Hotel AI Visibility Methodology

The short version

Every day, we ask the major AI assistants the real questions travelers ask, record their answers word-for-word, and measure whether your hotel is named, where it ranks, how it's described, and which sources the AI used to decide. Every number on your RevPARGenius Hotel AI Visibility dashboard links back to the actual answer that produced it.

What changed — June 2026

Methodology updated June 2026: (1) per-cell sentiment denominator corrected — sentiment is now computed across mentioned cells only, not total cells in a run; (2) citation categorisation is now Claude-driven (was URL-pattern matching), which increases accuracy for ambiguous domains; (3) Recommended Actions now auto-complete on scan-confirmed resolution, with a Before/After snapshot preserved in the timeline.

How it works

A simple loop, run on a daily schedule.

We define the questions

A set of real traveler prompts for your market and guest type — "best boutique hotels in Hoi An," "where to stay in Queenstown for a couple," "family hotels near the old town."

We ask the AI engines

Each prompt is sent to the live, web-connected version of each assistant — so we capture what a traveler would actually see today, not the model's older training data.

We record everything

The full answer text and the exact source URLs each engine cited are stored, timestamped, and tagged with the model that produced them.

We measure

For every answer: were you named, at what position, in what tone, and against which competitors.

We track the trend

Results are stored every day, so you see movement over time — not a single snapshot that could be a fluke.

What the numbers mean

Four metrics, each defined the same way every time.

Visibility

The share of your tracked prompts where the AI names your hotel. Your headline number.

Position

When you're named, where you rank in the answer — first, third, eighth.

Sentiment

How the AI describes you when it mentions your hotel — scored 0–100 from negative to positive. Computed across mentioned cells only, not total cells in a run, so a single positive mention doesn't wash out in a sea of gaps.

Cited sources

The websites the AI read to build its answer — Booking.com, TripAdvisor, your own site. This shows what's shaping your reputation.

The engines we track

The assistants travelers actually use, queried through their official interfaces with live web search — each returning real citations.

ChatGPT Perplexity Gemini Grok

Google AI Overview, Google AI Mode, and Microsoft Copilot don't offer a direct interface, so we capture them from live search results and label them as such — measured a little differently, and shown transparently.

Why you can trust it

Built on evidence, not assertion.

—Every number has a receipt

Click any metric to read the actual AI answer and the citations behind it. Nothing is a black box; we report what the engine literally said.

—Citations come from the engines themselves

The AI tells us which sources it used. "Booking.com cited 31 times" is the engine's own data, not our interpretation.

—We run daily and report trends

AI answers naturally vary between runs, so a single check is noise. We measure rates across many runs over time, so the trend is real.

—"Mentioned" is a hard match

Whether your hotel was named is decided by exact text-matching against your name and its variants — provable, not a matter of judgment.

—Consistent, dated, reproducible

Every result is timestamped and tagged with the model that produced it, defined the same way each time — so the numbers are comparable week to week.

What we're honest about

Confidence comes from knowing the edges.

AI answers personalize. Results vary by individual user and location. RevPARGenius Hotel AI Visibility measures a consistent, defined baseline for your market — not every traveler's exact screen.

The engines change. Models are updated frequently, which can shift results independent of anything you do. We flag large movements rather than overstate them.

This is decision-support. AI visibility is directional market intelligence to guide strategy and content — a sharp instrument for where to act, not a guarantee of any single outcome.

How citations are categorised

Claude-driven classification, replacing URL-pattern matching as of June 2026.

Every URL cited by an AI engine is classified into one of six categories: OTA (Online Travel Agent — Booking.com, Expedia, Agoda, etc.), review platform (TripAdvisor, Google Reviews, Yelp), own site (the hotel's direct domain), trade press (hospitality media, industry publications), community (Reddit, travel forums), and other. Classification is performed by Claude, not by URL-pattern lookup, which means ambiguous domains — hotel groups that are also OTAs, regional meta-search engines — are correctly categorised based on page content and context.

The OTA share figure — the percentage of all citations coming from OTA platforms — is compared against the Cloudbeds AI Hotel Recommendations Study (2025) baseline of 55.3%. An OTA share below 55.3% indicates a healthier direct-booking pipeline than the industry average.

Cloudbeds 2025 baseline: 55.3% OTA citation share. This is the median OTA share measured across hotel-related AI answers in the Cloudbeds study. RevPARGenius shows your property's OTA share against this number directly in your citation breakdown. Source: Cloudbeds AI Hotel Recommendations Study, 2025

How recommendations are generated and auto-completed

Automatically from scan data. Auto-resolved when your fix lands.

Recommended Actions are generated automatically at the end of each scan by Claude, which analyses the full result set: which prompts the property was invisible on, which competitor patterns dominated those prompts, and which technical signals — schema coverage, citation source mix, position data — were weakest. Actions are ranked by estimated mention-rate impact within 30 days, assigned a priority (Critical, High, Medium, Low), a category (Content, Schema, Technical, Profile), and a rough effort estimate.

Actions persist across runs — they are not regenerated each time. When a subsequent scan shows the property is now mentioned on a previously-gap prompt, the Recommended Action for that prompt is auto-completed. A Before/After snapshot (the prompt, the previous gap response, and the new cited response) is stored in the timeline so you can verify the improvement.

This design means the action list is durable: if you mark an action as In Progress, that status persists. If you complete the fix and the next scan confirms it, the action closes itself. See the sample report for an example of four actions generated from a real Brisbane scan, and the Hotel AI Visibility overview for details on how to act on them.

Frequently asked questions

How does RevPARGenius measure AI visibility?

Every day we send real traveler prompts to the live, web-connected versions of ChatGPT, Perplexity, Gemini and Grok, record their answers word-for-word with the sources they cited, and measure whether your hotel is named, at what position, with what sentiment, and against which competitors. Every number links back to the actual AI answer that produced it.

Which AI models does RevPARGenius Hotel AI Visibility track?

ChatGPT, Perplexity, Gemini and Grok — queried through their live, web-connected interfaces so we capture what a traveler would actually see today. Google AI Overview, Google AI Mode and Microsoft Copilot are captured from live search results and labelled separately.

How accurate and trustworthy are the results?

Every metric links back to the actual AI answer and the citations behind it — nothing is a black box. Whether your hotel was named is decided by exact text-matching against your name and its variants, not judgment. Results are timestamped and tagged with the model that produced them, defined the same way each time, so the numbers are comparable week to week.

How often do you run the checks?

Three times per week — Monday, Wednesday and Friday. Each run captures the live response and citations, timestamped, so you can track changes over time rather than rely on a single snapshot that could be a fluke.

Can AI results be gamed or manipulated?

Not sustainably. AI assistants draw from indexed web content, review platforms and citation signals — the same sources that determine long-term search authority. Short-term tactics that don't reflect real content quality or consistent business signals don't hold. RevPARGenius tracks the rate over many runs so one-off fluctuations don't distort your score.

← RevPARGenius Hotel AI Visibility overview · See pricing →

How we measure AI visibility.