There is a conversation happening about your hotel right now, and you are not in the room.
A traveler opens ChatGPT and types “best boutique hotels in your city for an anniversary.” A few seconds later they have a shortlist, a vibe summary, and a confident little paragraph about each property. Some of those properties are your competitors. Maybe one of them is you. Maybe the paragraph about you is wrong. You have no idea, because you have never actually looked.
That is the whole problem. Most independent hoteliers obsess over their Booking.com ranking and their TripAdvisor stars, both of which they can see, and completely ignore the AI answers they cannot see. Which is a little like checking your reflection in one mirror while a different mirror behind you shows something else entirely to everyone walking past.
The fix is not complicated, and you do not need a tool subscription or an agency to do the first pass. You need a repeatable audit you run once a month. This post is that audit, step by step, with a scoring table you can copy. If you have not already read Is Your Hotel Invisible to ChatGPT?, start there for the why. This post is the how.
Why this matters more every quarter
Quick reality check on scale. Search volumes for the discipline of optimizing for AI answers have exploded: “AEO” (answer engine optimization) pulls roughly 27,100 US searches a month, “AI SEO” around 8,100, and “generative engine optimization” about 5,400. Those are marketers searching, sure, but marketers search because their customers are behaving differently. Travelers are increasingly asking an assistant before they ever open a booking site.
And here is the uncomfortable part: when a traveler trusts the AI’s shortlist, they often skip the comparison-shopping rabbit hole entirely. That can actually be good news for your margins. Every booking that starts with an AI recommendation and ends on your own booking page is a booking that did not hand 15 to 25 percent to an OTA. You will never fully escape the OTAs, and nobody serious is promising you that. But showing up accurately in AI answers is one of the cleaner ways to win back more direct bookings and depend a little less on the channels that tax every reservation.
You cannot influence what you have not measured. So let’s measure.
The fixed prompt set (this is the heart of it)
The single biggest mistake people make is “auditing” by typing one random question into ChatGPT, getting a meh answer, and concluding the whole thing is noise. The trick is a fixed prompt set you run identically every month. Same prompts, same order, every time. Consistency is what turns vibes into a trend line.
Here is the starter set. Adapt the bracketed bits to your property, but keep the structure.
Category 1 - Discovery (does the AI surface you at all?)
These are the prompts where the traveler does not know your name yet. This is where you are competing for a spot on the list.
- “What are the best boutique hotels in [your city]?”
- “Where should I stay in [your city/neighborhood] for a romantic weekend?”
- “Recommend a [pet-friendly / family-friendly / adults-only - pick your niche] hotel in [your city].”
- “I want a small independent hotel in [your city], not a big chain. Any suggestions?”
- “Best hotels near [your nearest landmark or demand driver].”
Category 2 - Branded (does the AI know YOU, and get you right?)
These are the prompts where the traveler already has your name. This is where misinformation hides.
- “Tell me about [Your Hotel Name] in [your city].”
- “Is [Your Hotel Name] a good place to stay? What are the pros and cons?”
- “Does [Your Hotel Name] have [your key amenity - pool, parking, spa, restaurant]?”
- “How much does it cost to stay at [Your Hotel Name]?”
- “What do guests say about [Your Hotel Name]?”
Category 3 - Head-to-head (how do you stack up?)
- “[Your Hotel Name] vs [Direct Competitor Name] - which is better for a couple?”
- “Cheaper alternatives to [a big upscale property in your market].”
Run all twelve. Yes, it takes a bit. No, you do not skip the boring ones, because the boring ones are where you find out the AI thinks you closed in 2019 or that you do not allow dogs when your whole brand is dogs-welcome.
A prompt set is only useful if it never changes. The moment you start “improving” your prompts every month, you lose your ability to compare month over month. Lock the wording. Improve the hotel, not the questions.
Run it across the big three (they disagree, and that’s the point)
Do not audit ChatGPT alone. Run the same twelve prompts through at least three assistants, because they pull from different places and they genuinely disagree:
- ChatGPT - huge reach, blends training data with live web results. Often the default for casual travelers.
- Gemini - tied into Google’s ecosystem, so it leans on Google Business Profile, Maps, and Google’s index. If your local presence is strong, this is often where you show up first.
- Perplexity - the citation machine. It shows its sources inline, which makes it the most useful for diagnosis. If Perplexity cites a competitor’s blog post and a stale directory listing but not your site, you have just found your homework.
