There is no rank to check inside an AI answer. The same question asked twice can return two different paragraphs with different sources. That makes AI search visibility harder to measure than a Google position, but it is measurable. What works is a fixed set of questions, asked on a schedule and recorded the same way every time.
What AI visibility means
AI visibility is how often a brand appears inside the answers that ChatGPT, Perplexity, Gemini and Google’s AI Overviews write, and what those answers say about it. It covers three separate things:
- Mentioned. The brand is named in the answer.
- Cited. The answer links to the brand’s own site as a source.
- Described correctly. What the answer says about the brand is accurate.
A brand can be mentioned without being cited, and cited while being described wrongly. Each of those needs different work, so measure them separately. The work of improving them is often called GEO, and the overlap with classic search is larger than the acronym suggests. I covered that in GEO vs SEO.
Why rank tracking does not transfer
A Google ranking is one position for one query in one location. An AI answer has no fixed position. It changes with the wording of the prompt, with the conversation before it, and between runs of the identical prompt.
So a single screenshot proves very little. Visibility here is a rate: out of a set number of questions and runs, how often you appeared.
Step 1: Build a prompt set
Write 20 to 30 questions your buyers actually ask before they contact someone like you. Pull them from sales calls, support emails and the queries in Search Console that already bring you traffic.
Split them into three groups:
- Category questions. “What should I look for in a B2B payments provider?” No brand named.
- Comparison questions. “Best alternatives to [competitor].” “[You] vs [competitor].”
- Brand questions. “What does [your company] do?” “Is [your company] a good fit for enterprise teams?”
Category questions tell you whether you are in the conversation at all. Brand questions tell you whether what is said about you is true. Most companies only check the second group, and it is the less important one.
Freeze the set. If the wording changes every month, the trend means nothing.
Step 2: Decide where and how often
Pick the engines your buyers use. For most B2B companies that means ChatGPT, Google AI Overviews and Perplexity, with Gemini and Copilot where your market leans that way. Keep the counts per engine, because the same site can be cited seven times more often on one than on another.
Run each prompt at least three times per engine, in a clean session with no history and no memory. Once a month is enough for most sites. Weekly is worth it only during a launch or right after a significant change.
Step 3: Record the same fields every time
A spreadsheet is enough. For every run, record:
| Field | What to write |
|---|---|
| Mentioned | yes / no |
| Cited | yes / no, and which URL |
| Position | first brand named, among others, or last |
| Accuracy | correct, outdated, or wrong, with the wrong claim quoted |
| Competitors named | every brand that appeared |
| Sources cited | every domain the answer linked to |
The last column is the one most people skip, and it is the most useful. It shows which third-party sites the engine trusts for your category. Those are the sites where a correct description of you matters most.
The numbers worth reporting
- Mention rate. Runs where you were named, divided by total runs. Report it per prompt group.
- Citation rate. Runs where your own site was linked.
- Share of voice. Your mentions against the mentions of the three or four competitors that appear most often.
- Accuracy issues. A count, plus the list. One wrong claim that repeats across engines is a priority on its own.
Leave out anything that pretends to be more precise than the method. A “visibility score” of 67.4 built on 30 prompts suggests a precision the sample does not have.
Signals outside the answer
The prompt set measures presence. These measure whether presence turns into visits and enquiries:
- Referral traffic in analytics. Filter sessions by referrer for
chatgpt.com,perplexity.ai,gemini.google.comandcopilot.microsoft.com. Part of this traffic arrives with no referrer and shows up as direct, so treat the number as a floor. - Landing pages from those referrers. They show which of your pages the engines actually send people to.
- The contact form. A “how did you find us” field with an AI option costs nothing. It is the simplest way to see enquiries that started in an AI answer, like the first ones reported in this law firm case study.
- Search Console. Google’s documentation on AI features says this traffic is counted inside the Performance report, under the Web search type, with no separate line. Impressions that rise while clicks stay flat on informational queries are often an AI Overview sitting above the result.
Where tools help
Software that runs prompts at scale saves time once the prompt set is stable and you are tracking many markets or many brands. It does not replace deciding which questions matter, and it cannot tell you whether a description of your company is accurate. Build the manual version first. Then you will know what you need a tool to automate.
What to do with the results
Each pattern points to a different fix:
- Not mentioned in category questions. The engines do not connect you with the category. This is entity and third-party work: consistent descriptions of what you do on the sites the engines cite.
- Mentioned but never cited. Your own pages do not answer the question directly enough to quote. This is content structure: a direct answer under each heading, then the detail.
- Described wrongly. Find the source of the wrong claim in the “sources cited” column and correct it there. Correcting your own site alone rarely changes the answer.
Three months of consistent records give you a real trend. That is less than the dashboards promise and more than most companies know about themselves today.
If you want this set up for your market, and the fixes carried out once the numbers are in, that is the core of my AI search optimization work. Or tell me which questions matter in your market and I will tell you what I would measure first.
The first measurement for this site, 31 October 2026: 20 prompts, four engines. Zero mentions on the 18 generic prompts, accurate answers on the 2 brand prompts, and one engine that could not find the domain at all because the host was blocking its crawler. A first baseline usually looks like that. The point is the second one.
FAQ
What is AI visibility?
AI visibility is how often a brand is named or cited inside AI-generated answers from ChatGPT, Perplexity, Gemini and Google’s AI Overviews, and whether what those answers say about it is accurate.
How do you measure AI search visibility?
Use a fixed set of 20 to 30 buyer questions. Run each one several times per engine in clean sessions, once a month, and record whether you were mentioned, whether your site was cited, which competitors appeared and whether the description was accurate. Report the results as rates.
What is the difference between GEO and SEO?
SEO gets a page into the list of search results. GEO gets a brand named inside the answer an AI engine writes. Most of the technical groundwork is shared. The full comparison is in GEO vs SEO.