or Remove it. - **Questions** lists each question with its intent and how often you're named. Open one for its own page. See [Questions and keywords](https://citeroot.app/docs/measure/prompt-panels). - **Answers** lists every answer. Open one and it appears in a popup over the list, with your brand highlighted and its sources beneath. **Open the full page** shows it on a page of its own. - **Sources** shows which domains the answers cite and, on Pro and Max, the gaps. With fewer than five answers a number is shown but labelled "not enough data yet". Treat early percentages as a first look, not a measurement. ## 5. Check your website Your first scan runs at set-up. Open **Website health** to read it, or to scan again. It reads your home page and as many more pages as your plan allows: 25 on Free, 100 on Grow, 250 on Pro, 500 on Max and 1,000 on Enterprise. You don't need to verify the domain or add anything to your site. See [Site Audit](https://citeroot.app/docs/diagnose/site-audit). ## 6. Keep going Checks repeat on your plan's schedule, so the trend fills in without you. Add competitors, tighten your questions, and use **Article Studio** (articles) and **Insights** (beta) when you know what to fix. --- ## Core concepts Source: https://citeroot.app/docs/getting-started/concepts Updated: 2026-11-16 > Projects, prompts, engines, runs, answers, sources and findings — the vocabulary used throughout Citeroot. ## Project A project is one brand and its domain, with competitors, a list of questions (its panel) and its history. Agencies typically create one project per client (Pro includes five, Max ten). The question limit covers all projects together. ## Prompt and panel A **prompt** is a question a buyer might ask. In the app, prompts are called **Questions**. A **panel** is the set of prompts you track. Prompts are tagged with a **stage**: | Stage | Purpose | | --- | --- | | Problem | Does the engine know the problem you solve? | | Shortlist | Are you named in "best X for Y"? | | Comparison | How do you appear against named rivals? | | Constraint | Do requirements (price, integrations, compliance) favour you? | | Purchase | What is said about you directly? | ## Keyword and intent Each question carries a short search keyword (the phrase it stands for) and its **intent**: what the searcher wants (Info, Commercial, Transactional or Navigational). The AI writes both with the question. See [Questions and keywords](https://citeroot.app/docs/measure/prompt-panels). ## Engine An AI platform (engine) is where Citeroot collects answers: **ChatGPT** and **Gemini** as their own websites answer, **Google AI Mode** and **Google AI Overviews** from Google. Every plan includes all four, and Enterprise can add others, such as **Claude** and **Perplexity**, on request; in Project settings you choose which ones a project uses. [Engines and models](https://citeroot.app/docs/measure/engines-and-models) explains each one. ## Check A **check** asks every active question on every engine of the project once. The first check runs when you add your website; after that, checks follow your plan's schedule: every week on Free, every day on Grow, Pro and Max. See [Checks and allowance](https://citeroot.app/docs/measure/runs-and-allowance). ## Answer The stored result of one prompt on one engine: the full answer text, the model that produced it, the sources it cited (and any it only consulted), the search queries it ran, and what Citeroot measured in it. Answers are kept as collected; Citeroot never rewrites them. ## Allowance Each plan includes a monthly number of answers, because each one is a paid request to our data provider. It covers every scheduled check. See [Checks and allowance](https://citeroot.app/docs/measure/runs-and-allowance). ## Source A domain or page cited in an answer. Sources are typed by kind — owned, competitor, review site, community, editorial, documentation, marketplace, directory, video or social — using a rules-based classifier you can inspect. ## Finding A result from Site Audit: a check on a page that passed, needs attention or failed, with the evidence and a suggested fix. ## Article and insight An **article** is a draft page for your own website that answers one question where AI doesn't name you. An **insight** is a written investigation of your current results. Both are drafts for a person to review. Older content outlines, called briefs, are kept too. See [Articles and insights](https://citeroot.app/docs/diagnose/briefs-and-insights). --- ## Metric definitions Source: https://citeroot.app/docs/measure/metrics Updated: 2026-10-27 > Exactly how named, cited, recommended, position, share of voice and the 95% interval are calculated — and where they can be wrong. Every metric in Citeroot has a stated numerator and denominator. This page is the reference. All of them are computed from **completed** answers; answers that failed are not counted as absences, and they don't use your allowance. ## Named (mention rate) ``` Named = completed answers that name the brand ÷ completed answers ``` An answer names your brand when its text contains your brand name, one of the other names you added, or your domain, as a whole word (so "Fern" does not match inside "Fernhill"). Matching ignores case. Citation markup such as `[1]` or `([site](url))` is removed first, so a brand that only appears inside a link label isn't counted. Matching looks for whole words, so it works best for languages that put spaces between words; in Chinese, Japanese or Thai a brand written directly against other characters may be missed — read those answers rather than relying on the rate. ## Your site cited ``` Cited = completed answers that cite a page on your domain (or a subdomain) ÷ completed answers ``` Only sources the engine actually cited count. Pages it merely looked at ("consulted") are stored but not counted. A competitor's page about you is not an owned citation. ## Recommended ``` Recommended = completed answers that recommend the brand ÷ completed answers ``` This is a **heuristic**. An answer counts when a sentence naming your brand also contains a recommending phrase ("I recommend", "best choice", "well suited", "top pick", "stands out" …) and no negation ("don't recommend", "avoid", "not ideal"). The sentence is stored as the evidence and shown on the answer page. It will sometimes be wrong; read the answer before you act on it. ## Position Position is the order in which brands are **first named**, among the brands you track (you plus your competitors): 1 means you were named before any tracked competitor. It is reported as an average next to the rates, not blended into them. Brands you don't track aren't counted. ## Share of voice ``` Share of voice = your mentions ÷ (your mentions + every tracked competitor's mentions) ``` A mention is each time an answer names a brand, so an answer that names you three times counts three. It needs competitors to be meaningful: add them on **Competitors**. Adding or removing a competitor changes everyone's share. Home's **Who AI names** chart and a question's own page show something simpler: the share of answers that name each brand (an answer counts once per brand). ## Citation share ``` Citation share = citations of your own site ÷ all citations ``` A citation is one site cited in one answer: a site cited twice in the same answer counts once, and all of your own pages and subdomains count as one site. ## Ranks Next to visibility, share of voice and citation share, Questions and Answers show your **rank**: your place among you and the competitors you track (for citations, among all the sites cited). #1 is the most named or most cited. Two brands with the same number share a place, and the next one skips: 1, 2, 2, 4. When nobody is named there is no rank, shown as a dash. ## Where to see them - **Questions → By topic:** visibility, share of voice, average position and citation share for each topic, with your rank in each, and the same for every question inside it. Click a column heading to sort. - **Answers → Brand matrix:** how often each brand is named (or its share of voice), by topic, question or platform. - **Answers → Source matrix** and **Source ranking:** each site's share of the citations, and the ten most cited sites in order, by topic, question or platform. - **Sources:** where AI's citations go by kind of site, and every cited site and page. Each view has a platform menu and a choice of the last 7, 28 or 90 days. A number based on fewer than 5 answers is shown faded: it is too early to read much into it. ## All AI platforms, or one Rates cover **All AI platforms** (ChatGPT, Gemini, Google AI Overviews and AI Mode) unless you pick one from the platform menu or filter. The per-platform bars on Home always show each platform separately. ## 95% intervals Rates are shown with a Wilson 95% interval, which widens when there are few answers. "Named 40%, interval 25–57%" means the data fits anywhere in that range. Intervals describe sampling noise in the answers we collected; they don't capture bias from your choice of questions or engines. ## The AI Visibility score Home shows one number from 0 to 100 and a word: **Low**, **Fair** or **Good**. It is a summary, not a measurement. It combines three of the measures above over the last 28 days: ``` Score = 60% × named + 25% × cited + 15% × recommended ``` Low is under 25, Fair is 25 to 59 and Good is 60 or more. The weights and the bands are Citeroot's own starting choices, not an industry standard, and we may change them once there is more data. We show no score until there are at least 5 answers. A single number hides which signal moved, so use the separate measures, with their ranges, to decide what to do, and read the answers. ## Not built yet Accuracy of what the answer says about you, and excluding individual runs from calculations, are not available. --- ## Questions and keywords Source: https://citeroot.app/docs/measure/prompt-panels Updated: 2026-11-16 > How to choose, add and limit the questions you track, and what a question's keyword and intent tell you. The **Questions** page is the list Citeroot asks AI about. Together your questions are your panel. Next to each one you see its intent, its buying stage and how often AI names you for it. When you add your website, Citeroot writes the first questions for you. ## Design - Start from real language: sales calls, support tickets and what customers type into search. - Cover every stage of the buying journey (Problem, Shortlist, Comparison, Constraint, Purchase). - Write them the way people ask: a short question, or a plain search phrase such as "best accounting software for freelancers". Skip jargon. - Keep most of them free of your business name. People who already know you are not the ones you need to win. See [Questions that name your business](#questions-that-name-your-business). - Use your plan's room well: 5 on Free, 30 on Grow, 150 on Pro, 300 on Max, shared by all your projects. Every question is asked at every [check](https://citeroot.app/docs/measure/runs-and-allowance). The free [buyer-prompt builder](https://citeroot.app/tools/prompt-builder) drafts a balanced starter set you can copy in. ## Adding questions There are three ways to add questions on the **Questions** page. - **Suggest questions.** Say what you sell and who buys it (both start from what Citeroot read on your website), then press **Suggest 10 questions**. The AI model writes questions the way people search. None names your business or contains your web address. You can ask up to six times an hour. On the Free plan, and whenever the AI model is not available, Citeroot fills in standard templates with what you sell instead. - **Add a question.** Type one question and pick its buying stage. - **Paste several.** One question per line. Start a line with a stage and a colon, such as `Comparison:`, to set the stage for that line. Other lines get the stage you pick. [Citeroot Agent](https://citeroot.app/docs/measure/citeroot-agent) can also offer a question to track. Press **Add to my questions** on its answer. Duplicates are skipped, and questions over your plan's limit are left out, with a message saying how many. Each new question gets a search keyword and intent within a minute or so, and is asked at the next check. ## Keyword and intent Behind each question sits a short **search keyword**: the phrase, in lower case, that people type for it. The AI writes it with the question, along with the question's **intent**: what the searcher wants. - **Info**: how to, what is, why. - **Commercial**: best, compare, reviews, alternatives. - **Transactional**: pricing, free, buy, sign up. - **Navigational**: looking for a particular site or brand. Intent is not the same as the buying stage. The stage (Problem to Purchase) is where the buyer is in their journey. Intent describes the search itself. Articles in [Content](https://citeroot.app/docs/diagnose/briefs-and-insights) use the keyword in their title. Citeroot doesn't show search volume. AI apps don't publish what people ask them, so nobody can say how often a question is asked in ChatGPT or Gemini. ## Topics Every question belongs to a **topic**: what it is about, in a few words ("Pricing and plans", "Integrations"). Topics let you read your numbers by theme: where AI already connects you with a need, and where competitors lead. - Questions that **Suggest questions** writes come with a topic. - **Sort into topics** (on the **By topic** view) gives every question without a topic one, in a single AI step. **Sort all again** regroups every question. It is on the paid plans; on Free, set topics by hand. - To set or change one question's topic, open the question and edit **Topic**. The box suggests the topics you already use, so one topic is spelled one way. ## Three views The **Questions** page has three views: - **By topic:** four headline numbers for all your questions (visibility, share of voice, average position and citation share, with your rank against your competitors), then a table with a row per topic. Open a topic to see its questions. Click a column heading to sort. See [Metrics](https://citeroot.app/docs/measure/metrics) for how each number is worked out. - **All questions:** the full list, with how often AI named you, your average position and the buttons to pause or delete. The forms to add questions are here too. - **Searches AI ran:** before answering, an AI platform often searches the web. This list shows the searches it ran for your questions, most frequent first. Pages that answer these phrases have the best chance of being read. ChatGPT shows its searches; Google's AI Overview and AI Mode mostly don't, so most of the list comes from ChatGPT. Each view has a platform menu and a choice of the last 7, 28 or 90 days. ## Question pages Click a question to open its own page. - **At the top:** the question, its intent, its buying stage, its keyword and its topic (which you can change). - **Named**, **Your site cited** and **Recommended:** how many of the last 28 days' answers show each. - **By AI platform:** the same three measures for each platform, with the latest answer from each. Click one to read it in a popup. - **Who AI names for this question:** you and the competitors you track. - **Websites AI uses for it:** the sites its answers cite, with their type and how many answers cite each. - **Latest answers:** the last twelve. Each opens in a popup. - **Draft an article:** shown when AI names you in fewer than half of the last 28 days' answers. It opens [Article Studio](https://citeroot.app/docs/diagnose/briefs-and-insights) on this question. You can pause, resume or delete the question from its page. ## Pausing and deleting Pause a question to keep its history but stop asking it. A paused question is not asked, so it uses none of your answer allowance, but it still takes a place on your list until you delete it. Deleting a question keeps the answers already collected. ## Keep the list stable A panel is only comparable over time if it stays stable. Changing questions changes the numbers, so add new ideas deliberately and read the trend with that in mind. Citeroot doesn't version panels for you; the Questions page shows each question's own named-rate over the last 28 days so you can see which ones moved. ## Questions that name your business A question like "Is Fernhill worth it?" is useful: it shows what is said about you directly. But it is answered from your own website, so it will almost always name you. These questions are included in the totals, so a list with many of them inflates your named-rate. So Citeroot writes them in a fixed mix, at set-up and in **Suggest questions**: about four in five name no brand, and about one in five name your business ("Is Fernhill any good?", "Fernhill reviews"). None uses your web address. Read them apart with the **question type** filter on Home, Questions, Answers, Sources and Competitors: - **Your brand**: questions that name your business. - **Other brands**: questions that name a competitor you track ("alternatives to …"). - **No brand (generic)**: everything else. This is the number that shows whether AI recommends you to people who don't know you yet. ## Limits Questions per plan, for all your projects together: 5 on Free, 30 on Grow, 150 on Pro and 300 on Max; on Enterprise, as many as your agreement sets. A question asked on several AI platforms is still one question: platforms multiply the answers per check, not the number of questions. --- ## Sources Source: https://citeroot.app/docs/measure/sources Updated: 2026-10-27 > Where AI gets its information: citations by kind of site, every cited site and page, and gap lists. **Sources** shows where AI gets the information behind its answers to your questions, over the last 90 days. ## Sources by type The bar at the top splits every citation by the kind of site cited: your own site, competitors' sites, review sites, communities and social sites, video, directories, stores and documentation, and editorial and other sites. A citation is one site cited in one answer, so the shares add up to 100%. Under the bar, each kind shows its share, how many citations and how many different sites. Choose one to list only its sites below; choose it again, or **Show every kind**, to clear it. Two short lines under the bar say how your own site compares with competitors' sites, and which other kind of site AI leans on most, with how often you are named in those answers. **View every type as a table** has all ten types with the numbers behind them. ## Sites and pages The list below has two tabs: - **Sites:** each cited domain, most cited first, with its icon, its type, the share of answers citing it, who is named when it's cited, the AI platforms that cited it and an example page. - **Pages:** each cited page, most cited first. A page counts once per answer, however it was linked (ChatGPT's tracking tag and #fragments are ignored). Each site's icon is fetched by Citeroot from the site itself and kept on Citeroot, so no other company learns which sites you look at. A letter shows when a site has no icon. ## The table Each row of **Sites** is a domain, with: - **Type** — owned, competitor, review site, community, editorial, documentation, marketplace, directory, video or social. Types come from a rules-based classifier (known review sites, forums, video and social hosts, your competitors' domains, and so on). Unknown domains default to editorial, which is sometimes wrong. - **Share of answers citing it** — the share of answers that cite at least one page on the domain. A domain is counted once per answer however many pages it cites. - **Who is named when it's cited** — in how many of those answers you were named, and in how many a competitor was named. - **Engines** that cited it, and a sample page. ## Gap lists A **gap** is a domain, other than your own, that is cited in answers naming a tracked competitor but in **no** answer naming you. Gaps are where being present would most plausibly change what the engines say. Gap lists are included on Pro, Max and Enterprise, and during a free trial of one of them. On other plans Source Map shows how many gaps exist and the first one in full. ## Using gaps well 1. Open the cited page and understand why it is cited. 