Short answer: the AI visibility score is the share of AI answers that name the business, across a fixed set of buyer questions, six AI assistants and repeated runs, recorded with the date. A fix sprint uses 20 questions × 6 assistants × 3 runs = 360 answers.
The score
AI visibility = answers that name the business ÷ total answers × 100.
| Free check | Fix sprint and re-checks | |
|---|---|---|
| Questions | 3 | 20 |
| Assistants | ChatGPT, Google AI Mode, Gemini, Perplexity, Claude, Grok | The same six |
| Runs per question | 1 to 3 | 3 |
| Answers scored | Up to 54 | 360 |
| What it's good for | A first signal and the 3 biggest fixes | A baseline you can track month to month |
What is recorded for every answer
| Field | What it means |
|---|---|
| Mentioned | Is the business named anywhere in the answer? |
| Position | Where it appears in the list, if the answer is a list |
| Framing | Recommended, neutral, hedged, or recommended against |
| Accuracy | Are the stated facts right: prices, ratings, location, services? |
| Named instead | Which competitors appear |
| Sources cited | Which pages and listings the assistant relied on |
How answers are collected
- A new chat for every run, logged out or in a clean account, so personal history doesn't skew results.
- Web search turned on where the assistant offers it.
- The city or area written into the question, rather than relying on the assistant to guess location.
- The same wording every time, and the date recorded with each answer.
- Answers saved, so you can read them yourself.
Mentioned is not the same as recommended
AI visibility is a ladder. A page can be read by an assistant without being shown, cited as a source, mentioned by name, or recommended as one of the best options. Reports show mentions and recommendations separately, because a business can be cited often and still be left off the shortlist.
A note on the case studies
The B2B platform result (4.7% to 57%) was measured across three assistants: ChatGPT, Gemini and Perplexity. New checks cover six. Both case studies use the same question set before and after the fixes.
Limitations
- AI answers are not deterministic. The same question can get a different answer a minute later. Repeated runs reduce this noise but don't remove it.
- Personalisation and location can change what a real buyer sees.
- Assistants change. A model update can move results for reasons unrelated to any fix.
- Small changes can be noise. A few points' difference between two checks may mean nothing. Look at the trend across repeated checks.
- Before and after is not proof of cause. The fixes are the most likely reason for a change, but other things change at the same time.
Want this done for you? The free one-page check shows what six AI assistants say about your business, who they name instead, and the three fixes that matter most. Sent within 48 hours.