How the AI answer check works
What we ask, which models answer, how often, and what the results can and cannot tell you.
Which models we ask
We call one API: the Perplexity Agent API, with live web search on. Through it each of your questions goes to four models:
| Shown as | Model |
|---|---|
| perplexity | Perplexity's own answer engine (its fast setting) |
| openai | OpenAI gpt-5.4-mini, required to search the web before answering |
| gemini | Google gemini-3.5-flash-lite, required to search the web before answering |
| claude | Anthropic claude-haiku-4-5, required to search the web before answering |
These are the models' API versions, not the ChatGPT, Gemini, Claude or Perplexity apps. When a provider renames or retires a model we switch to its closest replacement and update this page.
What we send
Each question exactly as you wrote it, with no extra instructions, no system prompt and nothing about your brand. Your brand and competitor names are used only on our side, to read the answers. Answers are capped at about 1,200 tokens (roughly 900 words).
How often
- Automatically once a month, on the 1st, for every question.
- By hand, once every 7 days, from the AI check page.
- Each account can run about five full checks a month in total; the page says when that is used up.
How we score an answer
- Named: your brand name or domain appears in the answer as a whole word, or the answer cites a page on your domain.
- Position: the order in which you are named among you and your listed competitors (1 = named first).
- Competitors named: the same test, for each competitor you listed.
- Sites cited: the domains of the sources the answer links to. We keep the first 600 characters of each answer.
No AI reads or judges the answers: this is plain text matching, so the same answer always gets the same score.
Limits
- Answers from the APIs can differ from what a person sees in the apps, which may use other models, their account history and their location.
- Answers change from day to day, even for the same question. One check is a sample, not a ranking; the trend over several checks says more.
- A brand mentioned in a different spelling, or only in an image, is not counted.
- We cannot see or change why a model names someone.
Your data
Your questions go to Perplexity, which passes them to OpenAI, Google or Anthropic to answer. Nothing else from your account is sent. See our sub-processors.
Frequently asked questions
Is this what ChatGPT says about me?
Not exactly. We ask OpenAI's API model with web search, which is close to but not the same as the ChatGPT app. The same is true for the other models.
Why did my result change when nothing changed on my site?
Answers vary between runs. Look at the trend across several checks rather than one result.
Can I choose the models?
Not yet. Every question goes to all four.
Start watching your site in minutes
Free for up to 10 sites, no card, nothing to install. Read-only access you can revoke at any time — never passwords, never admin access.