How we measure

aitani measures AI visibility by asking real questions to ChatGPT, Gemini and Google AI Overview from a Japanese IP address, using official APIs. Every measurement is logged raw and versioned so changes over time are comparable and reproducible.

The four-tier funnel

We evaluate visibility as a funnel with four levels. Knowledge: does the AI know your business exists? Category: is your business associated with the categories customers actually search (e.g. "family-friendly ramen", "halal", "open late")? Consideration: does the AI place you inside the shortlist it presents to the customer? Recommendation: does the AI pick you as the top or explicit suggestion? A gain in Recommendation without gains in Knowledge and Category is fragile — the whole funnel must move.

Question set

For each business we compose 45–50 customer questions covering demand across intent (browse, decide, book), audience (families, tourists, group, solo), constraints (allergy, budget, late-night), and language (Japanese, English, sometimes Vietnamese). The set is locked at baseline. Every re-measurement uses the same set, otherwise the comparison is meaningless.

How we ask

ChatGPT is queried via the OpenAI Responses API with web-browsing enabled, from a Japanese IP address. Google AI Overview and Google AI Mode are queried through DataForSEO with location and language forced to Japan. Gemini is measurable through the same provider but costs many times more per query, so we enable it on request — when it is off we record "not measured" rather than guessing a number. Each question is run several times, and we record the most common answer along with how much it varied between runs.

What counts as "recommended"

A mention ("Sakura Ramen is one of many ramen shops in Shibuya") is not the same as a recommendation. We only credit a hit when the business appears in an enumerated shortlist the AI presents in response to a decision-oriented question. This is stricter than most AI-visibility tools on the market.

White-hat principle

We only clean up and align real evidence: consistent hours across every source, structured schema, plain-text menus, factual business descriptions, honest photo captions. We do not inject fake reviews, purchased links, or misleading claims. Where we cannot get evidence, we mark the item "not measured" rather than guess.

Rate limits and spend caps

Each audit is bounded by a per-run spend cap (default USD 3.00) and a per-day account cap so no runaway API bill is possible. We use official APIs (OpenAI, DataForSEO, Google PageSpeed, Google Places) and do not scrape hidden endpoints. Terms-of-service compliance is a hard requirement.

Reproducibility

Every audit stores: the exact question set (versioned), method version, model version, IP region, timestamps, raw responses, and the derived score. If you dispute a result, we can replay the exact same query and show you the response.

For the full technical reference see llms-full.txt: llms-full.txt

Not sure it fits your business?

Write to us. A person reads and answers — including questions about scope, cost, or whether your industry is a fit at all.

Or start with the free check; it takes about two minutes.