Home / Share of Model
The metric
Share of Model
The percentage of relevant questions in which your property is actually named in an AI-generated answer. Measured per property, per engine, against a named competitive set, on a query set fixed for the engagement.
Why the metric exists
Every previous shift in hotel marketing arrived with an instrument. When OTAs became dominant, hoteliers could see the commission line. When Google mattered most, they could see impressions, positions and click-through rates. Each shift was measurable, and therefore manageable.
AI recommendation arrived without one.
A traveller asks a model where to stay. It names four properties. If yours is not among them, no impression is logged, no position recorded, no report generated. A decision was influenced and nothing anywhere records that it happened.
The distinction that matters
In search, ranking eleventh means page two — disappointing, survivable, and visible in your analytics. In an AI answer there is no eleventh place. You are in the answer or you are absent. Share of Model exists to make that binary countable over time.
The definition
Share of Model is the percentage of a defined set of relevant queries in which a property is named, cited or recommended in AI-generated answers.
It is expressed per property and per engine, reported monthly, and always read against two references: your own baseline, and the same measurement applied to five named competitors.
How it is measured
The methodology is published because a measurement nobody can audit is not a measurement.
| Parameter | Standard | Why |
|---|---|---|
| Query set | Minimum 25 queries per property | Drawn from six categories: destination plus segment, amenity-led, comparison, occasion, near-me, and price-band. Built from how guests actually phrase questions, not from keyword tools. |
| Engines | Four minimum | ChatGPT, Google AI Mode, Perplexity and Gemini as a baseline. Vertical travel agents added where the segment justifies it. |
| Geography | Run from your source markets | Results diverge substantially by location. A query run from Mumbai about a London hotel does not reflect what a London traveller sees. |
| Competitive set | Five named competitors | You nominate them. Relative movement is more meaningful to a commercial team than an absolute figure. |
| Recording | Presence, position and source | Third of five is not equivalent to first of five. And the source cited tells us where to invest next month. |
| Evidence | Every query screenshotted and archived | This is what makes a twelve-month trend defensible rather than assertable. |
| Query set lock | Fixed for the engagement | Changing the query set invalidates the trend. A metric that can be reshaped is not a metric. |
| Cadence | Same week, every month | Consistency of method removes variation that would otherwise be mistaken for performance. |
What Share of Model does not tell you
Stated plainly, because a metric sold beyond its limits eventually damages the person who sold it.
- It does not prove bookings. AI referral traffic is largely untraceable through standard analytics. Share of Model measures whether you are in the consideration set — not whether a specific reservation resulted from it. We will not claim otherwise, and we would be sceptical of anyone who does.
- It is a sample, not a census. Twenty-five queries represent the shape of demand in a market; they are not every question ever asked.
- Generative answers contain variance. Any single reading includes noise. This is precisely why we fix the query set, run it monthly and report the trend rather than a one-off score. A single measurement is an anecdote; twelve are a trend.
- It does not replace your other metrics. It sits alongside occupancy, ADR, RevPAR and channel mix. It is one instrument, measuring one thing that previously had none.
Common questions
How is Share of Model actually calculated?
Count the queries in which the property appeared, divide by the total queries run, express as a percentage. Reported per property and per engine, with position recorded separately. The arithmetic is deliberately simple — the rigour is in the consistency of the query set and the method, not in the formula.
Why twenty-five queries and not more?
Twenty-five per property across four engines is 100 measurements per property per month — enough to establish a stable trend without becoming unaffordable across a forty-property group. For groups above twenty-five properties we measure a representative sample and disclose the sampling method in the report.
Can we audit your methodology?
Yes, and we would encourage it. The query set, the engines, the competitive set and the geography are all shared at the start of an engagement. You can re-run any query yourself and get the same answer.
What if a model just changes how it ranks sources?
It will, periodically and without notice. That is why we measure across four engines rather than one — no single change is fatal to the picture. It is also why the work is ongoing rather than a project.
Do you measure in languages other than English?
Yes, where source markets justify it. A Munich property is asked about in German; a Barcelona property in Spanish. Measuring only in English produces a misleading number for European groups.
NextSee the framework that turns the measurement into work.
The CITED framework