There are two ways to find out what AI says about your hotel. You can open ChatGPT, type a question, and read the answer — a snapshot, one engine, one moment, gone the second the model rolls the dice again. Or you can do that systematically: thousands of questions, every major assistant, every day, scored and tracked over time. The second is measurement. This is how it works.

Step one: ask the questions travelers actually ask

Travelers don't search "Hotel Bellevue Lisbon." They ask "best boutique hotel in Lisbon for a romantic weekend near the water." So that's what we ask. The questions — we call them prompts — come in three kinds:

  • Discovery questions name no hotel at all: "best family resort near Lake Tahoe," "where to stay in Austin for SXSW under $300." These are the ones that decide whether you get discovered, and they're the only kind that feeds your visibility score — naming yourself doesn't count.
  • Property questions ask directly about you: "tell me about Hotel Bellevue." These reveal whether the AI knows you accurately and which sources it cites.
  • Booking questions ask how to reserve: "how do I book Hotel Bellevue?" These feed the booking-path read — does the AI point to your own site or an OTA?

Discovery questions are anchored to your reality — your neighborhood, your price band, the landmarks and airports near you, the things you're genuinely known for — because that's how travelers phrase them. When a detail is missing the question gracefully widens (from "near Alfama under €250" to "in Lisbon") so the measurement stays stable rather than breaking.

Step two: ask as different travelers

Here's the part most spot-checks miss. The same question returns different hotels depending on who's asking. So we re-ask each prompt as a set of personas — couples, families, business travelers, budget-conscious guests, the segments you actually compete for. The AI sees a traveler profile and tailors its answer, exactly as it would for a real person.

This is how you discover that AI recommends you confidently to couples but overlooks you for families — a gap you'd never see from a single test, and the most actionable thing in the whole report. Lite includes 3 personas, Pro includes 5, and you can add more as you need them.

Step three: ask every engine that matters

Travelers don't all use the same assistant, so we don't track just one. Sigtrip queries ChatGPT, Perplexity, Google AI (Gemini), Anthropic's Claude, and DeepSeek. Each engine carries a weight reflecting its reach — ChatGPT counts most — so your headline score reflects where travelers actually are. Free and Lite plans track ChatGPT; Pro adds Perplexity and Google AI at no extra cost; any engine can be added on a paid plan.

When an engine is switched off, it doesn't drag your score down — the math rebalances across the engines that are running, so adding or removing one re-weights rather than penalizes.

The fan-out is bigger than it looks

Prompts × personas × engines multiplies fast. Even a handful of prompts, re-asked for every persona across every engine, every day, generates thousands of AI answers a month. That volume is the whole point: it's what turns a noisy, one-roll-of-the-dice snapshot into a stable, trackable signal.

Step four: read the answer — twice

An AI answer is free-flowing prose, not a tidy list. Catching whether you appeared — and where — takes two passes.

Pass one is fast and literal: it scans the text for your name and its variants. Deterministic, but brittle — it can miss "the Tivoli" when your full name is "Tivoli Avenida Liberdade," or trip over accents and abbreviations.

Pass two is an LLM that actually reads the answer the way a person would. It catches the paraphrases pass one misses, judges your rank among the properties named, and — critically — extracts the rest of the picture: which competitors were recommended alongside or instead of you, the tone of the mention, the sources the AI leaned on, and whether the booking path points to you or to an OTA. Pass two is the authoritative read; pass one is the fast first filter and a cross-check.

Step five: turn answers into scores

From those thousands of reads, a handful of numbers fall out. Each answers a different question:

  • Visibility Index (0–100) — your headline score. Rank- and engine-weighted, so being named first counts more than being named last, and a mention on a high-reach engine counts more than one on a niche engine. Built from discovery questions only.
  • Mention Rate (0–100) — how often you appear at all, rank-blind. "Do they ever bring us up?"
  • Blind Spot (0–100) — the mirror image: where you're missing. Broken down by theme and engine so it reads as a to-do list, not just a number — "invisible on family queries in Google AI."
  • Share of Voice — your mentions versus your competitors' on the same questions. True competitive share, counted the same way for everyone.
  • Booking Path Integrity (0–100) — when the AI explains how to book, does it send the guest to your own site (100) or only to OTAs (0)? Being recommended but handed to Booking.com is a half-win, and this score makes that visible.

Add per-persona and per-engine breakdowns underneath each, and "are we visible?" becomes a map precise enough to act on.

What measurement can't do — and why we say so

We're measuring a living, stochastic system. Ask an AI the same question twice and you can get two different answers — that's the model, not a flaw in the method. So day-to-day numbers wobble, and any honest methodology has to say so out loud rather than smooth it into a comforting straight line. We estimate the size of that normal wobble for your specific hotel and only flag moves that clear it. That's a whole topic on its own — why your score moves, and how to read the real signal.

The discipline is the same one that's always governed measurement: be precise about what you're counting, honest about the noise, and consistent enough day to day that the trend means something. The reward is that "how do we show up in AI?" stops being a matter of opinion you argue about in a meeting, and becomes a number you can move.

Want to see it on your own property? Run a free scan and read the first set of answers for yourself.