Measurement

Citations aren't recommendations: what to actually measure in AI search

Being cited by AI, being mentioned by AI, and being recommended by AI are three different outcomes — and only the last one puts a client in front of you. Most visibility tools count the wrong one, and a single spot-check counts nothing, because AI answers change from run to run. What you want to know is how often you're named as the answer, per engine, across many runs.

Updated July 26, 2026Reading time 5 min

Three outcomes people call the same thing

Ask most tools whether you’re “visible in AI” and you’ll get a number. The problem is that the number usually blurs three very different things:

Only the third one sends you a client. You can be cited as a source the engine read and still not be the business it recommends. You can be mentioned in a list of five and lose to the one described as the best choice. A visibility score that counts citations, or counts any appearance of your name, will tell you you’re doing fine while the recommendation goes to someone else.

Why a single check tells you nothing

Here’s the part almost every do-it-yourself check gets wrong: AI answers are not stable. Ask the same engine the same question three times and you can get three different shortlists. The model samples; the web it reads shifts; the answer moves.

So a screenshot of ChatGPT naming you once is not a position — it’s one roll of the dice. And a screenshot of it not naming you once isn’t a verdict either. What you actually want is the rate: across many runs of the question your customers really ask, how often are you the answer? That number is stable enough to act on. The single result is just noise dressed up as a finding.

The share-of-voice trap: a popular way to report AI visibility quietly drops the answers that name no business at all from the denominator — which flatters everyone left in it. If half the answers to your money question name nobody, a metric that ignores those reads far higher than reality. Count every run, including the ones that recommend no one, or the number lies to you.

What to measure instead

The honest metric is narrow and boring, which is why it works: the rate at which you’re recommended for a specific customer question, on a specific engine, across repeated runs — counting every run. Reported per engine, never blended into one score, because the engines disagree and a blended average hides exactly the surface you’re losing.

That’s the whole reason our reports run every high-intent question at least three times on each of the six engines and hand you the rate, with the runs on file. Not a citation count. Not a single lucky screenshot. The number that answers the only question that matters: when your next customer asks who to hire, how often is the answer you?

Common questions

Isn't getting cited by AI the goal?
It's a means, not the goal. A citation is a link or source the engine used; a recommendation is your business named as the answer to 'who should I hire.' You can be cited as a source and still not be the business the engine recommends — and it's the recommendation that sends you the client. Measure whether you're named, not just whether a page of yours was read.
How many times should you run a query before trusting the answer?
More than once, always. AI answers vary from run to run on the same question — the same prompt can name three different businesses across three runs. A single check tells you what happened once, not what a customer will typically see. We run every high-intent question at least three times per engine and report the rate, because the rate is the real signal and the single result is noise.

The report is free

See how often AI names you — not just links you.

Twenty-five real customer questions across six engines, three runs each, every run on file.

Get your AI visibility report