How the score works
If a venue disputes its score, this page is the answer. Everything below is published because everything below is checkable.
1. What we measure
For each reading we ask three AI assistants (ChatGPT, Gemini and Claude) a fixed set of 50 questions a real traveler, wedding planner or event organizer would ask about the market. Every question is asked three times per engine, in fresh sessions through each provider's API with live web search turned on. The model versions are listed with every reading. The API answers can differ from the consumer apps, which add their own settings and personalization. That is 450 answers per reading: 50 queries × 3 engines × 3 runs.
Each live market has had its first reading, and later readings repeat on a schedule that this page will announce. A market marked first reading pending has its venue list and questions published, and it gets scores once its first reading runs.
For every answer we record the model and version, the date and time, the full text, every venue named, the order venues appear in, any links or citations shown, and the language used about each venue.
2. What counts as a mention
- A venue is "mentioned" when its name, or an unambiguous reference to it, appears in the answer text. Aliases and common misspellings are mapped in a published alias table.
- A mention of the town a venue is in does not count.
- A venue appearing only as a link still counts, and is also scored as a citation.
- Negative mentions count for visibility but not for prominence. "Avoid X" makes X visible without recommending it.
3. The four components
Presence: 40 points. The share of the 450 answers in a reading in which the venue appears at all. It is a plain count that anyone can verify against the archived transcripts.
Prominence: 25 points. How the venue appears when it does. Each mention is graded:
| Grade | Condition | Value |
|---|---|---|
| Lead | First venue named, or the single recommendation | 1.00 |
| Featured | Named with descriptive language: why it's good, who it's for | 0.75 |
| Listed | Named in a list with no description | 0.40 |
Citation: 20 points. The share of mentions with a link to the venue's own website. Links to OTAs, review sites or listicles do not count, because a booking through a middleman pays commission. Citation is measured only on engines that showed links in that reading, and each reading lists them. In the first reading, Claude's answers carried no links, so citation there comes from ChatGPT and Gemini.
Stability: 15 points. Whether the recommendation survives repetition. For each of the 150 query-engine pairs, we ask three times. Stability is the average, across pairs where the venue appears at least once, of runs-with-mention divided by 3. A venue that flickers in and out scores low; a lucky single answer is not a reputation.
4. Limits
- AI answers are noisy. The same question can get different answers minutes apart. That is why every query runs three times per engine and why Stability exists. Even so, a score describes one reading only.
- The index measures what models say about a venue. A superb venue with a thin web presence will score low. That gap is the point of the index, and it is fixable.
- Models change. Providers update models without notice. We log version strings and note known model changes on the affected reading.
- The query set is editorial. We chose the 50 questions. The full set is published so anyone can see the frame. Set changes are versioned and announced.
- Coverage is a seeded universe. Each table covers venues in its market drawn from public listings. Any venue in the market can request inclusion; it is free and takes effect at the next reading.
5. Commercial independence
Scores cannot be bought, and no commercial relationship with Vael Creative affects any score. Clients and non-clients are scored by the same runs from the same transcripts.
Properties marked † on the table and on their score pages are Vael Creative clients. Scores are computed from the published query set before any client work begins, and no client input affects scoring. A script applies the mark from the venue seed file, so nobody can add or remove it for a single reading.
6. Disputes
Any venue can request a re-score. We re-run the full battery for the disputed reading and publish the result, whatever it shows. Raw transcripts behind every published score are available on request: brian@vaelcreative.com.