Test the way a guest asks, not the way you'd search
The instinct is to type your hotel's name into ChatGPT and see what comes back. That tells you what the engine knows about you once it already knows who you are — useful for checking accuracy, but it's not how guests discover you. Guests almost never name a property they haven't chosen yet. They describe the trip and let the engine nominate the options.
So test the discovery question, not the brand question. Ask the way a guest would: "best quiet resort in Phuket for a honeymoon," "family-friendly hotel in Bangkok near the river," "where to stay in Hong Kong for a first business trip." Then look at whether your property is one of the names that comes back — and if it isn't, note which properties won the slot instead. Those are your real AI competitors, which may not be the competitors you assume.
- Brand question ("tell me about Hotel X") tests accuracy — whether the engine describes you correctly once it knows you
- Discovery question ("best resort in… for…") tests visibility — whether the engine nominates you at all
- Discovery is where bookings are won or lost; run those questions first
- Note who wins the slot when you don't — that's your actual AI competitive set
Why a single check will mislead you
One prompt in one engine on one day is anecdote, not signal, for three reasons. First, engines disagree: a property can be named consistently by Gemini and absent from Perplexity, or strong in the Chinese engines and invisible in ChatGPT. Checking one engine tells you almost nothing about the others. Second, phrasing swings the result: "luxury resort" and "quiet honeymoon resort" and "5-star hotel" can surface different shortlists for the same destination, so a single wording under- or over-states where you stand. Third, results move over time as engines update, so a one-off check has no baseline to compare against.
This is why an offhand test is reassuring or alarming in equal measure and reliable in neither direction. What you actually need is coverage — the same set of guest questions, run across every engine that matters, repeated on a schedule so you can see the trend and prove whether a change worked.
From spot-check to real measurement
A structured audit is the difference between a hunch and a number. Salience runs a full set of the real questions your guests ask — across discovery, comparison, and decision stages — against eight engines (ChatGPT, Perplexity, Gemini, Claude, plus DeepSeek, Qwen, GLM, and Kimi). It records the actual answers, scores whether and how prominently you're named, and benchmarks you against the properties winning the slots you're missing.
Because it uses the same questions and the same competitor field every time, the audit gives you a baseline you can re-measure against after you make changes — so "we improved our AI visibility" becomes a specific before-and-after per engine and per question, not a claim. You can start with a free, no-login report on your own property to see exactly where you stand before deciding what to fix.
The DIY spot-check above is a fine first look and costs nothing but a few minutes. When you want the real picture — every engine, every guest question, a score, and a benchmark — that's the audit. Contact sales@pgintel.dev or request your property's report.