2. Decide whether there is a legitimate way to be included: a review you can earn, a listing you can claim, a correction to request, a community thread you can join openly. 3. Prioritise sources that appear near Comparison and Purchase questions. Citeroot doesn't contact anyone for you. ## Limits of this view It's built from the answers you collected, so it reflects your panel, not the whole market. Citation behaviour differs between engines and between runs; a source that appears once is weak evidence. --- ## Reading changes and variance Source: https://citeroot.app/docs/measure/variance Updated: 2026-10-14 > How Citeroot separates real movement from run-to-run noise. AI answers vary between runs even for identical questions. Citeroot is built to show that rather than hide it. ## Intervals Every rate carries a Wilson 95% interval that depends on the number of answers behind it. Few answers means a wide interval. ## "Within normal variation" Home compares the **last 28 days** with the **28 days before**. A change is called **up** or **down** only when the two 95% intervals do not overlap. Otherwise it's labelled **within normal variation** — it's not a finding. With fewer than five answers in either window the comparison says there isn't enough data. ## What isn't compared - Citeroot compares across the whole panel and all engines. Use the per-engine and per-stage tables to look at one slice, and remember those have fewer answers each. - It doesn't automatically mark when a provider changes the model behind an engine. The model name is stored on every answer, so you can check it on the answer page. - It can't tell you why a number moved. ## Reading a drop Before assuming a cause, check: 1. Did the number of answers change? 2. Did you change the prompts or competitors? 3. Did the engines or country in Project settings change? 4. Did competitors' mentions rise? 5. Did a crawler-access setting on your site change? See also the blog post [Why your AI visibility score moved](https://citeroot.app/blog/why-your-ai-visibility-score-moved). ## Run-to-run variation as a baseline ## Causality Citeroot records what the answers said and when. It doesn't claim that anything you did caused a movement; that judgement stays with you. --- ## AI platforms Source: https://citeroot.app/docs/measure/engines-and-models Updated: 2026-11-16 > The AI platforms Citeroot tracks, how each answer is collected, and how close it is to what a person sees. Citeroot tracks the AI platforms people use to find businesses, and collects each answer the way that gets closest to what they see. | Platform | Collected from | Shown as | | --- | --- | --- | | ChatGPT | ChatGPT's own website | App view | | AI Overview | The AI Overview above Google's results | Google search | | AI Mode | Google AI Mode | App view or Google search (each answer says which) | | Gemini | Gemini's own website | App view | Every plan, including Free, tracks all four. Answers are collected for your project's country. ## Who collects the answers Two data providers collect the answers: **Bright Data** and **DataForSEO**. Citeroot uses one of them for each platform, and changes it when the other does a better job for that platform. Every answer records the platform, how it was collected, which provider collected it, whether it searched the web (when the platform shows it), the sources it cited and the country used. You see all of that on the answer. A change of provider can move a platform's numbers a little, because the two providers collect in different ways. Answers keep the label of the provider that collected them, so you can always tell. Perplexity and Microsoft Copilot were tracked for a few days in October 2026 and were dropped: the collection did not work reliably for them. Answers collected for them before that stay readable. ## All AI platforms, or one at a time Home and Sources show **All AI platforms** combined by default, and a menu switches to one platform. The Answers list has the same choice as a filter. The per-platform bars on Home always show each platform separately, so you can see where you're strong and where you're missing. ## How close is this to what a person sees? **ChatGPT, AI Mode and Gemini.** Each question is asked on the platform in a fresh chat with no earlier history, from your project's country. So the answer includes the platform's own instructions, model choice and web search. What it can't include is a signed-in person's memory, chat history, custom instructions and settings, so one person's answers can still differ. Where the provider documents it, the answer says it is what a visitor without an account sees. **AI Overview.** What Google shows on its results page in your project's country. Many searches show no AI Overview at all; those are recorded as "no AI Overview shown" and never count as "not named". Answers also vary from run to run, which is why Citeroot shows intervals and compares like with like. Treat the numbers as a consistent measuring stick for **your** questions, not as an exact count of what every person sees. ## How collection works Scheduled checks send questions to the provider and take the answers as they come back. The first check, right after you add your website, is asked in a way that shows answers quickly. Most answers arrive within minutes, and some platforms can take longer. You can leave the page while a check collects. ## Not tracked on the standard plans Claude, Perplexity, Microsoft Copilot, Grok, Meta AI, DeepSeek, Qwen, Mistral, Naver and others. On Enterprise, we can add these platforms for your account on request: [tell us](https://citeroot.app/enterprise) which ones your buyers use. ## Your plan and platforms Every plan, including Free, tracks all four platforms. In Project settings you choose which ones a project uses. --- ## Checks and allowance Source: https://citeroot.app/docs/measure/runs-and-allowance Updated: 2026-11-16 > When Citeroot asks your questions, what a check is, and what counts toward your monthly answers. ## What a check is A check asks every **active** question on every AI platform in the project once. Ten questions on all four platforms make a check of 40 answers. Most answers arrive within a few minutes. You can leave the page: collection carries on without you. Home shows the progress while a check is running. ## When checks happen There is no button to run a check. Checks follow a fixed schedule, so every result is comparable with the last: | Plan | Checks | | --- | --- | | Free | Every week | | Grow, Pro, Max and Enterprise | Every day | - **The first check** runs as soon as you add your website. You don't need to do anything. - **After that,** checks start at the same hour as the first one (on Free, also on the same weekday). Home shows when the next check starts, in your own time zone. - A check starts within 15 minutes of its time. - **Questions you add** are asked at the next check. - A check that can't run at its time (checks were paused, or your email address isn't confirmed yet) waits for the next one. It doesn't run late, so there is never more than one check a day. - You can pause checks for a project in **Project settings**. - Checks skip free accounts that haven't signed in for 30 days, so abandoned accounts don't use the provider's capacity. Signing in starts them again. ## The monthly allowance Each answer is a real, paid request to one of our data providers, so each plan has a monthly allowance. It is the most a month can need: every question, on every platform, at every check of the longest month (31 daily checks, or 5 weekly ones). | Plan | Questions | Answers a month, at most | | --- | --- | --- | | Free | 5 | 100 | | Grow | 30 | 3,720 | | Pro | 150 | 18,600 | | Max | 300 | 37,200 | | Enterprise | Set in your agreement | Set in your agreement | - It resets on the **1st of each month (UTC)**. - Only answers that **succeeded** count. Failed answers and retries are free. - You only run out if questions were changed many times in one month (each new project's first check counts too). Checks then wait until the 1st, and Home says so. - Past answers stay readable whatever your plan. - Article drafts, insights and Citeroot Agent share one monthly allowance of AI credits on the paid plans. See [Trial, plans and billing](https://citeroot.app/docs/account/plans-billing). ## Failures and retries If a provider is rate-limited or briefly down, the answer is retried after a delay that grows each time (three attempts in total). If it still fails, or the provider refuses the question, it's marked failed with the reason and shown on the check. The next check asks again. --- ## Competitors Source: https://citeroot.app/docs/measure/competitors Updated: 2026-11-16 > Add the businesses AI might recommend instead of you, find the ones your answers already name, and see how often each one is named. The Competitors page shows who AI names next to you. It sits in the **Track** group of the menu. ## What the chart shows For each brand, the share of your answers that name it over the last 28 days. A table below gives the count ("named in 23 of 112 answers"), a 95% range, and the change against the 28 days before. We only show "up" or "down" when the two ranges don't overlap. Otherwise it reads "No clear change". With fewer than 5 answers in either window it says there are too few answers to compare. ## Adding competitors There are three ways, all in the **Your list** card. You can track up to 10 competitors per project. Remove one at any time. ### Named in your AI answers **Find competitors in my answers** lists the brands your recent answers recommend that you don't track yet. Each one shows how many answers name it, with a **Track** button that adds it to your list. - On a paid plan Citeroot Agent does this, and it uses a little of your AI credits. Citeroot collects the names in your answers itself; the AI only checks that short list against what your business sells, so a tool you work with or a place name is not listed as a competitor. - On Free, and when your AI credits are used up, the button reads **Look for brands in my answers** and uses no AI: it lists the names your answers repeat in bold or as list labels. - It reads up to your 60 most recent answers from the last 60 days. It needs at least two answers, so run a check first. - A brand is listed only when at least two answers name it. The list shows up to 12 brands, most named first. The counts are made in code, with the same whole-word matching used everywhere else in Citeroot. - Forums and social sites (such as Reddit, Quora and YouTube), review and comparison sites (such as G2 and Trustpilot), publishers, directories, marketplaces, encyclopaedias and AI assistants are never suggested. - The AI model spots the brand names. On the Free plan, and whenever the model is not available, Citeroot uses the names that the answers themselves show in bold or as list labels in at least two answers. That is less precise. - Press **Look again** for a fresh look. You can look up to four times an hour. The page shows when you last looked. ### Suggested by AI - **Find competitors for me.** Our AI helper reads your business and your questions and suggests rivals. We list only businesses whose website responds, skip the ones you already track, and show a short reason for each. You choose which to add. It's limited to six requests an hour, and it is on the paid plans. - **Source links.** When the helper has searched the web for rivals, instead of answering from memory, each suggestion has a **Source** link to a page it found. Open it and check that the business really competes with you before you add it. ### Add your own Type a name, and optionally a website. A website helps us tell a competitor apart from a similar name. ## Things to know - A new competitor shows from your next check. Answers we already collected are not read again. - Suggestions come from an AI model and can be wrong. Check each one before you add it. A brand can be named in answers without being a real rival. - Competitors appear in the Home chart, the Sources "gaps" and the trend line. - When your list is full, remove one to add another. --- ## Citeroot Agent Source: https://citeroot.app/docs/measure/citeroot-agent Updated: 2026-11-16 > An AI agent that reads your Citeroot numbers, searches the web and writes reports and plans you can edit in Docs. Free has a simple chat. **Citeroot Agent** answers questions about your AI visibility and does the work that follows. Ask "Why is a competitor named more than us?", "Which article should I write first?" or "Write a report for my team", and it looks things up, explains what it found and saves anything long to **Docs**. ## Where to find it - **The agent page.** It is in both menus: first in **Agent space**, and under **Start** in **Analytics**. Start from the box at the bottom, a topic, one of the ideas, or a job from your **What to do next** list. - **The chat bar.** At the top of the agent page, the chat's name opens a list of your chats. **New chat** starts a fresh one. - **The side panel.** The **Citeroot Agent** button at the top right of every other page opens a panel on the right. It takes its own space, so the page gets narrower rather than covered. Drag its left edge to make it wider (or focus the edge and use the arrow keys); past about a fifth of the screen it lies over the page. Each message tells it which page you are on, so "this page" or "this answer" means what you are looking at. The chat carries on while you move between pages. - **Recents.** Your latest documents and chats are under **Recents** in the Agent space menu. A document opens in the document panel on the page you are on; **All chats** lists every chat, with rename and delete. ## What it is for Citeroot Agent works on your marketing: being named in AI answers and search, SEO, content, competitors, messaging and channels, and how to use Citeroot. It declines other requests, such as general programming or homework, in a sentence. Small technical pieces that marketing needs are fine: schema markup, a `robots.txt` or `llms.txt` file, tracking links. ## What the agent can do - **Read your numbers.** Visibility by platform and buyer stage, your tracked questions, single answers, competitors, the sources AI cites, your website health scan and your to-do list. It says where each number comes from. - **Search the web and read pages.** For facts outside Citeroot, such as a competitor's pricing page or what ranks for a topic. It says when something came from the web. - **Write documents.** Reports, plans and briefs are saved to **Docs** as drafts. Its reply carries a card for each one; click it to open the document in the document panel and edit it straight away. - **Draft articles and write insights.** Article Studio's **Draft article** and Insights' **Generate insight** hand the job to the agent. It uses the same writers as before (built from your own website and your answers, with anything specific left in [brackets]), and each one uses AI credits, the same ones the agent uses. There is no separate limit for articles or insights. - **Use skills.** A [skill](https://citeroot.app/docs/measure/skills) is a set of instructions for one kind of work: an article, a Reddit post, a YouTube script, a LinkedIn post, an FAQ section. The agent loads the right one before it starts, and every message you send from Article Studio carries the article skill. You can also add one yourself: press the plus button in the message box, choose **Add a skill** and pick up to three for that message. You see a "Skill attached" or "Loaded the skill" step when one is in use. Ask it to change a skill ("make my articles shorter") and it saves your own version. - **Do the legwork for an article.** Ask for an article and the agent searches the web, reads the best pages and passes the real sources to the writer, which names and links them and writes the title tag and meta description. You are left only what you alone know. - **Plan your work.** On the [Plan](https://citeroot.app/app/plan) page, ask it to plan your next 7 days or the month. It puts jobs on the board as suggestions, each with a short line on what it is for; keep or dismiss each one. It can read your board and attach what it writes to a job: ask "write the article for today's planned topic and attach it to the card". On any card, the pen button (or the day) lets you edit the description and move the job to another day. Attach documents to a message and it reads them. - **Read what you attach.** Press the **+** button in the message box to upload files, or to attach documents from Docs. It reads images (PNG, JPEG, GIF, WebP), PDFs and text files (TXT, Markdown, CSV, JSON): up to 4 files a message, 4 MB each. You can also paste a screenshot into the box. Files are kept with the chat so follow-up questions work, and are deleted when you delete the chat. Reading files uses credits like any other message; images and long PDFs use more. - **Suggest next steps.