"Are we visible in ChatGPT?" is really three questions wearing one coat. Treating them as one number is the most common reason AI-visibility reports end up arguing with themselves.
#The three signals
Mention. Your brand is named in the text of an answer. No link required.
Citation. The answer links to a source. The source may be your page, a review of you, or a competitor's documentation. A citation of your domain doesn't mean your brand is named, and vice versa.
Recommendation. The answer explicitly suggests your brand for the use case in the prompt. It's a stricter bar than a mention: "Brightloom, Fernhill and Northgrid are options" names you; "For multi-team approvals, Fernhill is a strong choice" recommends you.
All three can happen independently:
| Case | Mentioned | Cited | Recommended |
|---|---|---|---|
| Answer lists you, links to a review site | Yes | No (not your domain) | No |
| Answer links to your docs but names only a rival | No | Yes | No |
| Answer says "choose Fernhill" and links to your pricing page | Yes | Yes | Yes |
#Why they move differently
Mentions follow what a model has learned and what retrieval surfaces. Citations follow which pages a system retrieves and decides to link. Recommendations follow how comparison content frames the category. A page fix can lift citations while mentions stay flat; a wave of review coverage can lift mentions without a single new citation of your own domain.
#How to report them
- Give each its own denominator. Mention rate is mentions divided by valid answers. Owned-citation rate is answers that link to your verified domains, divided by valid answers.
- Count only valid answers. If a run failed, it's not an absence. Report failures separately.
- Segment by prompt stage. Problem-stage questions behave very differently from "X vs Y" comparisons. A blended number hides both.
- Keep the evidence one click away. Anyone who doubts a metric should be able to read the answer behind it.
#What to do with the difference
If you're mentioned but never cited, look at who gets cited instead — often review sites and editorial lists — and whether your own pages state the key facts plainly. If you're cited but not mentioned, your documentation may be doing the work while the brand name is absent from the surrounding copy. If you're neither, start with access: can the relevant crawlers reach you at all?
A good report keeps all three in view, and treats a divergence between them as the most interesting finding on the page.