How we measure what AI says about a property
Every AuraScope result comes from the same instrument: a fixed set of traveller questions, asked of ChatGPT, Gemini and Perplexity under recorded conditions, and asked again after anything changes. This page sets out what we commit to, what a result can and can't show, and what we keep private.
Our commitments
The question set is fixed before the baseline.
Questions are chosen for the property's real markets, guest intents and competitors, then frozen. Changing a question starts a new series. We do not splice old and new together.
Conditions travel with every answer.
Each answer is stored with the engine, the model version where the engine shows it, how it was accessed, where it was asked from and when. A change to any of these is a change to the instrument, and is logged as one.
One answer is never a result.
Engines give different answers to the same question from one run to the next. We report a change only when it holds across repeated runs. We show the spread alongside it.
Everything else that changed is on the record.
Model releases, changes to how engines retrieve sources, changes to our own system and the property's own work sit on one dated log, so any movement can be read against what happened that week.
Comparison before credit.
A result is only credited to the property's work when a comparison supports it. Until our comparison sets are running for a property, its results are labelled directional.
No single score.
We do not fold results into one visibility number. A composite hides which engine moved, on which question, and why.
What we record
For every answer, four things.
Whether the property appears, and how.
- Missing
- Not named.
- Mentioned
- Named, without being suggested.
- Recommended
- Suggested as a place to stay.
- Cited
- The property's own site or listing is given as a source.
What the answer says about the property.
Facts, descriptions and claims, so wrong or out-of-date details can be found.
Which sources shaped the answer.
The types of site and the domains cited.
Where the answer sends the guest.
Towards an OTA, the property's own site, or neither.
What a result can and can't show
| Rung | What it tells you | What it can't tell you | Status |
|---|---|---|---|
| Answers | What it tells youWhether the engines name the property, how, and on whose evidence | What it can't tell youWhy the answer changed | Built |
| Crawling | What it tells youWhether AI crawlers can reach and fetch the property's site | What it can't tell youWhether anything fetched was used in an answer | Built |
| Visits | What it tells youGuests arriving at the site from AI assistants | What it can't tell youWhether our recommendations caused those visits | Built |
| Bookings | What it tells youBookings and booked value from those visits | What it can't tell youWhether they would have happened anyway, without a comparison | In development |
Crawling and Visits run with our design partners only.
Where the evidence stops
If the evidence can't separate the property's own work from changes inside the engines, we say so, and the result is labelled directional. Visibility movement is never presented as booking growth or commercial lift.
Results describe the questions we asked, on the engines we measured, in the weeks we measured them. They aren't a ranking of the market, and we don't publish one.
What we keep private, and why
We publish the commitments on this page. We don't publish the question library, how answers are classified, the thresholds that separate a change from noise, or how comparisons are chosen.
That work is the product, and there's a second reason. A published question set stops being a fair test: once properties and agencies write to the questions, the answers measure the writing.
Partners see their own questions, every answer we collected for them, and the log behind each result.
Questions
Why only three engines?
They're the engines we can measure to this standard today. We add one when we can record its conditions the same way.
How many questions do you ask?
Enough to cover the property's markets, guest intents and competitors. The set is agreed with the property before the baseline and doesn't change during a series.
Can I see the raw answers?
Yes. Partners can read every answer behind every result.
Changes to this method
- Version 1.0 published.