** A question to track, a competitor to add or a job for your list. Nothing changes until you press the button on the suggestion. You can see each step it takes while it works. Press **Stop** to end a reply early; what it wrote so far is kept. ## The document panel Click a document anywhere (a card in a chat, a row in Docs, an article in Article Studio, an insight, a job on the Plan board or a document under Recents) and it opens in the document panel, over the page you are on. Each document you open gets a tab, up to four. **Open full page** shows it on its own page, with a tab for each open document and **All docs** to go back. Escape or the close button hides the panel. On a phone it covers the page; close it to go back. ## Editing documents Documents are edited in place, in the document panel or on the document's own page: - Change the title at the top and type anywhere in the text. - The toolbar has undo and redo, text style (text and three heading sizes), bold, italic, strikethrough, code, links, bulleted, numbered and check lists, quotes, code blocks, dividers and tables. Inside a table you can add or delete rows and columns. - Keyboard shortcuts work too: Ctrl+B (bold), Ctrl+I (italic), Ctrl+K (link), Ctrl+Z (undo). On a Mac, use Cmd. Type `#` and a space for a heading, or `-` and a space for a list. - Changes save by themselves a moment after you stop typing. The line under the title says when it last saved. - **Export** copies the document as Markdown (or as HTML, for articles), or downloads it as a Markdown file. Hold Ctrl (Cmd on a Mac) and click a link to open it in a new tab. ## Credits The agent is included in Grow, Pro and Max, and counts its work in credits: - 1 million tokens read is 1 credit. - 1 million tokens written is 5 credits. - Each web search is 0.015 credits (about 67 searches is 1 credit). A search that fails costs nothing. By our estimate, a quick question uses about 0.01–0.03 credits and a researched report with web searches about 0.1–0.25. The exact amount shows under each reply. | Plan | Credits a month | |---|---| | Grow | 50 | | Pro | 200 | | Max | 400 | | Enterprise | Set in your agreement | Credits reset on the 1st of each month (UTC). You can see what is left at the top of the agent page, in the message box, and in **Settings → Plan & billing**. If credits run out part-way through a reply, it stops there and says so. Editing a document yourself uses no credits. ## On the Free plan Free has no Citeroot Agent and no AI features. It tracks your visibility on every AI platform; the agent, articles and insights come with Grow, Pro and Max. ## Docs **Docs** (in the Agent space menu) keeps everything written for your website in one list: reports and plans from Citeroot Agent, articles from [Article Studio](https://citeroot.app/docs/diagnose/briefs-and-insights), and your insights. Filter by kind or search by title. Click a document to open it in the document panel over the page and edit it there; **Open full page** gives it the whole screen, with a tab for each document you have open. A document the agent wrote links back to the chat it came from. ## Good to know - Replies are written by an AI model and can be wrong. Check anything important on the page it points to. - It works on the project you have open. Switch projects with the menu at the top right. - It never posts, emails or changes anything outside Citeroot. --- ## Weekly summary, alerts and fix kits Source: https://citeroot.app/docs/measure/emails-and-fix-kits Updated: 2026-10-14 > The emails Citeroot sends, how to turn them off, and the ready-to-paste fix for blocked AI crawlers. ## Weekly summary Once a week, Citeroot emails a short summary for each project: your AI Visibility score, how many answers named you with the 95% range, whether that changed from the 28 days before, which competitor was named most, and the next three jobs from your list. It is written from the answers Citeroot collected for your own questions, in code. No AI model writes it and it costs nothing. We send nothing when there are fewer than 5 answers in the last 28 days, because a number from so few answers would mislead you. New projects get their first summary after 3 days. ## Alerts An alert is sent only when a change is clear: the 95% ranges for the two 28-day periods do not overlap. Two things can trigger one: - you are named less often than before; - a competitor is named more often than before and is now ahead of you. You get at most one alert email a week per project, and never the same alert twice while it lasts. ## Turn them off Every email has an unsubscribe link. It opens a page with one button and works without signing in. You can also switch the weekly summary and the alerts on or off, separately, under **Settings → Weekly summary and alerts**. Emails about your account itself, such as password resets, are not affected. ## Fix kit: let AI crawlers read your site When your robots.txt blocks a crawler that AI assistants use to read pages (a search or user-request crawler), the **What to do next** list on Home shows **Let AI crawlers read your site**. It opens a fix kit with: - the crawlers that are blocked, who runs them and what they do; - the existing lines in your robots.txt that cause the block; - the exact lines to add, with a copy button; - how to check the result and run a new website health check. The kit never adds crawlers that collect content to train AI models. Whether to allow those is your choice, so the kit only lists them. Some hosts and security services block AI crawlers on their own, outside robots.txt; the kit says so. Editing robots.txt lets a crawler in. It cannot make an assistant choose to quote you. --- ## How AI describes you Source: https://citeroot.app/docs/measure/how-ai-describes-you Updated: 2026-10-12 > The real sentences AI used when it named your business, with the platform and how each answer was collected. Being named is one thing. How you are described is another. **How AI describes you** shows the sentences from your collected answers that name your business, word for word. ## What you see For each quote: the sentence, the platform (ChatGPT, AI Overview, AI Mode or Gemini), how the answer was collected (for example **App view**), the date, and the question that was asked. **Read the full answer** opens the whole answer and the sources it used. ## What it does not do - It does not rewrite or summarise. Nothing on this page is written by an AI model. Each quote is copied from one answer. Only markdown marks, link addresses and footnote numbers are removed so it reads as plain text. - It leaves out demo answers from the built-in simulator. - It uses the last 90 days of answers to your own questions, newest first, and shows no more than four quotes from one platform so a single platform does not fill the page. - It does not judge whether a description is good or bad. Read the quotes and decide. ## If something is wrong AI describes a business from pages it can read. If a quote is out of date or wrong, look at the sources on the full answer and at your own website: fix the page that says it, or the listing AI cites, and collect fresh answers later. See [Sources](https://citeroot.app/docs/measure/sources). ## The free check and the example report Two public pages show what Citeroot does before you sign up. The **free website check** (`/check`) reads your robots.txt and your home page once, tells you which AI crawlers are blocked and what to fix, and stores nothing. The **example report** (`/example-report`) shows a report for an invented business. Every number on it is made up and it says so. --- ## Skills Source: https://citeroot.app/docs/measure/skills Updated: 2026-11-13 > What a skill is, the skills Citeroot comes with, how to make one your own, import one from a file, or change one by asking Citeroot Agent. A **skill** is a set of instructions for one kind of work: how to write an article, a Reddit post, a YouTube script. Citeroot Agent loads the right skill before it starts, so the work comes out the same good way each time. You find them on the **Skills** page, in Agent space. ## The skills Citeroot comes with | Skill | What it does | | --- | --- | | **Article writing (SEO, AEO, GEO)** | An article for your site: answer first, sections that stand alone, real sources found and linked, and finished notes for publishing. Article Studio always writes with it. | | **Reddit post or comment** | Finds the threads AI quotes for your questions, then writes a useful post or comment as the real person behind the business, with who you are said plainly. It never writes as a fake customer. | | **YouTube video script** | A script that answers the question out loud, with the title, description, chapters and pinned comment. AI reads the text around a video, not the video. | | **LinkedIn post** | A short post from a named person that teaches one thing, with a first comment and ideas for the next posts. | | **FAQ section with schema** | Real questions in buyers' words, answers that start with the answer, and the FAQPage markup ready to paste. | These work as they are. You don't need to add them to use them: ask Citeroot Agent for "a Reddit post about …" and it loads the skill. ## Add a skill to a message Citeroot Agent picks the right skill for most requests. To choose it yourself, press the plus button in the message box, then **Add a skill**, and pick up to three. They show as chips above your message and apply to that message only. If you have your own copy of a skill, that is the one the agent follows. On Article Studio the message box always shows the article skill: every message sent from that page uses it. ## Make a skill your own Open a skill and press **Add to my skills**. You get your own copy. Change it as you like; Citeroot's version stays as it is, and nobody else sees yours. There are two ways to change your copy: - **Edit** it by hand. A skill is plain Markdown: a name, one sentence on when to use it, and the instructions. - **Change with Citeroot Agent.** Say what you want in plain words: "make my articles shorter and friendlier", "always sign Reddit posts with my first name". The agent rewrites the skill and saves your version. While you have your own copy, Citeroot Agent, Article Studio and any AI app you connected use it in place of Citeroot's. **Go back to Citeroot's version** drops your copy. A copy doesn't pick up later improvements to Citeroot's version. If you want those, go back to Citeroot's version and make your changes again. ## Import a skill Press **Import skill (.md)** and choose a Markdown file. Only `.md` files are accepted. The file can start with a short header: ``` --- name: Review reply description: Before you write a reply to a customer review. about: A short, human reply that thanks, answers and never argues. --- # Skill: reply to a review Use this when … ``` `description` says when to use the skill. Without a header, the name comes from the first heading. Importing a file with the same name as one of your skills replaces your version of it. **Download .md** on any skill gives you the file in this format, to keep or to share. ## Write a new one **New skill** opens a blank form. A good skill says when to use it, what to do step by step, and what to hand over at the end. Or ask Citeroot Agent to write one: "make a skill for podcast pitches". ## Good to know - You can have up to 30 skills of your own. Instructions can be up to 20,000 characters. - A skill tells the agent how to do the work. It can't switch off Citeroot's rules: the agent still works only on your marketing, never invents figures or customers, and Article Studio still checks every draft for invented figures and links. - Adding, importing and changing skills is on the paid plans, because the agent and Article Studio are. Everyone can read them. - Published skills in Citeroot's library are written and changed by Citeroot's team. --- ## Site Audit Source: https://citeroot.app/docs/diagnose/site-audit Updated: 2026-11-15 > What the crawl checks, how it protects websites without asking you to verify, and how scores are reported. Site Audit fetches pages from your site and returns page-level findings with the evidence attached. ## Nothing to set up You do not have to prove you own a website to scan it, and you do not add anything to it. Because anyone can point a crawler at a site, Citeroot protects websites in how it crawls instead: - It follows your `robots.txt` for its own user agent, `CiterootBot`, and stops if the homepage is disallowed. - It goes slowly: at most three pages at a time, with a pause between requests. - Each website has a **daily page budget** of 1,000 pages, shared by everyone who scans it. When it is used up, scans for that site wait until the next day. - Scans per project and per account are rate limited. ## What's discovered and checked Citeroot reads your robots.txt and sitemap (including sitemap indexes, compressed sitemaps excepted) to find pages. If the sitemap lists fewer pages than your limit it adds links found on the homepage. It stays on your host and prefers shallower pages. Per-audit page limits: 25 on Free, 100 on Grow, 250 on Pro, 500 on Max, 1,000 on Enterprise. | Group | Examples | | --- | --- | | **Indexing** | noindex directives, canonical references, response codes, redirects | | **Structure** | titles, descriptions, H1s, heading order, language | | **Content** | main text present in the raw HTML, opening answer, alt text, internal links | | **Markup** | structured data (JSON-LD) types and syntax, Open Graph | | **Crawler access** | robots.txt for 20 AI user agents, by purpose, and `llms.txt` | Pages are read as **raw HTML**, as most retrieval systems do. Pages that depend entirely on client-side scripts will look empty — which is itself a finding. ## Scores and findings Each page gets a score from its checks; the site score weights shallower pages more. Findings are grouped by check across pages, so the template issue affecting 40 pages appears once with a count. Each check shows pass, needs attention or fail, the evidence and a suggested fix. Comparing two audits shows whether the score moved. ## A polite crawler The crawler identifies itself as **CiterootBot**, honours robots.txt for its own user agent, caps speed and page size, only fetches public addresses, and re-checks every redirect. If your site blocks it, the audit says so and tells you how to allow it — which is also a hint about whether AI crawlers can get in. See [/bot](https://citeroot.app/bot). ## Try it free The [page-readiness audit](https://citeroot.app/tools/page-readiness) runs the single-page checks on any URL with no account. --- ## Crawler access Source: https://citeroot.app/docs/diagnose/crawler-access Updated: 2026-10-08 > How Citeroot reports AI crawler permissions by purpose and what the statuses mean. Site Audit tests your robots.txt against 20 AI-related user agents, grouped by what each is for. ## Purposes - **Search & answer indexing** — crawlers that build indexes assistants retrieve from. - **User-triggered fetches** — requests made because a person asked an assistant about a page. - **Training & model improvement** — crawlers and robots tokens that control training use. ## Statuses - **Allowed** — the tested paths are allowed. - **Partly blocked** — some tested paths are disallowed. - **Blocked** — all tested paths are disallowed. If robots.txt can't be fetched at all (for example your server returns an error for it), the whole report says it **couldn't be assessed** rather than guessing. That is never shown as "allowed". Each result shows the rule that matched, written as `Allow: /path` or `Disallow: /path`, and the user-agent group it came from. ## Matching rules Evaluation follows RFC 9309: the most specific user-agent group applies; within a group the longest matching path wins; Allow beats Disallow when equally specific; `*` and `$` wildcards are supported. ## Beyond robots.txt robots.txt isn't the only gate. CDN or WAF rules, rate limits, login walls and client-side rendering can all prevent access, and Citeroot only sees what its own crawler is served. If CiterootBot receives a 401, 403 or 429, the audit reports it as a possible block. ## Free checker Try the [AI crawler checker](https://citeroot.app/tools/ai-crawler-checker) without an account, and read the [crawler access guide](https://citeroot.app/blog/ai-crawler-access). --- ## Articles and insights Source: https://citeroot.app/docs/diagnose/briefs-and-insights Updated: 2026-11-15 > Article drafts for the questions where AI doesn't name you, written insights, and how to treat both. Both features turn your data into a written draft for you to review. **Articles** are in **Article Studio**. **Insights** are on the **Insights** page, which is still marked beta. ## Articles **Article Studio** lists the questions where AI names you in fewer than half of the answers, most searched first. For each one you see how often you are named and which of your tracked competitors AI names instead. Press **Draft article** to get an article for your own website. If you arrive from a question's page or from a job on Home, that question's row is highlighted and its **Draft article** button is the filled one. ### What goes into an article Citeroot gathers the facts first. The AI model then writes from them: - your business profile, as read from your website when you set up; - text read from your own website; - the question, its search keyword and its intent; - what AI answers to that question say today: who they name, and which websites they quote; - the searches AI ran while answering the question (when the platform shows them), other questions you track on the same topic, and the points current AI answers cover. These decide the sections and the FAQ. The writer may use general knowledge about the topic. About your business it may use only those facts, so it describes you only as your own site does. It must not invent prices, figures, customers, quotes or links. It may name your competitors, but it never states their prices, features or flaws as fact. ### The article skill Every article is written with Citeroot's **article skill**: one set of rules for a page that ranks in search (SEO), is lifted as the answer (AEO) and is quoted by AI assistants with your business named (GEO). The same skill is used by Article Studio, by Citeroot Agent when it writes or improves an article, and by an AI app you connect through the [MCP server](https://citeroot.app/docs/connect/ai-apps). What it asks for: - **Answer first.** The opening paragraph answers the question in about 40 to 60 words, in sentences that make sense when quoted alone. - **Sections that stand alone.** Each heading is a follow-up question or a plain label, and each section opens with its answer. Steps are numbered, options are compared in a table. - **The follow-up questions on the same page.** AI search turns one question into several searches, so the article covers them, in its sections and in a short FAQ. - **Real sources, found for you.** Figures, quotes and sources are what AI quotes most, but only real ones. When you ask Citeroot Agent for an article (the **Draft article** button does), it searches the web first, reads the most useful pages and gives the writer what it found. Each fact is checked against the page it came from before it is used, and the article names and links the source. Nothing is invented. - **As little left for you as possible.** What stays in [square brackets] is only what you alone know: your price, a customer story, a quote from a person on your team. Five at most. - **Plain, exact words.** No padding, and no repeating the keyword to fill space. - **Notes for publishing, written out.