How we measure something that isn't supposed to be measurable
AI answers are probabilistic. Any tool claiming a single exact "AI ranking" is selling you a screenshot. Here is precisely what we do, what the numbers mean, and the four things we refuse to claim.
Time. Every hour AI answers your market without naming you is an enquiry you never hear about.
Doing this yourself means writing 30 buyer questions, asking each of them in ChatGPT, Gemini, Claude and Perplexity, logging every name, opening every cited page, and then working out who was named more often than you. That is days of work, and it is stale the week after. We run it end to end in minutes and hand you the finished evidence — every answer, every competitor, every source page — so you can spend your time fixing your website instead of measuring it.
- 1You give us your business name and website. Nothing else is needed.
- 2We generate the fixed buyer question set for your industry and city.
- 3Every question goes to all four engines in fresh, logged-out sessions.
- 4We record each answer word for word, plus each page the engine cited.
- 5Names are merged to one business per key, so counts are not double-billed.
- 6You get the report in your inbox, with the raw CSV attached.
1. Prompt generation
From your industry and region we build 20–26 prompts across six intent families: broad discovery, suburb-specific, price-sensitive, urgent, review-led and treatment or service specific. The set is fixed per industry, so repeat runs are comparable.
2. Independent engine runs
Each prompt is sent separately to ChatGPT, Perplexity, Claude, Gemini. Fresh sessions, no chat history, no personalisation, no account signals. Region is supplied in the prompt text only — we never spoof location.
3. Name and position extraction
Every business name in each answer is recorded with its position in that answer. Aliases (trading name, agent name, group name) are merged. Prompts where an engine returns no names are excluded from averages and flagged.
4. Visibility scoring
Visibility score = appearance rate × position weight × engine breadth, normalised to 100. A business named first by all four engines on every prompt scores 100. The raw components are always shown next to the score.
Four honest limitations, printed in every report
We can't see what any individual customer was told
Answers vary by account history, device and timing. We measure the pattern, not one person's screen.
Citations are indicative, not exhaustive
Some engines cite sources explicitly, some don't. Where citations aren't given, we identify the likely source set and label it as inferred.
We can't guarantee movement
We measure, we do not promise placement. Anyone promising a top-three AI position is guessing.
Small markets produce small reports
If fewer than five businesses are ever named in your category and region, we tell you and refund on request rather than pad the document.