The one rule: the prompt must not contain your name
"What is acme.com?" is not a visibility test. Engines with live search will simply fetch the URL you handed them and describe it back to you. The answer is always yes, and it tells you nothing about whether you get recommended.
The only measurement that means anything is the unprompted question — the thing a real buyer types, with your brand absent. "Best accountant in Richmond BC", not "is my accounting firm any good".
Branded prompts are still worth running, but as a different instrument: they tell you whether the engine describes your category, city and specialism correctly. That is a positioning diagnostic. Reported as visibility, it is a flattering fiction.
Three mention states, not two
Counting "did it mention us?" collapses two opposite situations into one number. There are three states and the fix differs completely for each.
| State | What it looks like | What it needs |
|---|---|---|
| Absent | The engine does not know you exist. | Third-party citations. The ordinary starting point. |
| Present and correct | The engine describes you accurately. | Defend it. Grow share of voice against the names beside you. |
| Present and wrong | The engine confidently describes a business that does not exist. | Most urgent of the three. An authoritative entity record and consistent descriptions everywhere, quickly. |
A naive mention count scores the third state as a win. It is the opposite of one — the engine is actively misinforming buyers.
What to count
Four numbers, and they answer different questions. Mention rate is the core one; the rest exist because a single percentage hides too much.
| Metric | What it answers |
|---|---|
| Mention rate | Across every answer to every unprompted question, how often were you named. |
| Reach-weighted mention rate | The same mentions, weighted by roughly how many people use each engine — a mention on the largest engine is worth more than one on the smallest. |
| Share of voice | Of all the brands named across those answers, what share was you. |
| Cited link rate | How often an engine listed your site as a source. Only measurable where the API returns citations. |
Not every engine exposes its sources: of the 11 we measure, the Chinese ones largely do not return citation URLs at all.
Why a changed engine list breaks the comparison
If you measured nine engines last month and ten this month, the two readings were taken with different rulers. A new engine starts from nothing, so the headline average falls even when nothing about your brand got worse.
The fix is to pin the specification — questions × engines × repetitions — and state it on every report, then show the like-for-like reading on the old basis alongside the new headline whenever the spec changes. We pin ours and print it as, for example, 23p×10e×2r.
The same applies to question count and repetitions. Changing how many questions you ask changes the denominator, and a percentage whose denominator moved is not a trend.
Common questions
How often should I measure?
Monthly is the floor, because AI answers drift on their own and a single reading cannot tell drift from change. We run twice a month, on the 1st and the 16th. More often than that mostly buys noise, unless you have just shipped something and want to see it land.
Can I just use a free AI visibility checker?
They answer a narrow question — do any engines mention us at all — and many of them do it with branded prompts, which as above is not a visibility test. Useful as a smoke alarm, not as a baseline you will be judged against later.
Why do the numbers move when I changed nothing?
Because the engines changed. Answers are non-deterministic and the underlying models and indexes update without notice. This is why the reading is a trend line with a fixed specification rather than a single score, and why a small month-to-month move is not a result.
Scope and limits
We publish the full method — question design, the fixed specification, what we refuse to claim — at /method, and every report we issue states the specification it was measured under.
Last updated 2026-09-26.
Is your brand in the answer?
Avowd measures how eleven AI engines describe and recommend a brand, in English and Chinese, then does the work to change it. The free check takes about thirty seconds.