How we measure

Everything below is what actually happens. Where a number has a limit, the limit is stated next to it.

Last reviewed 30 September 2026.

What we ask

10 questions on AI Booster + AI Visibility report, built from fixed templates for your category and the services you offer, never improvised by a model, so the same question is worded the same way every time it is asked. About six in ten are about your neighbourhood and the rest about your city.

Your set is then frozen. We do not swap questions month to month, because a changing question set makes month-over-month comparison meaningless. If a question does change, your AI Visibility report says so and the trend for that question restarts.

Questions are never matched to a business by keyword. A competitor’s report once scored a massage business against mental-health-therapist questions and reported the resulting zero as a visibility problem. That is a category-matching bug presented as a finding, and it is the specific failure our review step exists to prevent.

Who we ask, and how often

Every question, every Monday, Wednesday and Friday on AI Booster + AI Visibility report, on each assistant your plan covers: ChatGPT and Google AI (Gemini).

We only count an answer when the assistant actually performed a web search for it. An assistant answering from memory is describing the internet as it was during training, not as it is today. When a search does not happen we ask once more, and if it still does not, the answer is stored and flagged rather than counted.

We do not list an assistant we cannot genuinely query. Some well-known ones have no public interface that performs a real search, so they are absent from our plans entirely rather than listed and quietly skipped.

What the numbers mean

Every count states what it is out of. The headline is the share of answers that named you: “30%, 12 of 40 answers” means that of the 40 answers that came back over the period, 12 recommended you. An answer that did not come back is left out of both numbers rather than counted as a miss.

We never reduce this to a single score. A score out of 100 is more shareable and tells you less: it hides which question moved, which is the only part you can act on.

Where an AI Visibility report uses an assumption, it is labelled as an assumption on the page. The numbers it is applied to are arithmetic on answers we actually checked.

What this does not tell you

Assistants personalise by location and history as well. Wording varies between checks and AI do not answer identically twice, so every count reflects the checks we made, not what any one customer will see.

We cannot tell you how many people asked, or how many became customers. Anyone claiming to is inferring it. Where we show a scenario, it is labelled.

And we cannot guarantee an outcome. We can tell you what assistants say now, what the businesses being named have that you do not, and what to change.

You can check our work

Every answer we count is stored exactly as the assistant produced it, and your scan log lists all of them with the time, the assistant, whether a real search happened, and a link to the raw answer. If a number looks wrong, you can read the thing it came from.