How aisle measures AI visibility

Written for humans. No score without a method behind it. Methodology version 1.1, last updated August 2026.

What we measure

For each product category we maintain a library of real buyer questions (“best air purifier for allergies?”, “best quiet running shoes for city streets?”). Every week we put those questions to four AI surfaces: ChatGPT, Claude, Google AI Overviews, and Google AI Mode, and record which brands each answer recommends.

Why we ask the same question many times

AI answers vary run to run: the same question can mention your brand once and skip it the next time. A single check is noise. We ask each question multiple times and report your visibility as a percentage with a 95% confidence range. If two numbers’ ranges overlap, we say so instead of pretending one is bigger. We only alert you when a change is statistically real.

Markets are never mixed

Shoppers in different markets can receive different AI answers, so we sample each selected market separately using market-specific locations, phrasing, price bands, brands, and device assumptions. Each market produces its own result and confidence range; those facts are never averaged into one global score.

What we honestly can’t measure

• Personalized answers. A person’s own chat history shapes their answers, and no tool can see into private accounts. We measure the logged-out baseline, the same way SEO rank trackers have always worked. We partially cover this with persona-flavored questions (“…for a beginner”, “…on a budget”).
• App vs. API differences. Our ChatGPT numbers come from OpenAI’s API with web search, a close proxy for the app but not identical to it. We plan periodic calibration checks against the real app and will publish the agreement rate here once that first study has run. It hasn’t yet.

What still needs to happen before this is fully live

Aisle’s sampling and aggregation engine (multi-provider sampling, Wilson-interval aggregation, per-market separation) is built, tested, and has run against real API data. What hasn’t happened yet: the calibration study referenced above. Gemini is temporarily outside the active surface set rather than fully removed. Google Search grounding, needed to keep its answers comparable to the other surfaces, isn’t available on our current billing tier. Demo numbers elsewhere on this site are clearly labeled and do not describe a real company.

Questions?

Ask us anything via the early-access form on the homepage. Methodology questions get answered first.