How measurement works
Every month, the certified reading of what AI systems say when they answer travel questions. One destination at a time.
What AI See You measures
Recommendation behaviour must be measured, not assumed.
Rankings are only meaningful when the prompt, model, context and evidence are known.
The goal is not to declare a universal truth from one answer. The goal is to produce a defensible view of how AI systems behave under defined conditions.
What sits behind every number
Evidence must be preserved close to the original response.
We store what the machines said verbatim and permanently; nothing is paraphrased, summarised or reconstructed.
One verbatim machine response, reproduced in full and unedited.
Without evidence, a ranking is an assertion. Without repeatability, it is only an observation.
The certified monthly Pulse
A month of measurement enters commercial use only when it passes written certification criteria.
Months that fail are not published quietly; they fail visibly, with reasons.
There is no response caching by design: each execution is an independent observation of a nondeterministic system, and caching would fabricate agreement that the instrument exists to measure.
Measurements, once certified, are permanent.
Traveller Categories
A Category is a kind of demand the platform asks about.
Categories are never ordered across destinations, because a Luxury Category in Sydney and one in Hobart sit on different populations.
Order is not a judgement in this product.
The AI platforms
Every measured prompt executes across the full active panel.
The same rendered prompt text goes to every platform; the platform does not tailor a prompt to flatter a model, because a per-platform prompt would make cross-platform comparison meaningless.
The platforms measured are named on each property page, taken from the certified reading itself and never from a list kept by hand.
Measured again, and again
Repeatability matters because one AI response is an observation, not a pattern.
Repeatability matters because a single response may reflect temporary model behaviour, prompt phrasing, sampling variation or incomplete context.
Two readings is a comparison, not a series.
Coverage expands with each monthly Pulse.
What AI See You does not claim to measure
The boundaries below are the ones a sceptical reader should hold us to.
- Not a review of your property
- AI See You measures what AI systems recommend. It does not evaluate the properties themselves.
- Not advice
- The product observes, measures, explains and evidences. It never recommends an action.
- Not an explanation of why
- It never interprets. No "which suggests", no "driven by", no cause. Those need a model to reason about why, and we do not measure why.
- Not a claim about how the assistants think
- Showing our prompt set says what we asked. We are evidencing our method, not their behaviour.
- Not a gap quietly filled
- Where our data cannot answer your question, the product says that, in the interface, where it is least convenient for us.
- Not for sale
- No commercial relationship, of any kind, can influence a published value, a ranking, an alert or a report.
- Not a different reality for different customers
- Personalisation changes scope and framing, never values; no customer, at any spend, sees a different reality.
- Not externally accredited
- We hold no external compliance certifications yet and claim none.
Which properties appear
A property appears here when its destination is one we measure and the property was inside the measured set for the latest certified reading.
From the visitor's side there is one fact: we measure this destination, or we do not.
Currently measuring 75 destinations across 4 countries.
Coverage expands with each monthly Pulse.