Start with what you're actually trying to find out
The question isn't "does ChatGPT know my hotel exists." Most AI models have crawled the open web and can describe your hotel if you ask about it by name. The real question is narrower and more useful: when a traveler asks a generic question, one that doesn't mention your hotel, does the AI bring your property up on its own? That's the behavior that actually decides whether a guest hears about you before they've heard of anyone else.
The core metric: detection rate
Detection rate is the share of tracked questions where your hotel gets named, out of the total number of checks run. If you ask five AI engines twelve real guest questions each (sixty checks total) and your hotel comes up in nine of them, that's a 15% detection rate. It's a plain, arithmetic number, and it's the one figure that tells you, at a glance, whether things are getting better or worse over time.
There's no single universal benchmark for what counts as "good": across a small set of hotels we've audited directly, ranging from a boutique property in Prague to a spa hotel in a smaller destination, detection rates have landed anywhere from a flat 0% to a little over 12%. What matters more than hitting some target number is knowing your own starting point and whether your own number is climbing.
Break it down by engine
A blended detection rate hides more than it shows. ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews each source and generate answers differently, so it's normal for a hotel to be strong on one and invisible on another. In case studies we've published, several hotels scored respectably on Gemini and Perplexity while getting a flat zero on Google AI Overviews specifically, a pattern that a single blended number would have completely papered over.
Mention rank matters too, not just whether you're named
An AI answer that names three hotels and puts you last still sends most of the booking interest to whoever's named first. Position within the answer (first mentioned, buried in a longer list, or the sole recommendation) is a second dimension worth tracking alongside the raw yes/no of detection. Two hotels can have the same detection rate and very different real-world outcomes if one is consistently named first and the other consistently named last.
The manual self-check method
You don't need a paid tool to get a rough read on where you stand. It takes about three minutes and gives you a real, if noisy, data point.
Why a single check isn't the same as measurement
The manual method above has a real limit: large language models are stochastic. Ask ChatGPT the same question twice on the same day and it can hand back two different lists of hotels. A single run tells you what happened once, not what usually happens. Treating one lucky mention as "we're visible now" is the same mistake as treating one bad run as "we're invisible," both are true only on average across many checks, which is why a repeated panel of questions, run on a schedule, gives a far steadier read than any single spot-check.
Turning a number into something you can act on
A detection rate on its own tells you there's a gap, not why it exists. The useful next step is looking at which specific questions you're missing (broad, generic ones versus narrow, occasion-specific ones tend to behave very differently) and cross-referencing that against the technical side: is your site readable by AI crawlers, does it carry structured data describing your hotel, is your name, address, and phone number consistent everywhere you're listed. Those are the levers that tend to move the number, not the number itself.
How often to check
Once is a snapshot. A monthly or weekly cadence, on the same set of questions, is what turns a single number into a trend line you can actually manage against, and what tells you whether a fix you made last month actually did anything.