AI visibility audit: what we measure, and what we don't
By the CiteTrace team · Updated 2026-11-10 · 6 min read
A number with no method behind it is a screenshot. This page is the method behind the report you get from the [free audit](/audit), so you can decide how much to trust it. We have tried to be blunt about the limits.
Inputs
- Your website. We read it to understand what you sell and who you compete with.
- 10 buyer questions. Generated from your category and use cases, a mix of category ("best X for Y"), comparison and alternatives questions. You can see all of them on the report.
- 4 engines per question: ChatGPT with web search, Perplexity, Gemini with web search, and a ChatGPT model with no web access as a baseline for what the model remembers without searching.
What we record for every answer
- Whether you are named, and your position in the list when you are.
- Which competitors are named.
- Sentiment toward you when named: positive, neutral or negative.
- Every URL the engine cited, and which of those are your own domain.
How the numbers are built
- Appeared in is the number of questions, out of 10, where at least one engine named you.
- Visibility is the share of all answers that named you.
- Share of voice is your mentions divided by all brand mentions across the answers.
- Top competitors are ranked by the number of questions in which they appear.
- Cited pages are grouped by domain so you can see who the AI reads, and which are third parties you could approach.
- Gaps are questions where competitors are named and you are not.
The headline score (0 to 100) is the average visibility across the three web-search engines, with the no-web baseline left out. It is a conversation starter. The questions and cited pages underneath are the useful part.
What a single run can't tell you
- Whether a result is stable. We ran one company's same 10 questions three times on 10 Nov 2026 and saw 5, 5 and 2 questions named. A free audit is one pass.
- Anything about Google AI Overviews or Copilot. We do not track them in the audit; Gemini with web search stands in for Google.
- Cause and effect. A page being cited tells you the engine read it, not that it influenced a purchase.
- How your answers look from another country or a logged-in account.
What happens after the audit
On a call we run your questions live and show you the report. If it makes sense to work together, the paid version tracks 50 questions daily, fixes the cited sources, and reports every Friday. If it does not, you keep the report.
We do not send anything to anyone on your behalf, and we do not use fake accounts, bought votes or fake reviews, ever.
Frequently asked questions
Is the free audit really free?
Yes. It costs us a few cents of model usage. There is no card and no obligation. We do not email you the report automatically; it is shown on screen.
Why 10 questions?
It is enough to show a pattern across categories, comparisons and use cases while staying fast (about two minutes) and cheap. The paid tracker uses 50.
Why include a baseline engine with no web access?
It shows what the model remembers from training, separate from what it finds by search. Gaps between the two tell you whether your problem is memory or findability.
Can I run my competitor's audit?
You can run any public website. The report is for the site you enter, based on public AI answers.
Where is my data stored?
See our privacy policy. We store the audit and the answers it produced, not anything about your visitors.
See who the AI recommends instead of you
A free 10-question audit: 4 engines, the pages they cite, three fixes to start with.