Your AI Choice Score is a number out of 100 that summarises how strongly AI answers favour your hotel in our benchmark. A higher score means stronger results across the five parts below.
Think of it as a report card: each part gets a mark out of 100, but some parts count more towards the final mark. A score of 71 does not mean a 71% chance of a booking or recommendation.
We test your hotel and your confirmed competitors using the same set of questions and selected AI models. A “prompt” simply means a question we ask the AI. We save the answers and record which hotels appear, their positions and any direct comparisons.
Only questions marked for measuring hotel choice feed the recommendation parts of the score. Questions used only to check facts or flag risks are excluded. Below, “scored answers” means the answers included in that choice calculation.
The five parts
The weights add up to 100%. For example, strong recommendations make up 35% of the score, so a mark of 80 contributes 28 points: 80 × 0.35 = 28.
Strong recommendations
Count the scored answers that place your hotel first, second or third. Divide by the total number of scored answers and multiply by 100. An answer where your hotel is absent or has no recorded position does not count as a top-three result.
Top three in 8 of 10 answers = 80/100.
Head-to-head wins
For recorded direct comparisons, a win earns 1 point, a tie earns ½ a point and a loss earns 0. Divide the points by the number of usable comparisons and multiply by 100. One answer can contain several comparisons.
5 wins + 2 ties + 3 losses = 6 points from 10 comparisons = 60/100.
High-intent prompts
These are questions about choosing between hotels, comparing features, finding a stay for a local need or resolving booking concerns. For answers where your hotel appears, turn its position into points using the scale below, then take the average.
Positions 1, 2 and 3 earn 90, 68 and 46 points. Their average is 68/100.
AI placement
Find your hotel’s average recorded position across scored answers, then turn that average into points using the same position scale. Answers without a recorded position are left out of this average.
An average position of 2 earns 68/100.
Proof strength
Combine the quality of the saved evidence with how much of the planned benchmark produced evidence. Deduct points for items flagged for review. This is a check on the evidence behind the run, rather than another count of recommendations.
High-quality evidence, 80% coverage and no review flags earn 87/100. The full calculation is below.
A worked example
Imagine a fictional hotel with the results shown above: top three in 8 of 10 scored answers; 5 wins, 2 ties and 3 losses; an average of 68 for choosing-style questions; an overall average position of 2; and proof strength of 87. All five parts are available.
| Part | Mark /100 | Weight | Points added |
|---|---|---|---|
| Strong recommendations | 80 | 35% | 28.00 |
| Head-to-head wins | 60 | 30% | 18.00 |
| High-intent prompts | 68 | 20% | 13.60 |
| AI placement | 68 | 10% | 6.80 |
| Proof strength | 87 | 5% | 4.35 |
| Total before rounding | 70.75 | ||
71/100
28 + 18 + 13.60 + 6.80 + 4.35 = 70.75. We round to the nearest whole number, giving an AI Choice Score of 71.
To see what a change means: if this hotel made the top three in 9 of 10 answers, its strong-recommendation mark would rise from 80 to 90. If the other four marks stayed the same, that would add 3.5 points and take the final score to 74.
Open a question for the exact rules, including how we handle gaps.
We use the recorded list positions instead. Your hotel earns a win if it is the only hotel from the comparison group to appear, or if its position is at least as high as the best-positioned alternative. A tied list position counts as a full win under this fallback rule; an explicitly recorded tie counts as half a win.
A positioned hotel also earns a win against alternatives that appear without positions. It earns no win when an alternative ranks higher, or when it is absent and an alternative appears. If your hotel appears without a position but another hotel has one, the comparison counts with no win. If there is no usable comparison, we leave that answer out of the win rate. Explicit comparisons marked uninformative are also excluded.
That answer still counts in the total used for strong recommendations, so missing the top three lowers that mark. For head-to-head results, an appearance by an alternative can count as a loss under the rules above.
High-intent and placement marks use appearances only: an absence is left out of those averages. In a high-intent answer, an appearance without a recorded position earns 42 points. For overall AI placement, an appearance without a position is excluded. If there are no qualifying appearances for a part, that part is unavailable.
Start with an evidence-quality value: high = 92, mixed = 64, low = 32, or unknown = 45. High quality means there are saved evidence items and none carry the low-confidence, weak-support, review, indirect-claim, weak-source or uncertain-hotel flags. Quality is low if there is no evidence, or at least half the items are low-confidence, weakly supported or flagged for review. Other cases are mixed.
Coverage is the percentage of all planned questions in that hotel’s run that produced saved evidence, including questions outside the choice-score subset. Multiply the quality value by 0.55 and the coverage percentage by 0.45. Add them, then subtract 4 points per item needing review, up to a maximum deduction of 24. Round the result and keep it between 0 and 100.
In our example: (92 × 0.55) + (80 × 0.45) − 0 = 86.6, rounded to 87. That contributes 87 × 0.05 = 4.35 points to the final score.
An unavailable part is left out, rather than treated as zero. We divide the points from the remaining parts by their combined weight. If only proof strength were unavailable in our example, the remaining points would total 66.4 and their weights would total 95%. The calculation would be 66.4 ÷ 0.95 = 69.89, rounded to 70.
We only publish a score when there are at least 3 eligible answers, your hotel appears in at least one, and at least 3 parts are available with a combined weight of 65% or more. Otherwise, we show “Insufficient evidence”. A measured zero is still included; it is different from an unavailable part.
Each of the five marks is rounded to a whole number before we apply its weight. For placement, we use the unrounded average position to calculate the mark. Evidence coverage is saved to three decimal places as a fraction, then converted to a whole-number percentage for proof strength. We add the weighted marks, adjust for any unavailable parts and round the final score to the nearest whole number. Half-points round up.
The score describes the answers captured for that benchmark. Compare scores over time using the same hotels, questions, AI models and scoring rules. A change in those conditions can change the score too.
The benchmark runs without live web-search tools. It reflects each model’s available knowledge and the information supplied in the test. AI answers are probabilistic: asking the same question again can produce a different answer.
These scores do not prove bookings or revenue impact. Use the five parts to understand where your hotel is being recommended and where its results are weaker.