AI visibility
How to measure it, and how much to trust the number.

AI visibility is how often ChatGPT, Perplexity, Gemini and Google's AI features name, cite or recommend your company when someone asks a question you should be answering. You measure it by asking a fixed list of real buyer questions, repeatedly, and recording what comes back. The result is a trend you can act on, not a precise score.

What to record

For every prompt, on every platform, on every run, write down:

• Whether your company is mentioned.
• Whether your site is cited as a source.
• Whether you are recommended, as opposed to just listed.
• Which competitors appear, and in what order.
• Which sources the answer leans on, yours and other people's.
• Whether what it says about you is accurate.

Mentioned, cited and recommended are three different things. An engine can name you without linking you, or cite your page while recommending a competitor. Keeping them apart is most of the value.

Build the prompt set

Use questions your buyers actually ask, in their words, not your keywords:

• Category: "best [what you sell] for [kind of company]".
• Problem: "how do I fix [the thing you solve]".
• Comparison: "[you] vs [competitor]", and "[competitor] alternatives".
• Branded: "what does [you] do" and "is [you] any good".

Keep the list fixed so months can be compared. When your offer changes, add a labelled second list instead of editing the first.

How many prompts, how many runs

Nobody agrees, and that is informative. On 16 October 2026 we asked ChatGPT (GPT-5.4) and Perplexity (Sonar Pro) the same question: "How do I measure my company's AI visibility in ChatGPT and Perplexity? What should I track and how reliable are the numbers?" ChatGPT suggested a panel of 50 to 500 prompts. Perplexity suggested 20 to 50. Both said to run answers repeatedly, because the same prompt can produce different answers.

We would start with 20 to 40 prompts and run each several times a month. That is a judgement call, not a standard, and more prompts reduce noise only if they are real buyer questions.

How reliable is it

Both models gave essentially the same verdict, and we agree with it: directional in aggregate, weak for any single answer.

It is good enough for:

• A trend over months.
• Comparing yourself with named competitors.
• Finding the prompts where you are absent and which sources win them.

It is not good enough for:

• An exact rank. Position inside a generated answer moves around.
• A number like "we own 17% of ChatGPT". There is no such denominator.
• Conclusions from one prompt on one day.

Google adds a caution of its own: it says no third-party tool has access to its internal ranking or AI systems. Any tool, ours included, is sampling answers from outside.

Free places to look first

Google's guide to generative AI features points to a Generative AI performance report in Search Console, which shows how your pages appear in Google's AI features. Check it before paying for anything. It covers Google only, not ChatGPT or Perplexity.

Your analytics will also show visits whose referrer is an AI product. That is the number closest to money, and it is usually small and easy to miss.

How we do it

For each client we ask the models what they answer about the sector and whom they cite, at the start and every month after. That is a sample, with the limits above, and we report it that way.

It is what the generative engine optimization work reports against. For the terminology around it, see GEO, AEO, AI SEO.

The part that costs us

If your site cannot be crawled, or says little that is worth quoting, a visibility report will only confirm that. A tracker does not fix either problem, and we would rather tell you that before you pay for monthly reporting. Fix access and the pages first, then measure.