** After the article, a **Before you publish** section gives the title tag, the meta description and the web address as finished text, then what only you can add, the author and date line, the schema to use and the pages on your site to link. Delete it before the page goes live. You can make the skill your own: see [Skills](https://citeroot.app/docs/measure/skills). The skill promises no ranking. It follows what Google and Microsoft publish about AI search and what tests on generative engines found, and it keeps Citeroot's own rule: nothing invented. ### What a draft looks like - A title that contains the search keyword, then a first paragraph that answers the question directly in about 40 to 60 words. - Four to seven sections, with short paragraphs, lists where they help, and a table when the question compares options. - A **Frequently asked questions** part with three to five questions, then a short invitation to the reader. - Usually 1,000 to 1,600 words, in plain words. - Anything specific that Citeroot cannot know, such as a price or a customer example, is left in [square brackets] for you to fill in. - A **Before you publish** section at the end, for you. ### Checks and the fallback Citeroot checks every AI draft before you see it. A draft is rejected if it has figures that are not in the facts, links other than to your own site and to the sources that were found and checked, web addresses that are not in the facts, HTML or images, no title, too few sections or no FAQ part, too few or too many words, or if it was cut off. A rejected draft is sent back once with the exact problems to fix. If it still breaks the rules, or the AI model is not available, you get a structured outline instead: the right headings, with your parts left in [square brackets] to fill in. An AI-written draft says which model wrote it, for example "Written with Gemini" or "Written with Claude". A draft written without the AI is labelled "Draft". ### Using an article - Read it before you publish. Fill in anything in [square brackets], check every claim about your business and add your own examples. Then publish it on your own site, at its own address. Citeroot publishes nothing unless you connect your website: then you can send the article there as a draft or live (see [Publishing](https://citeroot.app/docs/connect/publishing)). - Copy it as **Markdown** or **HTML**, or download it as a .md file. - Change the title, text or status (Draft, Ready or Archived) under **Edit the title, text or status**. - **What this article was built from** shows the question, what AI says today, the websites AI quotes and whether Citeroot could read your website. - To write about a topic of your own, use **Write about your own topic** in Article Studio. You can say what it should cover. ### Allowance Articles use AI credits, the same ones Citeroot Agent and insights use. There is no separate limit for articles. Free has no credits, so articles are on the paid plans. Credits reset on the 1st (UTC), and Article Studio shows how many you have left. An article written by an AI app you connected through the MCP server uses none. Where email confirmation is required, you need to confirm your address first. ### Older outlines Before articles, Article Studio (then called Content) wrote outlines called briefs. They are kept. They show in your list as **Outline**, and you can still open, edit, copy and delete them. ## Insights An insight is an investigation of the current state: where you are named, where competitors win, what to fix first. 1. Citeroot computes the facts in code: rates with intervals, the weakest questions, stages and platforms, competitor leads, source gaps and the latest website scan's top issues. 2. Those facts, and only those, are handed to the writer, with an instruction not to invent numbers, quotes, competitors or URLs. 3. You get a Markdown draft you can edit, copy or download. The facts table is shown next to the narrative so you can check one against the other. An insight needs at least five collected answers. Insights use AI credits, the same ones articles and Citeroot Agent use, so they are on the paid plans; there is no monthly count. ## Who writes the drafts An AI model writes the drafts when AI is switched on for your Citeroot. Without it, you get a plainer template built from the same facts. ## Read before you use A draft is a starting point. Check claims against your own product, don't publish text you haven't read, and remember that Citeroot only knows what the answers, your website and the scans showed it. --- ## Exports and roadmap Source: https://citeroot.app/docs/connect/exports-and-roadmap Updated: 2026-11-15 > What you can take out of Citeroot today, and the integrations that are planned but not built. ## Available today - **Answers as CSV** (Grow, Pro, Max and Enterprise) — from the Answers page, honouring the filters you've set. One row per answer with the engine, model, stage, question, whether you were named, your position, whether your site was cited, whether you were recommended, competitors named, cited domains and URLs, and the full answer text. Spreadsheet-formula characters at the start of a cell are neutralised. - **Everything as JSON** (every plan) — Settings → **Download my data** exports your profile, projects, questions, runs, every collected answer, website scans, articles and older outlines, and insights. - **Your own AI app** (Grow, Pro, Max and Enterprise) — connect Claude, ChatGPT or another app to your account through Citeroot's MCP server. See [Connect AI apps](https://citeroot.app/docs/connect/ai-apps). - **Your numbers as JSON, through the API** (Grow, Pro, Max and Enterprise) — for dashboards and scripts. Read-only. See [The API](https://citeroot.app/docs/connect/api). - **A report as a link or a PDF** (Pro, Max and Enterprise) — see [Your account](https://citeroot.app/docs/account/your-account). - **Articles to your website** (Grow, Pro, Max and Enterprise) — send an article to WordPress, Shopify or a webhook of your own, as a draft or live. See [Publishing](https://citeroot.app/docs/connect/publishing). - **Articles as Markdown or HTML** — copy any article as Markdown or HTML, or download it as a .md file. Older outlines can be copied or downloaded as Markdown. ## Planned, not built These are on the roadmap and **don't exist yet**. Nothing in the product depends on them. | Planned | What it would do | | --- | --- | | Google Search Console and GA4 | Bring in the queries that reach your pages and the visits assistants send. | | Bing Webmaster Tools | The same for Bing. | | Event webhooks | Signed delivery when a check finishes or a finding appears. (A webhook for articles exists today.) | | Slack, Looker Studio, Linear and Jira | Alerts, a ready-made reporting connector and task creation. | | Roles and single sign-on | Different rights for different members of a team. | To tell us which matter to you, join the waitlist on the [integrations page](https://citeroot.app/platform/integrations), or [contact us](https://citeroot.app/contact). --- ## Connect AI apps (MCP server) Source: https://citeroot.app/docs/connect/ai-apps Updated: 2026-11-17 > Use your Citeroot data inside Claude, ChatGPT and other AI apps: how to connect, what a connected app can do, and how to disconnect it. Citeroot has an **MCP server**. MCP (Model Context Protocol) is the standard way an AI app reaches an outside service. Connect your AI app once, and you can ask it things like "write an article for my weakest question", "plan my week" or "which sources cite my competitors and not me?" It answers from your Citeroot account. The app's own AI does the thinking and the writing. Citeroot gives it your data and Citeroot's instructions, the same ones Citeroot Agent works from. So a connected app uses **none of your AI credits**. It is on the paid plans (Grow, Pro, Max and Enterprise). ## How to connect Open **Settings**, then **Connect AI apps**. Copy your Citeroot address. It ends in `/mcp`. The steps for each app are below. The menus in these apps change now and then: look for "connector", "app" or "MCP server". In every case the app opens Citeroot in your browser, and you sign in and press **Allow**. ### Claude 1. Open Settings, then Connectors, then **Add custom connector**. 2. Paste your address and press **Connect**: ``` https://citeroot.app/mcp ``` 3. Sign in to Citeroot and press **Allow**. ### ChatGPT 1. Open Settings, then **Apps and connectors**. Turn on developer mode if it asks. 2. Create a connector, paste your address (`https://citeroot.app/mcp`) and choose **OAuth**. 3. Sign in to Citeroot and press **Allow**. ### Claude Code 1. Run this in your terminal: ``` claude mcp add --transport http citeroot https://citeroot.app/mcp ``` 2. Run `/mcp`, choose **citeroot** and authenticate. Sign in to Citeroot and press **Allow**. ### Cursor 1. Add this to `~/.cursor/mcp.json`, or add the server from Cursor's MCP settings: ```json { "mcpServers": { "citeroot": { "url": "https://citeroot.app/mcp" } } } ``` 2. Sign in to Citeroot and press **Allow**. ### VS Code 1. Add this to `.vscode/mcp.json`, or run **MCP: Add Server** and choose HTTP: ```json { "servers": { "citeroot": { "type": "http", "url": "https://citeroot.app/mcp" } } } ``` 2. Start the server when VS Code offers. Sign in to Citeroot and press **Allow**. ### Codex 1. Run these in your terminal: ``` codex mcp add citeroot --url https://citeroot.app/mcp codex mcp login citeroot ``` 2. Sign in to Citeroot in your browser and press **Allow**. ### Other apps Add a remote MCP server with your address. Most open Citeroot for you to press Allow. ### Then ask it for something Try "Write an article for my weakest question." The app reads the brief and Citeroot's article skill, writes the article and saves it to Docs as a draft. If Claude shows a set-up screen with choices, pick **Sign in now**. For the OAuth client, either **Use Claude's published identity** or **Register automatically** works. Leave the request headers empty. The app then opens Citeroot. Sign in if you are asked, read what the app will be able to do, and press **Allow**. You go back to the app, connected. Only press Allow when you started the connection yourself, in an app you trust. The page names the app and where you will be sent back to. ### Apps that want a token Some apps take a token in a header instead of opening a sign-in page. On **Connect AI apps**, type a name and press **Make a token**. Copy it at once: it is shown one time. Send it as `Authorization: Bearer `. A token gives the same access as a connected app and does not expire. Treat it like a password, and remove it when you stop using it. ## What a connected app can do | It can | Tool | | --- | --- | | List your projects | `list_projects` | | Read how often AI names, cites and recommends you, with ranges and the change | `get_visibility_overview` | | Read your tracked questions and how each one does | `list_questions` | | Read AI answers, as excerpts or one in full | `get_answers` | | Read your competitors' share of answers | `get_competitors` | | Read the websites AI cites, and the gaps | `get_sources` | | Read your to-do list and your website health scan | `get_todo`, `get_site_health` | | Read your Plan board and your documents | `get_plan`, `list_docs`, `read_doc` | | Load Citeroot's skills, such as the article skill | `list_skills`, `get_skill` | | Get the brief for an article: your site's facts, what AI says today, and the article skill | `get_article_brief` | | Save a document it wrote to Docs, as a draft, or update one | `save_doc`, `update_doc` | | Put suggested jobs on your Plan board, for you to keep or dismiss | `plan_jobs` | | Attach a document to a job on the board | `attach_doc` | | Publish an article to the website you connected under [Publishing](https://citeroot.app/docs/connect/publishing), as a draft or live | `publish_article` | | Track new questions, within your plan's limit | `add_questions` | | Track a competitor | `add_competitor` | | Add a job to What to do next on Home | `add_todo` | | Change one of your skills, or make a new one | `save_skill` | The last five work as they do with Citeroot Agent, with the same rules: - **Publishing** goes as a draft unless you ask for live. An article with blanks in [square brackets] never goes live, and the "Before you publish" notes are never sent. - **Questions** must be ones a buyer would ask, where a good answer names brands, products or shops. Citeroot refuses a question that names your business (AI would answer it from your own site, so it says nothing about whether AI recommends you), one that compares two other brands, and one that only asks for advice. Your app gives each question its keyword, intent and topic, so Citeroot's own AI isn't used. A new question is asked at the next scheduled check. - **A competitor** you add is counted in the answers you already have, in the background. Most apps ask you before they use a tool that changes something. Say yes only to what you asked for. With more than one project, the app says which one it means; it can only name projects in your own account. Apps that support MCP prompts also show five ready-made requests: write an article, report on my AI visibility, plan my week, find my content gaps, and track new questions. ## What it can't do - See your password or payment details. - Change your plan or your account settings, or connect a website for publishing. You do those in Citeroot. - Delete anything: a question, a competitor, a document or a post on your site. (It can replace the text of a document in Docs with a new version, and update a post it published.) - Run a check outside the schedule. - Use Citeroot's own AI, web search or page reader. Your app uses its own, so nothing it does spends your AI credits. ## Writing an article from your AI app Ask your app to write an article for a question you track. It calls `get_article_brief`, which returns the facts Article Studio's own writer gets and the [article skill](https://citeroot.app/docs/diagnose/briefs-and-insights#the-article-skill). The app writes the article and saves it with `save_doc`. It then shows in Docs and in Article Studio as a draft, marked with the app that wrote it. One difference from Article Studio: your app can read the web, so it may add a source or a figure it found. Citeroot can't check those. Read every claim before you publish, as with any draft. ## If it doesn't connect - Check the address is exactly your Citeroot address, ending in `/mcp`, with no space after it. - If Claude says it couldn't check the server, wait a minute and add it again. Remove any half-added Citeroot entry from the list first. - If the app says the connector couldn't be added, it may be in your list already from an earlier try. Open the app's connectors, find Citeroot and press **Connect** to sign in. In Claude Code, run `/mcp`, choose Citeroot and authenticate. - If you are sent to a Citeroot page that says the link can't be used, start again from the app. That page never sends you anywhere it doesn't know. - Still stuck? Send us the exact message the app shows. ## Disconnecting **Settings**, then **Connect AI apps**, lists every connected app and token, with when each was last used. **Disconnect** or **Remove** stops it at once. You can also disconnect from inside most apps. An app that is not used for 60 days has to be connected again. ## Good to know - AI answers and web pages in your data are text written by others. A well-behaved app treats them as data, not as instructions. If an app ever does something you didn't ask for after reading an answer, disconnect it and tell us. - Each account can make 120 requests a minute through the server. - For developers: the server speaks MCP over HTTP (POST, JSON responses) at `/mcp`. Sign-in is OAuth 2.1 with PKCE; an app can register itself (dynamic client registration) or name itself by a web address (a client ID metadata document); the details are at `/.well-known/oauth-protected-resource` and `/.well-known/oauth-authorization-server`. --- ## The API Source: https://citeroot.app/docs/connect/api Updated: 2026-11-15 > Read your Citeroot numbers as JSON from a dashboard, a spreadsheet or a script: how to sign in, what you can read, and the limits. The API gives you your account's data as JSON, for a dashboard, a spreadsheet or a script of your own. It is **read-only**: nothing can be changed through it. It asks no AI platform and no AI model, so a request uses none of your answers or AI credits. It is on the paid plans (Grow, Pro, Max and Enterprise). ## Sign in 1. Open **Settings → Connect AI apps**. 2. Under **Tokens**, type a name (for example "Looker Studio") and press **Make a token**. 3. Copy the token. It is shown once. Send it with every request, in the `Authorization` header: ``` curl -H "Authorization: Bearer YOUR_TOKEN" https://citeroot.app/api/v1/projects ``` Treat the token like a password. To stop it, press **Remove** beside it on the same page. It stops at once. ## What you can read Every address starts with `https://citeroot.app/api/v1/`. All of them are `GET`. | Address | What it returns | Parameters | | --- | --- | --- | | `projects` | Your projects: id, name, website, country, competitors. | none | | `visibility` | Named, cited and recommended rates with their ranges, the change on the period before, and each AI platform. | `platform`, `days` (7 to 90, default 28) | | `questions` | Tracked questions with how often you are named. | `sort` (`named_low` or `newest`), `limit` | | `answers` | The latest answers, or one answer in full. | `id`, `question`, `platform`, `named` (`true` or `false`), `limit` | | `competitors` | You and each competitor: share of answers and change. | none | | `sources` | The sites AI cites, by type. | `type`, `gaps_only`, `limit` | | `site-health` | The latest website scan: score and findings. | none | | `todo` | The jobs on What to do next. | none | | `plan` | The Plan board's jobs. | none | | `docs` | The documents in Docs. | `kind`, `search`, `limit` | `GET https://citeroot.app/api/v1` lists the same addresses with every parameter. **More than one project?** Add `project` to the request, with the project's id, website or name: ``` curl -H "Authorization: Bearer YOUR_TOKEN" "https://citeroot.app/api/v1/visibility?project=example.com&days=28" ``` With one project you can leave it out. ## An example `GET /api/v1/competitors` returns, in outline: ``` { "window": "last 28 days, against the 28 days before", "answers": 832, "brands": [ { "name": "Competitor 1", "domain": "competitor-1.example", "named": { "pct": 63, "range": "59–66%", "of": "522 of 832" }, "before": 64, "change": "flat" }, { "name": "Your brand", "you": true, "named": { "pct": 38, "range": "35–42%", "of": "319 of 832" }, "before": 23, "change": "flat" } ] } ``` The names and numbers above are placeholders that show the shape. `change` is `up` or `down` only when the change is bigger than normal variation. See [Reading changes and variance](https://citeroot.app/docs/measure/variance). ## Errors and limits | Status | What it means | | --- | --- | | `400` | The request can't be answered as sent. The `error` text says why, for example which projects to choose from. | | `401` | No token, or a token that is wrong or removed. | | `403` | Your plan doesn't include the API. | | `404` | No such address. | | `429` | Too many requests. The limit is 120 a minute for each token. Wait a minute. | Every error is JSON with one field, `error`. ## What it doesn't do - It can't add or change questions, projects or competitors, and it can't start a check. Checks run on your plan's schedule. - There are no webhooks. To see new numbers, ask again after your next check. - The token reaches the account of the person who made it. Team members can't make one for the owner's account. - For AI apps such as Claude and ChatGPT, use the MCP server instead: it adds Citeroot's instructions to the same data. See [Connect AI apps](https://citeroot.app/docs/connect/ai-apps). The addresses under `/api/v1` will keep working. If a change would break them, it will get a new version. --- ## Publishing to your website Source: https://citeroot.app/docs/connect/publishing Updated: 2026-11-15 > Connect WordPress, Shopify or a webhook of your own, and send articles to your site as drafts or live: by hand, by the agent, or automatically. Connect your website once, and Citeroot can put an article on it. You choose whether it arrives as a **draft** for you to check, or **live** for visitors to see. It is on the paid plans (Grow, Pro, Max and Enterprise). Each project has its own connection, because a project is one website. ## Three ways an article gets there - **You press a button.