What people mean by “AI visibility”
In practice, the term covers four different questions: whether a brand is mentioned at all, whether it is recommended rather than listed, where it appears relative to named competitors, and which sources the answer used. A dashboard can show one number for all four; that number is usually the least useful part.
The useful unit is the answer, not the page. One question can name five brands in one run and four different brands in the next. The sources can change too: a directory in one answer, an editorial review in another, a vendor’s own documentation in a third. If you average those runs before you read them, you lose the reason the answer moved.
Why it is different from SEO
SEO measures whether a page can be discovered, understood and ranked in a search result. AI visibility measures what an assistant said after it selected and read its sources. The two overlap — crawlability, entity clarity, structured data and content quality matter to both — but they answer different questions.
A page can rank first in Google and be absent from the AI answer. A page can rank on page three and be the source an assistant quotes. The diagnostic starts from buying questions, not from the keyword-to-page map, because the answer is the thing being measured. For the comparison, see SEO audit vs AI visibility audit.
The four numbers that matter
- Mention rate: the share of sampled answers in which the brand appears at all. A mention is not a recommendation.
- Recommendation rate: the share where the brand is put forward as a choice, not just listed. This is a judgement call; the report shows the sentence behind every count.
- Position: where the brand first appears relative to the named competitors. Descriptive, not a universal ranking.
- Citation mix: the sources the answer used, classified by type. This is usually the most actionable number, because it tells you which page to fix.
The fifth number is variance: how much the answer changed between runs. A brand that appears once in three answers has a different problem from one that appears in all three at position six. We keep the runs separate. See the method for the sampling protocol.
How to measure without fooling yourself
- Fix the questions before you sample. Ten buying questions, including comparison and alternative questions. Do not add questions after you see the result.
- Log out. Account history, memory and personalisation change the answer. Use one location and one language, and write them down.
- Run each question three times over several days. Back-to-back runs are more likely to capture the same transient state. Spread them out.
- Keep the raw answer and the citations. A count without the sentence and the source behind it is not evidence.
- Re-run the identical set at day 60. A change in the answer without a change in the inputs is noise. A change that lines up with the work is evidence, not proof.
What to do with the result
Start with the questions where a competitor appears and you do not. Read the cited sources. If a directory or review site is doing the work, update that page; if a community thread is doing the work, participate honestly; if your own page is cited but the answer is wrong, correct the page first. A new blog post is rarely the first move.
The deliverable is a decision a marketer can defend: which question to fix, which page owns it, what success looks like, and when to re-check. That is why the diagnostic is a report, not a login. The synthetic 8-page sample shows the structure.
Trackers, audits and the difference
A tracker gives you a recurring number and a dashboard. It is useful if you already know which questions matter and you want to watch them weekly. An audit is a one-time reading with a source classification and a plan. Most teams need the audit first, because a trend without a baseline is just a screenshot with a subscription. For the vocabulary, see AEO vs GEO, ChatGPT brand monitoring and AI search metrics.
VisibleDiagnostic is the audit: 10 questions, 30 dated answers, three competitors, a source audit, three actions and a day-60 re-run. The free check on the homepage is three questions and one run; it is a demonstration, not the paid measurement.
Questions
Is AI visibility the same as an AI ranking?
No. Logged-out sessions do not expose a stable ranking comparable across users and runs. AI visibility is a record of what the answers said in a defined sample: mention rate, recommendation rate, position, citations and variance.
How many answers do I need before I can trust the result?
One answer per question is a demonstration. Three runs per question over several days shows whether a brand is consistently absent, consistently present, or moving. The diagnostic uses 10 questions and 30 dated answers.
What should I do after the audit?
Start with the questions where a competitor appears and you do not. Read the cited sources; fix the page that is actually doing the work. Re-run the identical questions at day 60.
Run the free three-question check · See the $850 diagnostic · Content audit vs AI source audit