How to measure AI visibility (and not fool yourself)

You fixed your data and you want to know if it worked. The honest answer is that AI visibility is measurable, but not with a single number you check once. It is a reading you take over time, per assistant. Here is how to do it without fooling yourself.

What "AI visibility" actually means

It is how often AI assistants mention, cite or recommend you when someone asks a question in your category. People shorthand it as AI share of voice: the percentage of answers in your category that name your brand, relative to everyone named.

Three words that get muddled and should not:

• Mentioned: the assistant names you in its answer.
• Recommended: it puts you forward as a good option, not just a name in a list.
• Cited: it uses your page as a source, usually with a link.

You can be recommended without being cited, or cited without being recommended. They are different wins, so track them separately.

The method

You do not need a lab. You need a repeatable panel of questions:

1. Write buyer-intent prompts. Ten to twenty to start, the real questions your customers ask: "best waterproof hiking boots under 150", "which running shoe for flat feet". More prompts give a steadier reading.
2. Ask each assistant. ChatGPT, Perplexity and Gemini retrieve and rank differently, so run the same prompts on each and record per assistant. The overlap between what ChatGPT and Perplexity cite is small, so an average across them hides more than it shows.
3. Record what you see. For each answer: were you absent, mentioned, recommended or cited? If cited, which page? Where in the answer did you appear? Being named first carries far more weight than fifth.
4. Repeat and watch the trend. Run the same panel on a schedule. One run is a snapshot, not a verdict, and the sources assistants pull from shift a lot month to month.

The trap: reading too much into one run

Answers vary between runs of the same prompt, so a single good or bad result is noise. Watch consistency: how reliably you show up across repeated tests of the same question. Low consistency means the assistant has a weak link between your brand and the topic, which is a data and content problem, not luck.

Where this connects to the fixes

Visibility is the scoreboard; structured data, identifiers and feeds are the game. If the number is low, the fix is usually upstream: missing schema, weak identifiers, or a retrieval crawler that cannot read you. Start by finding those gaps with a free audit, then track the trend as they close.

This is exactly what our GEO monitoring does: it runs a panel of your category questions across assistants and tracks whether you are mentioned and recommended over time, so you are not copy-pasting prompts by hand. See the Agency and Business plans, and for the fundamentals, what AEO is.