** Open an article in Docs or Article Studio and press **Send to your site as a draft** or **Publish live**. - **You ask the agent**, or an AI app you connected (Claude, ChatGPT and others). For example: "write an article for my weakest question and send it to my site". It sends a draft unless you say to make it live. - **Automatically.** In **Settings → Publishing**, tick **Send every new article automatically**. Every article the agent or Article Studio finishes then goes to your site. ## What is sent, and what is not - The article's title, its body as HTML, a short description and the last part of its web address. - The **Before you publish** notes are for you. They are never sent. The title tag, meta description and web address written there are sent as fields where the site has a place for them. - An article with blanks left in **[square brackets]** never goes live. It is sent as a draft, and Citeroot tells you which blanks are left. - Sending the same article again updates the post already on your site. It does not make a second one. - Raw HTML and scripts in an article are sent as plain text, so nothing can run on your site. Start with drafts. Read each article on your site before visitors see it. ## WordPress You need a WordPress user who can write posts (an Author, Editor or Administrator), and an **application password** for that user. An application password is a separate password for one app. It is not your login password, and you can remove it at any time. 1. In WordPress, open **Users → Profile**. 2. Scroll to **Application Passwords**. Type a name, for example "Citeroot", and press **Add New Application Password**. 3. Copy the password it shows. 4. In Citeroot, open **Settings → Publishing**, choose **WordPress**, and enter your site's address, the user name and that password. Press **Connect**. Articles arrive under **Posts**. The short description goes in the post's excerpt. If you use an SEO plugin, copy the title tag and meta description from the article's notes into the plugin's fields. If it doesn't connect: - **"WordPress refused the username or application password."** Make a new application password. Use the user name, not the email address. - **"We couldn't find WordPress's REST API."** Check the address. A security plugin may have switched the REST API off. - **No Application Passwords section.** WordPress shows it only on sites that use https. Some security plugins hide it. ## Shopify Articles go into a blog in your store, under **Content → Blog posts**. Shopify changed how stores connect apps in January 2026, so there are two ways. **Stores set up since 2026: client ID and secret** 1. In Shopify's **Dev Dashboard**, create an app for your store. 2. Give it the **write_content** permission. 3. Install it on your store. 4. Copy its **client ID** and **client secret**. 5. In Citeroot, open **Settings → Publishing**, choose **Shopify**, and enter your store's address (the one ending in `.myshopify.com`), the client ID and the secret. The app and the store must be in the same Shopify organisation. **A custom app made before 2026: access token** If your store already has a custom app (Settings → Apps → Develop apps) with the **write_content** permission, choose **Access token** and paste its Admin API access token. Either way you can name the blog the articles go in. Leave it empty to use the store's first blog. The author name is shown on each article. The title tag and meta description are not set in Shopify's search engine listing. Copy them from the article's notes. ## Webhook For any other system, Citeroot can send each article to an address of yours as a signed `POST`. It works with your own code, and with Zapier, Make or n8n (use their "catch a webhook" step). In **Settings → Publishing**, choose **Webhook** and enter the address. It must start with `https://`. Citeroot sends a test request (`ping`) and saves the connection if your address answers `200`. It then shows a **signing secret** once. Copy it. **What we send** ``` POST https://example.com/hooks/citeroot Content-Type: application/json X-Citeroot-Event: article.draft X-Citeroot-Timestamp: 1794744000 X-Citeroot-Signature: sha256= { "event": "article.draft", "sent_at": "2026-11-15T12:00:00.000Z", "project": { "id": "prj_…", "name": "Your brand", "website": "example.com" }, "article": { "id": "brf_…", "status": "draft", "title": "…", "slug": "…", "title_tag": "…", "meta_description": "…", "html": "

…

", "markdown": "…" } } ``` - `event` is `article.draft`, `article.published` or `ping` (the test, with no article). - `article.id` stays the same when an article is sent again, so you can update the page you made before. - Answer with any `2xx` status when you have taken the article. If you answer with JSON such as `{ "url": "https://example.com/blog/my-page" }`, Citeroot shows that link on the article. Add `"live": false` if you kept a published article as a draft, and Citeroot will say so. **Check the signature** The signature is an HMAC-SHA256 of the timestamp, a full stop and the raw body, with your signing secret, written in hex: ``` const crypto = require("crypto"); const expected = "sha256=" + crypto.createHmac("sha256", SIGNING_SECRET).update(timestamp + "." + rawBody).digest("hex"); const ok = crypto.timingSafeEqual(Buffer.from(expected), Buffer.from(signatureHeader)); ``` Refuse a request whose signature is wrong, or whose timestamp is more than a few minutes old. To get a new signing secret, connect the same address again. ## Safety - What you type when you connect (a password, a token, a secret) is stored encrypted and is never shown again. - Citeroot only sends to the site you connected. It does not follow a redirect to another address. - **Disconnect** in Settings → Publishing stops new sends at once. Posts already on your site stay there. - To cut Citeroot off from the site's side, remove the application password in WordPress, or uninstall the app in Shopify. ## What it doesn't do - It doesn't upload images, set categories or tags, or schedule a post for later. - It doesn't delete or unpublish a post on your site. - It sends articles, not reports or plans. - An AI app connected through the [MCP server](https://citeroot.app/docs/connect/ai-apps) can publish too, when you ask it, with the same rules. A read-only [API](https://citeroot.app/docs/connect/api) token can't. --- ## Security and data handling Source: https://citeroot.app/docs/account/security-data Updated: 2026-11-15 > What we store, where it goes, and how to export or delete it. ## What we store Your account (name, email, company, a hashed password), your projects, questions (with their search keywords, intent and topic) and competitors, every answer collected for you (with its sources), website scan results, articles, older outlines and insights, your Citeroot Agent messages and any files you attach to them (kept with the chat, deleted when you delete it), and billing identifiers from our payment provider (customer and subscription ids and status — not card details). We also keep a log of emails we send you. ## Who it goes to To collect an answer we send **the question text and your project's country and language** to our data providers, Bright Data and DataForSEO (one for each platform), which collect the answers from ChatGPT, Gemini and Google. We don't send your account details. If AI is switched on, an AI model provider (Anthropic or Google, whichever Citeroot has set up) receives the facts it needs to write: for example your business profile, your questions, the text of recent answers and, for an article, text read from your own website. That covers suggested questions and competitors, brands found in your answers, articles, insights and Citeroot Agent replies. A file you attach to a Citeroot Agent message is sent to that provider with the message, so it can read it. Hosting, the database, email and payments are handled by the services listed on the [subprocessors](https://citeroot.app/legal/subprocessors) page. We don't use your data to train models. ## Isolation and access Every query is checked against the signed-in account. Passwords are hashed with scrypt, sessions use httpOnly cookies and are stored hashed, and reset and confirmation links are single-use. On Pro and above an account can have several logins (team seats): each person has their own password and sessions, and the owner can remove one at any time. Role-based access, SSO and an audit log are not built. API and AI-app tokens are stored as hashes and can be removed in Settings. If you connect your website for publishing, its password or token is stored encrypted, is never shown again, and is used only to send your articles to that site. See the [security page](https://citeroot.app/security) for what we do and don't claim. ## Retention Answers, website scans and drafts are kept until you delete them, the project or your account. There is no automatic expiry today. ## Export and deletion - **Export** — Settings → Download my data (JSON), and CSV on the Answers page for paid plans. - **Delete a project** — Project settings. Removes its questions, answers, website scans, articles, outlines and insights. - **Delete your account** — Settings. Removes your account and everything under it, and cancels any active subscription immediately. Billing records the payment provider is legally required to keep stay with them. ## Reporting a vulnerability See the [security page](https://citeroot.app/security) and [/.well-known/security.txt](https://citeroot.app/.well-known/security.txt). --- ## Trial, plans and billing Source: https://citeroot.app/docs/account/plans-billing Updated: 2026-11-16 > How free trials work, what happens when one ends, and how to subscribe, change plan or cancel. ## Plans Citeroot has five plans: Free, Grow, Pro, Max and Enterprise (by arrangement). See [pricing](https://citeroot.app/pricing) for the current prices and what each plan includes. Limits are enforced by the app, not just described. | What you get | Free | Grow | Pro | Max | Enterprise | | --- | --- | --- | --- | --- | --- | | Price | $0 | $59 a month, or $49 a month billed yearly | $199 a month, or $169 a month billed yearly | $399 a month, or $339 a month billed yearly | By arrangement | | Projects | 1 | 1 | 5 | 10 | Custom | | Tracked questions (all projects together) | 5 | 30 | 150 | 300 | Custom | | AI platforms | All 4 | All 4 | All 4 | All 4 | All 4, and more on request | | Automatic checks | Every week | Every day | Every day | Every day | Every day | | Answers a month, at most | 100 | 3,720 | 18,600 | 37,200 | Custom | | Website health pages per scan | 25 | 100 | 250 | 500 | Custom | | AI credits a month (Citeroot Agent, articles and insights share them) | None | 50 | 200 | 400 | Custom | | Connect your own AI app (MCP server) | No | Yes | Yes | Yes | Yes | | Article drafts and insights | No | Yes | Yes | Yes | Yes | | Citeroot Agent | Not included | Agent | Agent | Agent | Agent | | CSV export of answers | No | Yes | Yes | Yes | Yes | | Read-only report links per project | No | No | Yes | Yes | Yes | Enterprise has no fixed limits: projects, questions, answers, pages, team seats and AI credits are set to what you need and written into the agreement. The answers row is the most a month's checks can need: every question on every platform at every scheduled check. You never have to count. See [Checks and allowance](https://citeroot.app/docs/measure/runs-and-allowance). ## Free trials - Every account starts on the Free plan, with no card. - A paid plan can come with a **free trial**. The [pricing page](https://citeroot.app/pricing) and **Plan & billing** show which plans have one and how many days it lasts. - The trial needs a card. You add it at Paddle's secure checkout from **Settings → Plan & billing**, nothing is charged until the trial ends, and you can cancel before then and pay nothing. We ask for a card so that trials are for real people, one each. - If you do nothing, the plan starts billing when the trial ends. - You get that plan's full limits during the trial. The top of the app shows the days left, and warns you in the last three. - You get one trial per account. A trial keeps the length it started with, even if the offer changes later. - Trials that started before card trials were introduced keep running without a card, and then move to the Free plan. ### When the trial ends - **If you do nothing,** the plan you chose starts billing on the day the trial ends, using the card you added. Paddle, our payment partner, sends the receipt. - **If you cancel before then,** you are not charged, and when the trial ends the account moves to the **Free** plan. Nothing is deleted. Your projects, questions and collected answers remain readable. Limits drop to the Free plan's: extra questions are paused (not removed), you keep all four AI platforms, and checks become weekly. - Cancel from **Plan & billing → Manage billing**. ## Subscribing Payments are open. Open **Plan & billing**, choose Grow, Pro or Max, monthly or yearly, and pay through Paddle's secure checkout. Paddle is our payment partner and the merchant of record: it takes the payment, adds sales tax or VAT where it applies, and sends the receipt. We never see your card details. If the plan you choose comes with a free trial and you have not had one, your subscription starts with it. Otherwise billing starts straight away. Paying yearly lowers the monthly rate and is billed once a year. Prices are in USD and exclude taxes. Enterprise is by arrangement. [Talk to us](https://citeroot.app/enterprise) about higher limits, a contract or a DPA review. ## Changing plan or cancelling Use **Manage billing**, which opens Paddle's billing portal: update your card, download invoices and receipts, switch plan or cancel. Cancelling takes effect at the end of the period you've paid for; you keep the plan until then. A failed payment keeps your access while the payment is retried, with a warning in the app. See the [cancellation policy](https://citeroot.app/legal/cancellation-policy) and [refund policy](https://citeroot.app/legal/refund-policy). ## Downgrades If a lower plan allows fewer questions than you have, the extra ones are paused. They aren't deleted; resume them after you upgrade or pause others. Nothing else is deleted either: your projects, answers, articles and insights stay. ## Monthly allowances Each plan's answer allowance covers its scheduled checks. See [Checks and allowance](https://citeroot.app/docs/measure/runs-and-allowance). Everything AI writes for you on a paid plan uses the same AI credits: Citeroot Agent, article drafts and insights. There is no separate limit for articles or insights; you can write as many as your credits allow. A quick question uses a little, an article or an insight more, a researched report the most. The ring beside the **+** button in Citeroot Agent's message box shows how much is used, and so does the menu's foot. Free has no credits, so no Citeroot Agent, articles or insights. Everything resets on the 1st of each month (UTC). --- ## Your account Source: https://citeroot.app/docs/account/your-account Updated: 2026-11-15 > Email confirmation, passwords, team seats, sharing a read-only report or a PDF, and your own name on reports. ## Email confirmation Where the deployment sends email, you confirm your address from a link we send after sign-up. Until you do, you can look around and edit prompts, but you can't start collections or audits — those cost real money, and confirmation keeps abandoned or fake sign-ups from using it. **Settings → Email** lets you send the link again. ## Password Passwords must be at least ten characters. Changing your password in **Settings** signs out your other devices. If you forget it, use [Forgot password](https://citeroot.app/forgot-password); the link is valid for an hour and works once. ## Team seats Pro has three logins, Max ten and Enterprise more, yours included. Free and Grow have one. - **Invite someone:** open **Settings → Team**, type their email address and press **Send invitation**. We email them a link. It works for 14 days, once, and only for that address. - **Joining:** they open the link and create a login with that address, or sign in if they already have one. Then they press **Join**. - **What a member can do:** everything inside your projects. They see the questions, answers, competitors, sources, website health, Docs, the Plan board and the agent's chats, and can add a project or change its settings within your plan's limits. Their work uses your plan's answers and AI credits. - **What stays with you:** the plan and billing, inviting and removing people, connecting AI apps and the API, and deleting a project. - **Removing someone:** press **Remove** beside their name. They lose access at once. A member can also leave from the same page. - **If your plan changes:** on a plan with fewer seats, the people who joined first keep their seat. The others see their own workspace until you upgrade or free a seat. There is one kind of member today. Roles, single sign-on and an audit log are not built. A login belongs to one team at a time. Someone who joins keeps their own Citeroot account, and gets it back when they leave. Someone with a paid plan of their own must cancel it first, or be invited at another address. ## Share a read-only report On Pro, Max and Enterprise, you can make read-only report links for a project. Anyone with a link sees the parts of the report you picked, with live numbers for the last 28 days. They don't need a login, and nothing on the page can change anything. Send one to a client, a teammate or your manager. - **Make a link:** press the **Share report** button at the top of the dashboard, next to the project's settings. Under **What to include**, keep everything or pick the parts you want: General overview, Questions, Answers, Competitors, How AI describes you, Sources and Website health. Then press **Create link** and **Copy link**. - **Several links:** each link keeps the parts it was made with, so a client can have the full report and a manager only the overview. They are listed under **Report links you made**. - **Cut off a link:** **Turn off** stops it at once. **Undo** turns the