How AEOscore measures AI visibility
How does AEOscore measure AI visibility?
Every number AEOscore publishes, in a report, a case study or a public comparison, traces back to this page. It sets out exactly how we measure AEO (answer engine optimisation): the prompt panel, the engines, detection, scoring, the publication standards we hold ourselves to, and the limits we do not pretend away.
How is the prompt panel built?
Each business we score gets a panel of buyer-style questions: it starts at 25 and can grow to 40 active prompts. The panel splits three ways: brand questions ask whether AI knows you, your reputation; non-brand questions ask whether AI would recommend you to a stranger, your competitive strength; and soft-brand questions sit between the two. Where a client provides Search Console data, panels are grounded in their real search demand.
Which AI engines does AEOscore track, and why not Copilot?
Answers are collected from ChatGPT, Claude, Gemini and Perplexity via their official APIs, and from Google AI Overviews and AI Mode via search-surface data. API answers can differ from what a consumer sees in an app, so we treat scores as directional and say so. We also tested Microsoft Copilot and have shelved it for now: its answers cannot yet be captured reliably enough to score honestly, and we would rather track six engines well than seven badly. It returns when it can be measured properly.
How does AEOscore detect a mention?
Every answer is parsed for whether the business is mentioned, whether it is cited as a source, and how prominently. Brand matching uses spelling variants and trading names only, never parent or group companies: a mention of your parent conglomerate is not a mention of you.
How is the AI visibility score calculated?
Prompts can run multiple times per engine per scan. Mention rates are weighted per prompt-engine cell with Wilson confidence intervals, so small samples cannot masquerade as certainty. The headline score runs from 0 to 100.
How does AEOscore know whether a fix worked?
When a fix goes live, we compare mention rates before and after, engine by engine, and give a plain verdict: improved, too early to tell, no clear change yet, or down since the fix. Volatile engines cannot swing a verdict on their own, and most fixes take 4 to 12 weeks to show in the answers, so we say “too early” more often than a dashboard would like.
What are AEOscore's publication standards?
We never publish a comparison or a named-brand claim from a scan window in which any engine returned errors. We archive the raw answers behind every published number. We phrase findings as what our sample found, not as universal fact. Any named brand in a public piece gets 48 hours of private right of reply before publication.
What are the limits of the methodology?
Engines change weekly and the sample is finite, so an AI answer is never a promise of customer behaviour. The score exists to make change visible.
What's your AEO score?
Send your domain and a named expert will run it through all six engines, score the results and email your report within 1 to 2 working days. Free, no card, no signup.
- ChatGPT
- Claude
- Gemini
- Perplexity
- Google AI Overviews
- Google AI Mode