What is AI visibility?

The short definition

AI visibility is whether an AI assistant names your business when someone asks it for a solution like yours. Not whether the assistant knows you exist when prompted with your name — whether it brings you up unprompted, to a buyer who has never heard of you.

The distinction matters because the two are measured completely differently and usually give opposite answers. Ask ChatGPT “what does Acme Software do?” and it will probably tell you, more or less accurately. Ask it “what’s the best warehouse management software for a mid-sized distributor?” and Acme may not appear at all. The first is recognition. The second is visibility, and it is the one that decides whether you are in the running.

Three outcomes people collapse into one

Most discussion of this topic treats “my brand appeared” as a single event. It is three, and they are worth very different amounts.

A tool that reports “you were mentioned in 60% of answers” without separating these is reporting a number that could mean almost anything. Being named eleventh in a list is a mention. So is being the single recommendation. Averaging them together produces a figure that moves for reasons nobody can explain afterwards.

Citations are the most useful of the three for deciding what to do next, and the most often ignored. If four answers in your category all cite the same comparison article, that article is doing more for your competitors than their homepage is.

Buyer questions are the ones that count

A buyer question is one a potential customer would type without knowing your name. “Best gym management software in India.” “Who builds MVPs for early-stage SaaS?” “How much does an ERP implementation cost?”

The test is simple: could the answer be written without naming any business at all? If yes, it is not a buyer question and it cannot measure visibility. “How do cloud applications sync across devices?” has a perfect answer that names nobody. You cannot be missing from it, so counting it tells you nothing.

The questions that quietly measure a market you aren’t in

The subtler failure is a question that is comparative but belongs to a different category. We watched a small software firm’s question set return Infosys, TCS and Tech Mahindra — real companies, genuinely named, and completely irrelevant to a twelve-person team. The question was measuring the Indian IT services market, not theirs.

This is why the businesses named beside each question matter as much as your own result. If you do not recognise the companies in the answer, the question is pointed at the wrong market and should be dropped.

What can and cannot be measured

“AI search” is not one surface. The engines answer differently, and some are not reachable at all.

That last row is the one the industry is least honest about. Google publishes no API for AI Overviews. Measuring it means buying scraped search-results data from a vendor, with the accuracy and terms-of-service questions that implies. Any tool claiming to track it is doing that, and should say so.

The difference between model knowledge and search-backed answers is also worth keeping separate rather than averaging. One tells you what the model has absorbed about you over time; the other tells you what it finds today. A new company will be absent from the first and can still win the second.

Why a single score misleads

Every tool in this space, ours included, can produce a number between 0 and 100. It is a convenient headline and a poor instrument, for a reason that has nothing to do with the maths: the score moves when the question set moves.

The figures worth watching are plainer and harder to game: how many buyer questions mentioned you, how many named a competitor instead, and which of those changed since last time. Three counts you can check by opening the answers.

What AI visibility does not tell you

• It is not traffic. Being recommended does not mean anyone clicked. AI answers frequently satisfy the question without a visit at all.
• It is not causal. If your numbers improve after you publish a page, the honest claim is that they improved after you published a page. Models retrain, indexes refresh, and answers drift on their own.
• It is not stable to the decimal. Ask the same question twice and the wording differs. Whether the outcome differs is an empirical question, and worth testing before you act on small movements.
• It is not a ranking. There is no position 1 through 10. A business is named or it is not, and where it sits in a list is weak evidence at best.

None of that makes it unmeasurable. It makes it measurable in counts and comparisons rather than in a single moving number — which is the same conclusion most people reach about SEO metrics eventually, just earlier in the cycle.