same link back on. - **If your plan changes:** on Free or Grow the links pause and show "This report isn't available". They work again, with the same addresses, if you move back to Pro or above. - **Save as PDF:** open a report link and press **Save as PDF**, then choose "Save as PDF" as the printer. The PDF is on white paper and holds every section of the link, with the folded answers opened. Search engines are told not to index the report, and the page doesn't pass its address on to the sites it links to. Still, treat a link like a password: anyone who has it can see the report. ## Your own name on reports On Max and Enterprise, your report links can show your name and logo in place of Citeroot's. Open **Settings → Report branding**, type the name, add a logo if you have one (PNG, JPEG or WebP, up to 200 KB) and press **Save**. Every report link of the account changes at once, the PDFs too. Clear the name to put Citeroot's back. The logo is shown small, on a black page and on white paper, so pick one that reads on both. ## Several brands Pro includes up to five projects and Max up to ten, which agencies can use for different clients. Your plan's question limit is shared by all of them. Switch between them, or add another, from the project menu at the top left of the dashboard. It shows how many of your plan's projects you have used. ## Deleting your account **Settings → Delete account** permanently deletes your data and cancels any subscription immediately. You'll be asked to type your email to confirm. If we can't confirm with the payment provider that the subscription is cancelled, we stop and leave the account intact rather than risk charging you for nothing. --- # Guides ## What is generative engine optimization (GEO)? Source: https://citeroot.app/blog/generative-engine-optimization Updated: 2026-11-21 > A source-checked guide to GEO: what the term means, how it overlaps with SEO and AEO, what you can influence, and how to measure it without overclaiming. **Generative engine optimization (GEO)** is the work of improving how a business and its content show up in answers written by AI systems. It includes making information easy to find and use, supplying evidence an answer can rest on, and measuring whether the result is a correct, useful appearance — not just *any* appearance. If a buyer asks an assistant about a problem you solve, does the answer include you, describe you correctly and send them somewhere useful? GEO is the practice of making that more likely and knowing whether it worked. ## Key takeaways - GEO is about being named and cited inside an AI answer, where SEO is about a position in a list of links. - It builds on SEO. Crawlable, clear, well-sourced pages are the base for both. - Assistants answer from the pages they retrieve, so being mentioned on other sites matters as much as your own pages. - You cannot control which pages an assistant picks or whether it cites you. You can make both more likely. - Measure it as separate rates (named, cited, recommended) on a fixed set of questions, with ranges. ## Where the term comes from The phrase was formalised in a 2023 research paper, *GEO: Generative Engine Optimization* (Aggarwal et al.), accepted at KDD 2024. The authors tested ways to change a source's visibility inside generated answers, rather than its position in a list of links, and reported improvements of up to roughly 40% on their visibility metrics in their experimental setting. That result is useful as evidence that content choices can matter. It isn't a promise that any tactic will lift your visibility by a fixed percentage on today's live products. Treat it as a hypothesis to test against your own questions. ## GEO, AEO and SEO Three terms get used interchangeably: - **SEO** — making content discoverable and useful in search. - **AEO (answer engine optimization)** — work aimed at being the source of direct answers, including featured snippets and assistant responses. - **GEO** — work aimed at generative systems that compose answers from retrieved sources. | | SEO | AEO | GEO | |---|---|---|---| | Where you appear | A list of links | A direct answer, such as a featured snippet | An answer an AI system writes | | What success looks like | A high position and a click | Your passage shown as the answer | Your brand named, cited or recommended | | What decides it | Relevance, links, page quality | A clear answer in the right place | The pages the system retrieves and what they say | | How you measure it | Rankings, clicks | Snippets won | Share of answers that name or cite you | They overlap heavily and aren't three independent switches. The pragmatic view: SEO fundamentals — crawlability, relevance, authority, clear pages — remain the foundation, and GEO adds new questions about what answers say and cite. ## What generative engines cite: our own data Generative engines cite forums and video as much as company websites. We track 28 buyer questions about AI visibility tools on ChatGPT, Gemini, Google AI Overviews and Google AI Mode. In 2,778 answers over the 90 days to 21 November 2026, Reddit was cited in 638 answers (23%) and YouTube in 450 (16%). Our own site was cited in 181 (7%). ![The sites AI cites most on AI visibility questions, out of 2,778 answers over 90 days: reddit.com 23%, semrush.com 17%, youtube.com 16%, tryprofound.com 13%, otterly.ai 11%, peec.ai 9%, ahrefs.com 9%, zapier.com 8%, citeroot.app 7%](https://citeroot.app/img/blog/sites-cited-ai-visibility.png) This is one topic and one set of questions, so your market will have a different list. The point it makes holds more widely: GEO work that stays on your own website misses most of the pages the answers are written from. A study of 75,000 brands by Ahrefs found the same direction for Google's AI Overviews, where brand mentions on other websites correlated with visibility (0.664) far more strongly than backlinks did (0.218) ([Ahrefs](https://ahrefs.com/blog/ai-overview-brand-correlation/)). A correlation is not proof of cause. ## Three questions that organise the work Rather than a checklist of tactics, ask three questions about each important topic. ### 1. Can it be found? Can the relevant systems reach your pages at all? - **Crawler access.** Search, training and user-triggered bots are different permissions. See [the AI crawler access guide](https://citeroot.app/blog/ai-crawler-access). - **Indexability.** Canonicals, response codes and noindex directives still apply. - **Raw HTML.** Many retrieval systems read what the server returns without running scripts. Key content that appears only after JavaScript runs may never be seen. ### 2. Can it be trusted? Would an answer feel safe resting on this page? - **Specifics over adjectives.** "Supports SSO via SAML, requires the Business plan" is more useful than "enterprise-ready". - **Corroboration.** Claims repeated by independent sources are stronger than claims made only on your own site. Review third-party descriptions of your company and correct outdated ones through legitimate channels. - **Freshness and ownership.** Dates, authors, versions and a clear owner for each page. ### 3. Can it be quoted? Could a system lift a passage and have it stay true? - **Answer-first structure.** An opening paragraph that states the answer, the audience and the key limit, as in [answer capsules](https://citeroot.app/blog/answer-capsules). - **Sectioned content.** Headings that match the questions people ask. - **Tables and lists with inclusions and exclusions** rather than marketing summaries. ## What you cannot control - Which pages a system chooses to retrieve for a given question. - Whether an answer cites you even when you're retrieved. Retrieval, selection and citation are separate decisions. - Day-to-day variance: the same prompt can return different brands on different runs. - The product's behaviour next quarter. Documentation and behaviour change; re-check. Be wary of anyone who promises a guaranteed citation position. A promise like that needs a defined question, surface, time and verification method to mean anything. ## Do you need special files or markup? For Google's search experiences, Google's guidance says no special AI text files or GEO-specific markup are required. Other systems differ. [llms.txt](https://citeroot.app/glossary#llms-txt) is a proposed convention that not every engine honours. Evaluate any proposed technical change against the documentation of the system you want to reach, not against a blog post's claim. ## How to measure it Start with separate measurements rather than one unexplained score: - **Mention rate** — valid answers that name the brand ÷ valid answers. - **Owned-citation rate** — answers that link to your own domains ÷ valid answers. - **Recommendation rate** — answers that explicitly recommend you for the prompt's use case. - **Source coverage** — which independent domains and URLs appear. - **Referral outcomes** — identifiable visits and conversions, measured separately. Define the denominator, prompt set, engines, location and time window. Report failed or unavailable observations separately rather than counting them as absence. See [how to measure AI visibility](https://citeroot.app/blog/measure-ai-visibility) for a repeatable workflow. ## A realistic first month 1. **Week 1:** baseline a fixed prompt panel for one business area. 2. **Week 2:** separate access problems from content gaps; look at the sources. 3. **Week 3:** ship one bounded improvement. 4. **Week 4:** re-measure under comparable conditions and write down what happened. The [full plan](https://citeroot.app/blog/first-month-geo-plan) has the detail. It's a workflow, not a promise that four weeks proves an effect. ## Frequently asked questions ### Does GEO replace SEO? No. GEO builds on SEO. Keep the search fundamentals, and add measurement of what AI answers say and cite. ### What is the difference between GEO and AEO? AEO (answer engine optimisation) aims at being the source of a direct answer, including featured snippets. GEO aims at systems that write a new answer from several retrieved pages. In practice the work overlaps and many people use the two words for the same thing. ### Can I guarantee that ChatGPT will cite my site? No. Access and useful content do not guarantee selection for a particular answer. Anyone who promises a fixed position should say for which question, on which platform, when, and how it will be checked. ### How long does GEO take to work? There is no evidence-backed timetable. Fixing blocked crawlers can change what an assistant is able to cite within days. Being named on other sites takes weeks or months. Set review dates around things you can observe. ### How do I get started with GEO? Write 20 questions your buyers ask, see who AI names today, check that AI crawlers can reach your pages, and fix one gap. [How to rank in ChatGPT](https://citeroot.app/blog/how-to-rank-in-chatgpt) walks through the steps. ## Sources - Aggarwal et al., [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735), KDD 2024. - Google Search Central, [Optimizing for generative AI features in Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide). - OpenAI, [Overview of OpenAI crawlers](https://developers.openai.com/api/docs/bots). - Perplexity, [Perplexity crawlers](https://docs.perplexity.ai/docs/resources/perplexity-crawlers). - [llms.txt proposal](https://llmstxt.org/). --- ## AI citation tracking: how to track citations in AI answers Source: https://citeroot.app/blog/ai-citation-tracking Updated: 2026-10-02 > How to track which sources AI answers cite, why a citation isn't a mention, how to compare competitors' sources and what to do with the findings. **AI citation tracking** is the practice of recording which sources an AI answer links to — and how that changes over time, across prompts and between competitors. It answers a different question from mention tracking. A mention tells you whether your brand is *named*. A citation tells you which *pages* the answer relied on or pointed to. Either can happen without the other. ## Why citations matter - Citations show **where answer evidence comes from**. If three review sites and one comparison article account for most links in your category, those are the pages shaping the answer. - They show **who is trusted** by the engine for a topic — and whether that's you, a competitor or a third party. - They give you **a concrete work list**: pages to improve, sources to earn, claims to correct. ## What to record For every answer, store: | Field | Why it matters | | --- | --- | | Prompt, stage and wording | Results are only comparable for the same prompt | | Engine and model version | Different engines cite differently | | Country and language | Sources vary by market | | Date and run number | Answers vary run to run | | Cited URLs, in order | Order and presence both carry information | | Anchor or passage context | Shows *what* the source supports | | Whether the brand was named | Keep mentions separate from citations | Store answers **as collected**. Rewriting or summarising them loses the evidence you'll want when someone challenges a number. ## Classify sources A flat list of 400 domains isn't useful. Classify each domain so patterns appear: - **Owned** — your own domains. - **Competitor** — rivals' domains. - **Editorial** — publications and independent guides. - **Review and comparison sites.** - **Community** — forums, Q&A, social. - **Documentation and reference.** - **Marketplaces and directories.** - **Video.** Then ask which types each engine leans on for each stage of the journey. ## Metrics - **Owned-citation rate** — answers that link to your domains ÷ valid answers. - **Source share** — a domain's share of all citations in a prompt set. - **Gap sources** — domains cited alongside competitors but not for you. - **Corroboration** — whether a claim about you appears on independent sources as well as your own. Report the denominator every time, and keep citation rates separate from mention rates. ## Turning findings into action 1. **Fix access first.** If your own pages can't be fetched, nothing else matters. See [AI crawler access](https://citeroot.app/blog/ai-crawler-access). 2. **Make your pages the best source for a specific question.** State the answer plainly, add evidence and keep it current. 3. **Close gaps legitimately.** For sources that cite rivals but not you: is there a real reason to be listed or reviewed? A better source to offer? A factual correction to request? 4. **Avoid manufactured endorsements.** Fake comparison sites and paid pseudo-reviews mislead readers and tend to backfire. ## Common mistakes - Treating the **order of citations** as a ranking. It may reflect how the answer was composed, not source quality. - Counting a **citation to a competitor's page about you** as an owned citation. - Ignoring **variance**. A source that appears in one run of ten is different from one that appears in nine. - Comparing **different prompt sets** between periods. ## Tooling options You can start with a spreadsheet and a handful of prompts. Once you need repeat runs, competitors and history, a purpose-built tool saves time. Source Map in Citeroot ([learn more](https://citeroot.app/platform/source-map)) classifies domains and produces gap lists from the answers it collects. ## Sources - OpenAI, [Overview of OpenAI crawlers](https://developers.openai.com/api/docs/bots). - Google Search Central, [Optimizing for generative AI features in Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide). - Aggarwal et al., [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735). --- ## AI search share of voice: definition, formula and pitfalls Source: https://citeroot.app/blog/ai-search-share-of-voice Updated: 2026-09-28 > How to calculate share of voice for AI answers, why the denominator matters, and how many answers you need before a difference means anything. **Share of voice (SOV)** in AI answers is a brand's mentions as a share of all tracked brands' mentions across a defined set of answers. It's a relative measure. If you and two competitors are named 24, 31 and 22 times in 60 valid answers, your SOV is 24 ÷ (24 + 31 + 22) ≈ 31%. ## Formula ``` SOV(brand) = mentions(brand) ÷ Σ mentions(all tracked brands) Mention rate(brand) = mentions(brand) ÷ valid answers ``` They answer different questions. SOV tells you how you split the conversation with named rivals. Mention rate tells you how often you appear at all. A category where nobody is mentioned much can give you a big SOV and a small mention rate. ## Define everything around it An SOV figure without its frame is not meaningful. Always state: - **The prompt set** — which questions, how many, which stage. - **The engines** — and the model versions where known. - **Location and language.** - **The time window** and number of runs per prompt. - **The brand list** — SOV changes when you add a competitor. - **What counted as valid** — failed collections are not absences. ## How many answers do you need? Small samples produce wide uncertainty. If you checked 20 answers and a brand appeared in 5, the plausible range for its true mention rate is roughly 11%–47% (a 95% Wilson interval). A brand at 7 of 20 overlaps almost entirely with it. Practical rules: - Don't call a difference real unless the intervals clearly separate. - Prefer more repeated runs of a stable panel over more prompts you won't act on. - Use our [share-of-voice calculator](https://citeroot.app/tools/share-of-voice-calculator) to see the intervals for your numbers. ## Pitfalls **Counting mentions per answer instead of answers.** If a single answer names a brand three times, it is still one answer mentioning that brand. Decide a unit and keep it. **Blending stages.** A strong showing in problem-stage questions can hide a weak one in comparisons. Segment before you average. **Position weighting without disclosure.** Some tools weight earlier mentions more. That's legitimate, but publish the weighting. **Mixing branded and non-branded prompts.** A prompt that names you guarantees you're mentioned. Keep branded prompts out of SOV. **Forgetting citations.