How we measure AI visibility.
A transparent account of what BlindSpot measures, how the two measurement layers relate, and what you can verify yourself.
Two dimensions, two scores.
We track how AI engines see hotels along two independent axes.
Unbranded queries like “best hotels in Singapore for business travelers”. Measures whether you appear in the answer and where in the list.
Direct queries about your property. Graded 0, 50, or 100 depending on whether the engine recognizes you and recalls verifiable facts.
Why our scan and your browser can show different things.
Modern AI engines respond differently depending on how they are queried. The same hotel can appear differently in our scan and in your browser, and each reading accurately reflects the AI engine’s behavior at that specific layer — they measure different things.
We query each engine in a controlled, repeatable way that captures its base knowledge of your property. This is the layer that travelers experience through mobile assistants, voice integrations, embedded chat, and any AI surface that is not actively searching the web during the response.
When a traveler types a question into a consumer AI chat interface, several engines will run a live web search before answering. This pulls in fresh content, recent press, and current rankings that the base training data does not contain. The browser layer often reflects more current information.
Each layer captures a real audience. The engine source data layer is what your property looks like to voice assistants, mobile shortcuts, embedded AI, and any user query that does not trigger live web grounding. The browser layer is what a curious traveler typing into a consumer chat interface sees.
If you open ChatGPT, Gemini, Claude, or Perplexity right now and ask about your hotel, you are seeing the browser layer. Our subscriber and bespoke tiers include the browser layer directly. Our free briefing and cohort reports use the engine source data layer, which is what makes them comparable across thousands of properties at once.
If you check our results against your own browser session and the answers differ, that discrepancy typically reflects the layer gap described above. The two layers measure different surfaces of the same engine, and a difference between them does not by itself indicate an error in our measurement or a flaw in the AI engine. Our testing indicates that the macro patterns (which hotels dominate, which are absent at the cohort level, how the market is moving) are generally consistent across both layers. Individual rankings vary.
The findings you can rely on, and the ones to take as directional.
- The dominant hotels in a market — the top one to three properties
- Concentration of AI attention — what share of mentions goes to the top of the list
- Cohort visibility (what proportion of hotels are absent from AI recommendations)
- Trend over time — week-over-week and month-over-month movement
- Relative ranking against your competitive set
- The exact rank of an individual hotel in a single response
- Which specific engine mentions you and which does not at any given moment
- Word-for-word brand descriptions returned by an engine
- Short-term swings smaller than a five-point movement
Discovery and brand scoring use weighted formulas (mention rate, position, fact-accuracy grading) applied identically across every property in every covered market. The methodology is held constant so trends are comparable.
What we track.
Engines tracked: ChatGPT, Claude, Gemini, Qwen, DeepSeek, Perplexity, Doubao, Meta AI. Languages cover the principal source markets for inbound APAC travel. Not every engine×language combination applies to every city — coverage matches real traveler population for that market.
Different tiers expose different layers.
Each product tier corresponds to a different combination of measurement layers and coverage. The deeper the layer, the closer to what an individual traveler experiences in their browser today.
Architecture is open, instruments are not.
We describe in detail what we measure and how we score it. We do not publish the specific query text, the model versions, the extraction routines, or the rate-limiting strategies. Those are instruments — if we published them, anyone could fabricate matching results, and our subscribers would have no way to tell our reading from a competitor’s replay of our public query.
This is the same posture that survey research, ratings agencies, and consumer-test publications take. The architecture is transparent; the instruments are protected so the measurement remains credible.
AI engines do not give the same answer twice.
Modern AI engines are probabilistic. The same question, asked twice by the same person, will often return somewhat different hotels in somewhat different order. Different users see different results based on prior conversations, account history, region, device, and time of day.
BlindSpot is designed for repeatable measurement under controlled conditions. We hold our questions and scoring constant so that trends are comparable across cities, engines, and weeks. That gives you a reliable directional read on how visible your property is and how it is moving — not a mirror of any single traveler’s session.
Questions about methodology, suggestions for additional coverage, or want to discuss what your readings mean for your property? Start with a free briefing, or write to research@blindspot.fit.
Methodology version 1.0 · Effective June 2026. We update this page when our coverage, scoring formulas, or measurement layers change materially. Historical versions are archived via the Internet Archive.