Engines and models
Which AI engines Citeroot asks, how, and how close an API answer is to what a person sees.
Citeroot collects answers by calling each provider's API with web search switched on, so the answer can cite live pages.
| Engine | Provider API | Search |
|---|---|---|
| ChatGPT | OpenAI Responses API | OpenAI's hosted web search tool |
| Claude | Anthropic Messages API | Anthropic's web search tool |
| Gemini | Google Gemini API | Google Search grounding |
| Perplexity | Perplexity API | Perplexity's built-in search |
The model behind each engine is chosen by whoever runs the deployment and is stored on every answer, so you can see exactly which model produced it. Providers rename and retire models; when that happens the operator updates the setting and the new name appears on new answers.
#How close is this to the consumer app?
Close, not identical. Differences you should expect:
- No personalisation. API calls don't carry a user's history, memory or account settings.
- Location is a hint. The country (and language) in Project settings is passed along where a provider supports it. It's not a real browser in that country.
- Different system behaviour. Consumer apps add their own instructions, tools and formatting that the API does not.
- Variation between runs. The same question can produce a different answer minutes later.
That's why Citeroot shows intervals and compares like with like. Treat the numbers as a consistent measuring stick for your panel, not as an exact count of what every person sees.
#Not supported
Google AI Overviews, Microsoft Copilot, Meta AI and Grok. There is no supported way to collect them reliably and we won't fake it.
#Your plan and engines
Scout runs one engine per prompt, Trail two and Summit all four. In Project settings you choose which, within that limit.
Last updated Oct 8, 2026 · Suggest an edit