** A brand may dominate mentions while a competitor dominates the *sources*. Report both. See [AI citation tracking](https://citeroot.app/blog/ai-citation-tracking). ## Reporting template | Item | Value | | --- | --- | | Period | Sep 1 – Sep 28 | | Prompt set | 48 non-branded prompts, 4 stages | | Engines | ChatGPT, Gemini, Claude | | Valid / failed runs | 432 / 11 | | Brand | Mentions | SOV | Mention rate (95% CI) | | Brand A | 118 | 31% | 27% (23–32%) | Fill in your own numbers; keep the structure. ## Sources - Wilson score interval: Wilson, E. B. (1927), "Probable inference, the law of succession, and statistical inference", *Journal of the American Statistical Association*. - Aggarwal et al., [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735). --- ## AI crawler access: robots.txt, llms.txt and rendering Source: https://citeroot.app/blog/ai-crawler-access Updated: 2026-10-06 > Which AI crawlers exist, what each is for, how to configure robots.txt by purpose, and why raw HTML still matters. Before an AI system can use your content, it has to be able to fetch it. This guide covers the access layer: who's knocking, what you can say about it and the common ways sites accidentally shut the door. ## Who's knocking Several providers publish separate user agents for separate jobs. Check each provider's current documentation, because names and behaviour change. | Purpose | Examples | | --- | --- | | **Search and answer indexing** | `OAI-SearchBot` (OpenAI), `PerplexityBot` (Perplexity), `Claude-SearchBot` (Anthropic), `Googlebot`, `Bingbot` | | **Training** | `GPTBot` (OpenAI), `ClaudeBot` (Anthropic), `Google-Extended` token, `Applebot-Extended`, `CCBot` | | **User-triggered fetches** | `ChatGPT-User` (OpenAI), `Perplexity-User`, `Claude-User` | A few clarifications: - `Google-Extended` is a robots.txt token, not a separate crawler. Google documents that it doesn't affect inclusion in Google Search. - Allowing a search crawler doesn't opt you into training, and vice versa, for providers that separate them. - User-triggered fetchers behave differently from background crawlers. Providers describe how, or whether, robots.txt applies; read their docs before assuming. ## Configure by purpose robots.txt works per user agent. Group the bots you treat the same: ``` # Allow search and user-triggered access User-agent: OAI-SearchBot User-agent: PerplexityBot User-agent: Claude-SearchBot User-agent: ChatGPT-User User-agent: Perplexity-User User-agent: Claude-User Allow: / # Opt out of model training User-agent: GPTBot User-agent: ClaudeBot User-agent: Google-Extended Disallow: / # Everyone else User-agent: * Disallow: /admin/ Sitemap: https://example.com/sitemap.xml ``` Rules are matched per RFC 9309: the most specific user-agent group applies, and within a group the longest matching path wins. Whether to restrict training is a business decision. Make it deliberately and document it. ## robots.txt is not access control It's an honour system for well-behaved crawlers. For private content use authentication. For abusive traffic use your firewall — but note that blanket WAF rules ("block anything with 'bot' in the user agent") can lock out the search crawlers you want. Verify a crawler's identity using the provider's published IP ranges or verification method before allowing it through. ## Rendering matters Many retrieval systems read the HTML your server returns and don't execute JavaScript. Check that: - The main content and key facts appear in the **raw HTML**. - Important pages don't depend on client-side navigation to be reachable. - Response codes are correct (200 for pages, 404 for missing ones, no soft-404s). A quick test: `curl` the page and read the output. If the answer isn't there, a crawler may not see it either. The [page-readiness audit](https://citeroot.app/tools/page-readiness) automates that check. ## What about llms.txt? `llms.txt` is a proposed convention: a Markdown file at `/llms.txt` summarising your site for language models. It's not a standard that all engines honour, and Google's guidance says special AI text files aren't required for its search experiences. If you maintain one, treat it as documentation. Generate one in a minute with our [llms.txt generator](https://citeroot.app/tools/llms-txt-generator). ## Common mistakes - Staging rules (`Disallow: /`) shipped to production. - Allowing bots in robots.txt but blocking them in a CDN or WAF. - Blocking by path patterns that accidentally catch documentation or pricing pages. - Redirect chains on robots.txt itself. - Assuming a successful crawl means inclusion in answers. Access is necessary, not sufficient. ## Check your site Run the free [AI crawler checker](https://citeroot.app/tools/ai-crawler-checker). It tests your robots.txt against 20 AI-related user agents grouped by purpose and shows the matching rule for each. ## Sources - IETF, [RFC 9309: Robots Exclusion Protocol](https://www.rfc-editor.org/rfc/rfc9309). - OpenAI, [Overview of OpenAI crawlers](https://developers.openai.com/api/docs/bots). - Perplexity, [Perplexity crawlers](https://docs.perplexity.ai/docs/resources/perplexity-crawlers). - Google Search Central, [Introduction to robots.txt](https://developers.google.com/search/docs/crawling-indexing/robots/intro) and [Optimizing for generative AI features in Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide). - [llms.txt proposal](https://llmstxt.org/). --- ## How to measure AI visibility: a 9-step workflow, with a worked example Source: https://citeroot.app/blog/measure-ai-visibility Updated: 2026-11-21 > Measure AI visibility by asking a fixed set of buyer questions on a schedule and counting who is named. Five metrics, nine steps, and our own numbers as the example. To measure AI visibility, write a fixed list of questions your buyers ask, put each one to the AI assistants several times, and count how often your brand is named, how often your site is cited and who is named instead. The result is a rate with a range, for one set of questions, on named platforms, in one country, over stated dates. ## Key takeaways - AI visibility belongs to a set of questions, not to a company. Change the questions and the number changes. - Track five separate metrics: mention rate, citation rate, recommendation rate, share of voice and accuracy. - Answers vary from run to run. Ask each question more than once and report a range, not a single figure. - Keep questions that contain your brand name apart. They always name you and prove nothing about being found. - A spreadsheet is enough to start. The method below is the same at any size. ## What does AI visibility mean? AI visibility is how often an AI assistant names, cites or recommends a brand when people ask questions the brand could answer. It is measured on a sample of answers, so every figure is an estimate. Three things are easy to mix up: - A **mention** is your brand named in the text of an answer. - A **citation** is a link to your own site among the answer's sources. - A **recommendation** is the answer saying you are a good choice for what was asked. An answer can cite your page without naming you, or name you while recommending someone else. Count them apart. Our guide to [mentions, citations and recommendations](https://citeroot.app/blog/mentions-citations-recommendations) has examples of each. ## The five metrics | Metric | Formula | What it tells you | |---|---|---| | Mention rate | Answers naming the brand ÷ valid answers | Whether you are found at all | | Citation rate | Answers linking to your site ÷ valid answers | Whether your own pages are used as a source | | Recommendation rate | Answers recommending the brand ÷ valid answers | Whether being named helps you | | Share of voice | Your mentions ÷ all tracked brands' mentions | Your size next to chosen rivals | | Accuracy rate | Answers describing you correctly ÷ answers naming you | Whether what is said is true | A "valid answer" is one that was collected successfully. A run that failed is not an answer that left you out, so keep failed runs out of the denominator and report how many there were. ## A worked example: our own numbers We track 28 buyer questions about AI visibility tools for our own product, Citeroot, on ChatGPT, Gemini, Google AI Overviews and Google AI Mode. In the 28 days to 21 November 2026 that produced 2,187 valid answers. - **Mention rate:** 184 of 2,187 answers named Citeroot, which is 8%, with a 95% range of 7 to 10%. - **Citation rate:** 144 of 2,187 answers linked to citeroot.app, which is 7%. - **Recommendation rate:** 101 of 2,187 answers recommended it, which is 5%. - **By stage:** every mention came from comparison and purchase questions. On problem and shortlist questions the rate was 0%. The same answers named Semrush in 1,113 (51%), Profound in 1,016 (46%) and Ahrefs in 866 (40%). ![Brands named in 2,187 AI answers about AI visibility tools: Semrush 51%, Profound 46%, Ahrefs 40%, Peec AI 39%, Otterly.AI 33%, SE Ranking 19%, Scrunch AI 16%, Writesonic 9%, Citeroot 8%](https://citeroot.app/img/blog/brands-named-ai-visibility.png) Two lessons from our own figures. The overall 8% hides the fact that we are invisible where buyers start, which only the split by stage shows. And the comparison with the 28 days before (9%) is flat: the ranges overlap, so that one-point move is noise. ## How to measure AI visibility in 9 steps ### 1. Choose the scope Pick one business area: a product line, an audience or a market. A company that sells three things has three visibilities. ### 2. Write the questions Write 20 to 50 questions across the buying journey: the problem, the shortlist, comparisons, constraints such as budget, and the purchase. Use the words from sales calls, support tickets and search data. Our free [buyer-prompt builder](https://citeroot.app/tools/prompt-builder) drafts a first list, and [how to build a prompt panel](https://citeroot.app/blog/build-a-prompt-panel) covers the method. ### 3. Fix the conditions Decide and write down the platforms you will check, the country and language, how many times each question is asked in a period, and the dates. Answers collected under different conditions cannot be compared. ### 4. Collect the answers and keep them whole For each run, save the question, the platform, the date, the full answer text and the list of sources. Do not summarise. Mark a failed run as failed. ### 5. Read and code each answer For every valid answer, record whether the brand is named and where in any list, whether your site is cited, whether you are recommended, which rivals are named, and whether what is said about you is right. Two people coding the same sample is the cheapest check on quality. ### 6. Calculate the rates, with ranges Work out the five metrics. For each rate, add a confidence interval so a small sample is not read as a firm result. Our free [share-of-voice calculator](https://citeroot.app/tools/share-of-voice-calculator) does both. Here it is with four brands from the example above: ![The share-of-voice calculator with 2,187 answers: Semrush 35.0% share of voice and 51% mention rate, Profound 32.0% and 46%, Ahrefs 27.2% and 40%, Citeroot 5.8% and 8%, each with a 95% range](https://citeroot.app/img/blog/share-of-voice-calculator.png) Notice that share of voice (5.8%) and mention rate (8%) are different numbers for the same brand. Share of voice depends on which rivals you include. [Share of voice in AI search](https://citeroot.app/blog/ai-search-share-of-voice) explains the difference. ### 7. Find the cause Pair the numbers with four checks: - **Access.** Can AI crawlers reach your pages? Test it with the [AI crawler access checker](https://citeroot.app/tools/ai-crawler-checker). - **Sources.** Which sites do the answers cite, and which of them never mention you? - **Demand.** What do people search for around these questions? Search Console shows it. - **Outcomes.** Which visits come from AI assistants, and what do they do? See [reading GA4 for AI referrals](https://citeroot.app/blog/reading-ga4-for-ai-referrals). In our case the sources explain the low rate. Across 2,778 answers over 90 days, Reddit was cited in 638 and named us in 25 of those. ![The sites AI cites most on AI visibility questions, out of 2,778 answers: reddit.com 23%, semrush.com 17%, youtube.com 16%, tryprofound.com 13%, otterly.ai 11%, peec.ai 9%, ahrefs.com 9%, zapier.com 8%, citeroot.app 7%](https://citeroot.app/img/blog/sites-cited-ai-visibility.png) ### 8. Report it honestly A report people can trust states the questions, platforms, country, language and dates; the valid and failed runs; each metric with its range; three example answers (one good, one bad, one surprising); and what you will change and when you will measure again. ### 9. Change one thing, then measure again Make one bounded change, record the date, and run the same questions under the same conditions. Call the result evidence, not proof. Other things changed in those weeks too. ## A reporting template ``` Scope: Period: Questions: () Platforms: · Country and language: <...> Runs: valid, failed Mention rate: % (range %) Citation rate: % Recommendation rate: % Most cited sites: Gaps: Accuracy problems: Next change: ``` ## Common mistakes - **One score for everything.** A single "AI visibility score" hides which of the five metrics moved. - **Checking once.** One answer is one sample. The same question names different brands on different runs. - **Changing the questions each month.** A trend needs the same questions. Add new ones as a separate set. - **Counting branded questions.** "Is [your brand] good?" names you every time and inflates the rate. - **Reading small differences.** If the ranges overlap, you have not measured a change. - **Counting failed runs as absences.** A run that returned nothing says nothing about you. ## Frequently asked questions ### What is a good AI visibility score? There is no universal benchmark, because the number depends on the questions you chose. Compare yourself with the rivals named in the same answers, and with your own earlier figures on the same questions. ### How many questions do I need to track? Start with 20 to 50. Fewer than that and the ranges are too wide to read. More only helps if someone will act on each question's result. ### How often should I measure AI visibility? On a fixed schedule: daily or weekly collection, read monthly. Answers change with model updates and with new pages on the web, so a single audit goes stale quickly. ### Can I measure AI visibility for free? Yes, by hand in a spreadsheet, for a small number of questions. It becomes slow once you ask each question several times on several platforms, which is when a tool earns its cost. ### Which AI platforms should I check? The ones your buyers use. For most businesses that means ChatGPT, Google's AI Overviews and AI Mode, and Gemini. Check each separately, because the same question gets different answers on each. ## Next step Write your first five questions and ask them by hand today. To have them asked on a schedule, Citeroot's Free plan checks five questions every week on ChatGPT, Gemini, Google AI Overviews and Google AI Mode, and keeps every answer. ## Sources - Aggarwal et al., [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735), KDD 2024. - Google Search Central, [Optimizing for generative AI features in Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide). - Citeroot's own tracking of 28 questions, 21 November 2026. The counts in this article come from it. --- ## AI search statistics for 2026: usage, clicks, traffic and what gets brands mentioned Source: https://citeroot.app/blog/ai-search-statistics Updated: 2026-10-09 > Sourced numbers on how many people use AI assistants, what AI answers do to clicks, how much traffic they send and which signals go with being mentioned. Every figure links to its source. This page collects the numbers people ask about most when they plan for AI search, with the date and source of each one. We only include a figure when we could trace it to the company that published it or to a named study, and we say when a number is a forecast, a correlation or a single study. **How to use this page:** quote each figure with its original source and date (links below). If the page saved you time, a link back to it is appreciated. We update it when a source publishes newer numbers; the date at the top shows the last review. ## Key figures at a glance | Figure | Value | Source and date | | --- | --- | --- | | ChatGPT weekly active users | 900 million (and 50 million paying subscribers) | OpenAI, February 2026 | | Google AI Mode monthly active users | More than 1 billion | Alphabet Q2 2026 earnings remarks, July 2026 | | Gemini app monthly active users | 1 billion | Google, August 2026 | | Clicks on a regular result when Google shows an AI summary | 8% of visits, versus 15% without one | Pew Research Center, July 2025 | | Change in position-one click-through rate when an AI Overview appears | About 58% lower | Ahrefs, 2026 re-run of its 2025 study | | Share of all website traffic that comes from AI assistants | Between about 0.3% and 1%, depending on the sample | SE Ranking 2026; Conductor 2025 | | Growth in AI-referred traffic to US retail sites | +393% year over year in Q1 2026 | Adobe, Q1 2026 data | | Strongest correlate of brand mentions in AI Overviews | Branded web mentions (correlation 0.664) | Ahrefs, 75,000-brand study | | Chance an assistant gives the same brand list twice | Less than 1 in 100 | SparkToro and Gumshoe study | ## How many people use AI assistants - **ChatGPT has 900 million weekly active users** and 50 million paying subscribers, OpenAI said on February 27, 2026, up from the 800 million weekly users it reported in October 2025 ([TechCrunch](https://techcrunch.com/2026/02/27/chatgpt-reaches-900m-weekly-active-users)). - **Google's AI Mode has passed 1 billion monthly active users** since it expanded globally, Sundar Pichai said in Alphabet's Q2 2026 earnings remarks. He also said AI Mode is adding to the total number of searches rather than only replacing them ([Google](https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q2-2026/)). - **The Gemini app reached 1 billion monthly active users in August 2026**, up from about 400 million in May 2025 and 950 million in July 2026. The count is for the app alone and does not include AI features inside Search ([iPhone in Canada](https://www.iphoneincanada.ca/2026/08/12/google-gemini-app-surpasses-1-billion-monthly-active-users/)). - **A forecast, for context:** in February 2024 Gartner predicted that traditional search engine volume would drop 25% by 2026 as people move to AI chatbots and virtual agents ([Gartner](https://www.gartner.com/en/newsroom/press-releases/2024-02-19-gartner-predicts-search-engine-volume-will-drop-25-percent-by-2026-due-to-ai-chatbots-and-other-virtual-agents)). It is a prediction, not a measurement, and Google reported all-time-high Search usage in mid-2026, so treat it as a direction of travel rather than a fact about 2026. **What it means:** user numbers are not the same as buying questions, but the audience asking assistants is now the size of a major search engine. It is worth knowing what they are told about your category. ## What AI answers do to clicks - **People click less when Google shows an AI summary.** Pew Research Center tracked the browsing of 900 US adults in March 2025: users clicked a traditional result in 8% of visits to a results page with an AI summary, against 15% without one. Links inside the summary itself were clicked in about 1% of visits, and people were more likely to end their browsing session after a page with a summary (26% against 16%). 58% of the panel ran at least one search that produced an AI summary that month ([Pew Research Center](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears/)). - **The top result loses more than half its clicks when an AI Overview appears.** Ahrefs' 2026 update of its click-through study found the top-ranking page's click-through rate about 58% lower on keywords that show an AI Overview than on comparable keywords without one ([Ahrefs press release](https://www.cadillacnews.com/ap/business/new-research-googles-ai-overviews-now-cost-websites-58-of-their-clicks/article_ee08befe-e058-53bc-9e2c-faed1fd43b84.html)). The 2025 version of the study found 34.5% ([Ahrefs](https://ahrefs.com/blog/ai-overviews-reduce-clicks/)). Ahrefs reports this as a correlation, not proof that AI Overviews alone caused the drop. **What it means:** for questions that trigger an AI answer, being named or cited inside the answer matters more than it used to, because fewer people scroll past it. ## How much traffic AI assistants send - **Still a small share of all traffic.** Conductor's benchmark of 13,770 domains found 1.08% of all website traffic came from AI referrals between May and September 2025, and ChatGPT sent 87.4% of those referrals ([Search Engine Land](https://searchengineland.com/ai-1-traffic-mostly-chatgpt-464653)). SE Ranking measured 0.32% in 2026, up from 0.24% in 2025 and 0.02% in 2024 ([SE Ranking](https://seranking.com/blog/ai-traffic)). The two numbers differ because the samples and methods differ. - **Growing quickly, and converting better.** Adobe reported that traffic from AI sources to US retail sites grew 393% year over year in Q1 2026, and that by March 2026 it converted 42% better than other traffic; a year earlier it had converted 38% worse ([Decrypt, citing Adobe](https://decrypt.co/364733/ai-traffic-us-retailers-jumps-q1-agentic-shoppers-outspend-humans)). - **Visitor value estimates vary widely.** Semrush estimated that the average visitor from an AI search platform is worth 4.4 times an average organic search visitor, measured by conversion rate ([Semrush](https://www.semrush.com/blog/ai-search-seo-traffic-study/)). Other studies put the gap anywhere from about 2 times to more than 20 times, mostly because they measure different sites and industries. **What it means:** most sites cannot judge AI visibility from referral traffic alone. Many answers name a brand without sending a click, and some assistants don't pass a referrer, so measure what the answers say as well as the visits you can see. Our article on [reading GA4 for AI referrals](https://citeroot.app/blog/reading-ga4-for-ai-referrals) covers the visits part. ## How buyers use AI to choose - **Most technology buyers now use AI during a purchase, and nearly all of them check it.** TrustRadius's 2026 B2B Buying Disconnect report, based on a January 2026 survey of 1,862 technology buyers, found that 63% used AI tools during their purchase journey and 94% of those fact-check the output at least some of the time. Buyers shortlisted 2.7 products on average ([MarketScale, on the TrustRadius report](https://www.marketscale.com/industries/business-services/94-of-b2b-buyers-fact-check-ai-research-outputs-and-vendors-are-underestimating-how-far-trust-has-fallen)). **What it means:** a short list is short. If an assistant leaves you off it, you may not get a second look; and because buyers verify, the pages and reviews the assistant cites need to back up what it says. ## What goes with being mentioned and cited - **Mentions across the web correlate with AI visibility more than links do.** In Ahrefs' study of 75,000 brands, the factors most correlated with a brand being mentioned in Google's AI Overviews were branded web mentions (0.664), branded anchor text (0.527) and branded search volume (0.392); branded traffic showed a weak correlation (0.274) ([Ahrefs](https://ahrefs.com/blog/ai-overview-brand-correlation/)). These are correlations, not proof of cause. - **Community and video sites are cited constantly.** Semrush ran 325,000 prompts in January and February 2026 and found Reddit was the most-cited domain across ChatGPT Search, Google AI Mode and Perplexity ([eMarketer, on the Semrush study](https://www.emarketer.com/chart/c/368764)). Trackers using other prompt sets rank YouTube and Reddit at the top as well, though the exact shares differ between studies. - **Content changes can measurably shift visibility in controlled tests.** The research paper that coined "generative engine optimization" reported visibility gains of up to about 40% in its experimental setup ([Aggarwal et al., KDD 2024](https://arxiv.org/abs/2311.09735)). Live products change constantly, so treat that as a reason to test, not a promised result. **What it means:** your own site is necessary but not sufficient. Reviews, community discussion, comparisons and press that mention you by name are what assistants tend to find and repeat. Our guide to [AI citation tracking](https://citeroot.app/blog/ai-citation-tracking) shows how to see which sources your engines lean on. ## How stable AI answers are - **Recommendation lists almost never repeat exactly.** Rand Fishkin (SparkToro) and Patrick O'Donnell (Gumshoe) had volunteers run 12 prompts 2,961 times across ChatGPT, Claude and Google's AI. The chance of getting the same list of brands twice was under 1 in 100, and the same list in the same order appeared less than 0.1% of the time. How often a brand appeared across many runs was much steadier ([Search Engine Land](https://searchengineland.com/ai-recommendation-lists-rarely-repeat-study-468076)). The study was not peer reviewed, and both authors work in the industry. **What it means:** a single screenshot of an AI answer proves very little. Track how often you appear across repeated runs and report a range. This is the approach Citeroot takes ([how we measure](https://citeroot.app/docs/measure/variance)). ## FAQ **How many people use ChatGPT?** OpenAI reported 900 million weekly active users in February 2026. **Do AI Overviews reduce clicks?** The evidence says yes for the results below them: Pew found people clicked a regular result in 8% of visits when an AI summary appeared, against 15% when none did, and Ahrefs found the top result's click-through rate about 58% lower on keywords with an AI Overview. **How much website traffic comes from ChatGPT and other assistants?** Roughly 0.3% to 1% of all traffic in recent studies, with ChatGPT sending most of it. The share is small but growing fast, and it tends to convert well. **What makes an AI assistant mention a brand?** No one can say for certain. The strongest published correlation is with how often the brand is mentioned across the web (Ahrefs). Being described consistently and favourably on sites the assistants cite, such as reviews, community threads and comparisons, is the practical lever. **Are AI answers consistent?** No. The same prompt rarely returns the same list twice, so measure visibility as a rate over many runs. ## Sources - OpenAI via TechCrunch, [ChatGPT reaches 900M weekly active users](https://techcrunch.com/2026/02/27/chatgpt-reaches-900m-weekly-active-users), February 27, 2026. - Google, [Q2 2026 earnings call: remarks from our CEO](https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q2-2026/), July 2026. - iPhone in Canada, [Google Gemini app surpasses 1 billion monthly active users](https://www.iphoneincanada.ca/2026/08/12/google-gemini-app-surpasses-1-billion-monthly-active-users/), August 12, 2026. - Gartner, [Gartner predicts search engine volume will drop 25% by 2026](https://www.gartner.com/en/newsroom/press-releases/2024-02-19-gartner-predicts-search-engine-volume-will-drop-25-percent-by-2026-due-to-ai-chatbots-and-other-virtual-agents), February 19, 2024. - Pew Research Center, [Google users are less likely to click on links when an AI summary appears in the results](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears/), July 22, 2025. - Ahrefs, [study of AI Overviews and click-through rates](https://ahrefs.com/blog/ai-overviews-reduce-clicks/) (2025 version, with a link to the 2026 update), and the [2026 press release](https://www.cadillacnews.com/ap/business/new-research-googles-ai-overviews-now-cost-websites-58-of-their-clicks/article_ee08befe-e058-53bc-9e2c-faed1fd43b84.html) via Business Wire. - Search Engine Land, [report on Conductor's AI referral benchmark](https://searchengineland.com/ai-1-traffic-mostly-chatgpt-464653). - SE Ranking, [AI traffic grew 16x from 2024 to 2026](https://seranking.com/blog/ai-traffic). - Decrypt, [AI traffic to US retailers jumps 393% in Q1](https://decrypt.co/364733/ai-traffic-us-retailers-jumps-q1-agentic-shoppers-outspend-humans) (Adobe data). - Semrush, [AI search traffic study](https://www.semrush.com/blog/ai-search-seo-traffic-study/), 2025. - MarketScale, [94% of B2B buyers fact-check AI research outputs](https://www.marketscale.com/industries/business-services/94-of-b2b-buyers-fact-check-ai-research-outputs-and-vendors-are-underestimating-how-far-trust-has-fallen) (TrustRadius 2026 report). - Ahrefs, [An analysis of AI Overview brand visibility factors (75K brands studied)](https://ahrefs.com/blog/ai-overview-brand-correlation/). - eMarketer, [chart of the most-cited domains in AI search](https://www.emarketer.com/chart/c/368764) (Semrush data, March 2026). - Aggarwal et al., [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735), KDD 2024. - Search Engine Land, [AI recommendation lists repeat less than 1% of the time](https://searchengineland.com/ai-recommendation-lists-rarely-repeat-study-468076) (SparkToro and Gumshoe study). --- ## AI visibility tools compared (2026): prices, engines and who each one fits Source: https://citeroot.app/blog/ai-visibility-tools-compared Updated: 2026-11-16 > An honest comparison of Profound, Peec AI, Otterly.AI, Semrush, Ahrefs Brand Radar and Citeroot: starting prices, which AI engines each tracks, and when to pick which. Written by Citeroot, with sources. **Disclosure:** we make Citeroot, one of the tools on this page. We've tried to describe every tool fairly from public information, including where another tool is the better choice. If something here is out of date or wrong, email [hello@citeroot.app](mailto:hello@citeroot.app) and we'll correct it. **About the prices:** they are list prices as published by each vendor and reported in public reviews between May and October 2026 (linked under Sources). Vendors change prices and limits often, so check the vendor's own pricing page before you buy. ## The short answer - **Want to start free, or need every major engine without paying for add-ons?** Citeroot has a free plan and tracks ChatGPT and Gemini as their websites answer, plus Google AI Overviews and AI Mode. - **Need Microsoft Copilot at a low entry price?** Otterly.AI starts at about $29 a month and includes it. - **An enterprise team with budget, SSO and many engines to cover?** Profound is built for that. - **A European team that wants to choose its models?** Peec AI, priced in euros, lets you pick the models you track. - **Already paying for Semrush or Ahrefs?** Their AI visibility add-ons keep everything in one place. ## Side by side | Tool | Starting price | Engines tracked | Free option | | --- | --- | --- | --- | | Citeroot | Free plan; paid from $59/month, or $49/month billed yearly | ChatGPT, Gemini, Google AI Overviews and Google AI Mode, on every plan; Claude, Perplexity and others on Enterprise, on request | Free plan with no card; free trial details on the pricing page (a card is needed to start one) | | Profound | About $99/month, billed yearly (Starter) | ChatGPT on Starter; Perplexity and Google AI Overviews added on Growth (about $399/month); more on Enterprise | No free plan reported | | Peec AI | About €89/month (Starter) | You choose 3 models on Starter; more as paid add-ons | Check with the vendor | | Otterly.AI | About $29/month (Lite, 15 prompts) | ChatGPT, Google AI Overviews, Perplexity, Microsoft Copilot; Gemini and Google AI Mode as add-ons | Free trial reported | | Semrush AI Visibility Toolkit | About $99/month per domain | Several AI platforms; see Semrush | Part of the Semrush toolset | | Ahrefs Brand Radar | From about $199/month; full AI platform coverage costs more | Up to six AI platforms on the top tier | Part of the Ahrefs toolset | ## Tool by tool ### Citeroot What it is: an AI visibility tool that records what ChatGPT, Gemini, Google AI Overviews and AI Mode say about a brand, and keeps every answer as it was collected. Each question is asked on the assistant's own website in a fresh chat, from the country you choose. Claude is on Enterprise, on request. It measures mentions, citations and recommendations separately and shows a range for each, because answers vary from run to run. It also maps which domains the answers cite, audits a site for AI-crawler access and answer-readiness, and drafts articles for the questions where AI does not name you. - **Pricing:** The Free plan costs nothing (1 project, 5 questions checked every week on all four platforms). Grow is $59/month ($49 billed yearly) with 30 questions checked every day. Pro is $199/month ($169 billed yearly) with 150 questions and 5 projects. Max is $399/month ($339 billed yearly) with 300 questions and 10 projects. The pricing page shows which paid plans come with a free trial (a card is needed to start one). [Current pricing](https://citeroot.app/pricing). - **Good fit:** small teams, founders and agencies who want a real baseline without a sales call; anyone who needs every major engine covered on a modest budget. - **Not a fit (yet):** Claude, Perplexity, Microsoft Copilot, Meta AI and Grok are not on the standard plans (Enterprise can add them on request). There are no Search Console, GA4 or Slack integrations, no roles and no SSO yet; a read-only API, an MCP server, publishing to WordPress, Shopify or a webhook, and team seats on Pro and above exist. Citeroot is a new, small company. ### Profound What it is: an enterprise-focused AI visibility platform. Public reviews describe self-serve plans billed yearly: Starter at about $99/month covering ChatGPT with 50 prompts, and Growth at about $399/month adding Perplexity and Google AI Overviews. Enterprise is quoted individually and reportedly covers up to 10 answer engines, with SSO and SOC 2 compliance. - **Good fit:** large brands and enterprise teams that need many engines, security reviews and account support. - **Consider instead:** if you need several engines on a small budget, the entry plan covers ChatGPT only. ### Peec AI What it is: an AI search analytics tool from Berlin, priced in euros. Reviews put Starter at about €89/month and Pro at about €199/month, with prompt limits that vary by source (roughly 25 to 50 prompts on Starter). You choose the models you track, with extra models available as paid add-ons. - **Good fit:** teams, especially in Europe, that want to pick their models and track prompts across several brands. - **Consider instead:** if you want every major engine included in the base price. ### Otterly.AI What it is: an AI search monitoring tool with low entry pricing. Reviews list Lite at $29/month for 15 prompts, Standard at $189/month for 100 prompts and Premium at $489/month for 400 prompts, with ChatGPT, Google AI Overviews, Perplexity and Microsoft Copilot included and Gemini and Google AI Mode as paid add-ons. - **Good fit:** small teams that need Google AI Overviews or Copilot from day one. - **Consider instead:** if Claude or Gemini matter to you, check the add-on costs; reviews disagree on whether Claude is offered. ### Semrush AI Visibility Toolkit What it is: AI visibility tracking inside Semrush, reported at about $99/month per domain. Reviews disagree on whether a core Semrush subscription is also required, so check before you buy. - **Good fit:** teams already living in Semrush who want AI visibility next to their SEO data. - **Consider instead:** if you don't use Semrush, a standalone tool is usually cheaper. ### Ahrefs Brand Radar What it is: Ahrefs' tool for brand mentions in AI answers. Ahrefs lists Brand Radar from about $199/month, with higher tiers (reported at about $398 and $699/month) for more AI platforms, and limited prompt tracking bundled into some core plans. - **Good fit:** Ahrefs customers who want AI mentions alongside their link and keyword data. - **Consider instead:** if you only need AI visibility, the full cost can be high. ## How to choose 1. **List the engines your buyers use.** If your buyers use Copilot at work, or you sell in a market where Google AI Overviews dominate, start there. If your buyers are developers or analysts, Claude matters more. 2. **Count prompts and answers, not just seats.** Ten prompts on four engines run daily is 1,200 answers a month. Compare plans on the number of answers they collect. 3. **Check how a tool handles variation.** AI answers change between runs. A tool that shows one "rank" from one run will mislead you. Look for repeated runs and ranges. [Why answers vary](https://citeroot.app/docs/measure/variance). 4. **Ask how answers are collected.** Most tools call the providers' APIs; some capture the consumer apps or Google's results page. Both are approximations of what a signed-in person sees. [API answers vs app answers](https://citeroot.app/blog/api-vs-app-answers). 5. **Start with a trial and your own prompts.** Use the same ten prompts in two tools for two weeks and compare what you learn. ## FAQ **What is the best AI visibility tool?** It depends on the engines you need and your budget. For a free start covering ChatGPT, Gemini, Google AI Overviews and AI Mode, Citeroot. For Copilot at a low price, Otterly.AI. For enterprise requirements, Profound. **Is there a free AI visibility tool?** Yes. Citeroot's Free plan costs nothing and collects up to 100 real answers a month. Citeroot's [free tools](https://citeroot.app/tools) (AI crawler checker, page-readiness audit, llms.txt generator) need no account. **Which AI visibility tools track Claude?** Citeroot adds Claude for Enterprise accounts on request. Peec AI lets you choose models, and some other tools offer it as an add-on; check each vendor's current list. **Which tools track Google AI Overviews?** Otterly.AI includes them, Profound includes them from its Growth plan, and Citeroot includes them on every plan. ## Sources - Profound: [Rankability review (2026)](https://www.rankability.com/blog/profound-ai-review/), [GEO Toolbox on Profound pricing](https://geotoolbox.ai/blog/profound-pricing); vendor site [tryprofound.com](https://www.tryprofound.com/). - Peec AI: [eCommerce Fastlane review (2026)](https://ecommercefastlane.com/peec-ai-review/), [Ryze comparison (August 2026)](https://www.get-ryze.ai/blog/profound-vs-peec-ai-2026); vendor site [peec.ai](https://peec.ai/). - Otterly.AI: [Capterra pricing](https://www.capterra.com/p/10023384/Otterly-AI/pricing/), [G2 pricing](https://www.g2.com/products/otterlyai/pricing); vendor site [otterly.ai](https://otterly.ai/). - Semrush: [Rankability review of the Semrush AI Toolkit (2026)](https://www.rankability.com/blog/semrush-ai-toolkit-review/), [Ryze on Semrush AI visibility pricing](https://www.get-ryze.ai/blog/semrush-ai-visibility-pricing-2026); vendor site [semrush.com](https://www.semrush.com/). - Ahrefs Brand Radar: [Ryze on Brand Radar pricing](https://www.get-ryze.ai/blog/ahrefs-brand-radar-pricing-2026), [Dageno review](https://dageno.ai/blog/ahrefs-brand-radar-review); vendor site [ahrefs.com/brand-radar](https://ahrefs.com/brand-radar). - Citeroot: [pricing](https://citeroot.app/pricing), [engines and models](https://citeroot.app/docs/measure/engines-and-models). ---