Pro move: in Perplexity especially, screenshot the citations. The sources it names are a literal to-do list of which pages are shaping the answer. If you want the deeper playbook on getting cited, our piece on the GEO trifecta of schema, entity, and citations goes into the mechanics.
One housekeeping note: use a logged-out or incognito session where you can, and turn off chat memory/personalization if the tool allows it. You want the answer a stranger gets, not the answer the assistant has tailored to you because you have been Googling your own hotel for six months.
How to log it: the scoring table
Open a Google Sheet. One row per prompt, repeated for each AI tool, with a fresh tab each month. For every prompt you are scoring four things:
- Mentioned? - Did your hotel appear at all? (Yes / No)
- Position - If it’s a list, where did you land? (1st, 2nd, “buried in a list of 10,” not present)
- Accuracy - Is what it said actually true? (Accurate / Minor errors / Wrong)
- Sentiment - Does it make you sound appealing, neutral, or meh?
Then roll those into a simple 0-3 score per prompt so you can total it and watch the number move. Here is the rubric:
| Score | What it means | Example |
|---|---|---|
| 0 | Invisible | Not mentioned at all, or a competitor is recommended in your exact niche |
| 1 | Present but flawed | Mentioned, but with wrong facts (closed, wrong amenities, wrong location) or a flat, unappealing description |
| 2 | Accurate but unremarkable | Correctly mentioned with right facts, but generic - nothing that makes a traveler pick you |
| 3 | Strong | Recommended prominently, accurate, and described in a way that actually sells the stay |
Tally your twelve prompts across three tools and you have a single AI Visibility number out of 108 (12 prompts x 3 tools x 3 max). It is a made-up index, not a law of physics, but it is your index, measured the same way every month, and that is exactly what makes it useful. A move from 41 to 53 is real progress you can point at.
Field tip: keep a separate “Misinformation Log” column that you never delete. Every wrong fact the AI states - wrong pet policy, a restaurant you closed, a pool you never had - gets logged with the date you first saw it and the date you fixed the source. This becomes your evidence trail when an answer finally corrects itself, and it tells you which fixes actually moved the needle.
Reading the results: what each pattern is telling you
A filled-in scoreboard is data. Here is how to turn it into action.
You score mostly 0s on discovery prompts. The models do not associate your hotel with the categories you compete in. This is an entity and authority problem - the AI does not have enough corroborated signals that you are “a boutique hotel in [city]” or “a pet-friendly stay.” The fix lives in your structured data and your third-party mentions. Start with structured data that makes your hotel quotable to AI.
You score 1s on branded prompts - mentioned, but wrong. This is misinformation, and it is the most urgent to fix because it actively costs you bookings. A traveler who reads “no parking” when you have a lot, or “permanently closed” when you are very much open, simply moves on. Almost always the AI is echoing a stale source: an old directory listing, a years-old article, an unupdated Google Business Profile. Your job is to hunt down and correct the source, not to argue with the chatbot.
You score 2s across the board - accurate but boring. Good news, you have no misinformation. Bad news, you are forgettable. The AI is describing you in beige. This usually means your own website talks about your hotel in features (“32 rooms, complimentary WiFi”) instead of in the language travelers and AIs both reward - who it’s for, what the experience feels like, what makes it specific. This is a content and positioning fix.
A competitor keeps showing up where you don’t. Go find out why. Run their name through Perplexity and read the citations. Nine times out of ten they have a clearer website, better-structured pages, or stronger coverage on the directories and local sites the models trust. It’s reverse-engineerable.
A realistic example of the loop in action
Imagine a 40-room inn in a small coastal town. They run the audit in month one and score a grim 38 out of 108. The pattern is brutal but clear: they are invisible on every “romantic getaway” discovery prompt, and on branded prompts ChatGPT insists they don’t allow dogs - even though dog-friendly is half their brand.
So they do three things. They rewrite their site’s about and rooms pages to lead with “adults-only romantic coastal escape” and “dogs stay free.” They add proper hotel structured data. And they fix the three old directory listings still showing “no pets.”
Month two, nothing moves much - which is normal, the models need time to re-read the web. Month three, the dog misinformation is gone from ChatGPT and Gemini, and they’ve started appearing in a couple of romantic-weekend lists. Month five, they’re at 61. That is the loop: measure, fix the source, wait, measure again. None of it is glamorous, and all of it compounds.
That whole “do travelers even trust this stuff” question is worth understanding too - our breakdown of how travelers actually use AI to pick hotels covers the behavior side.
Your repeatable monthly checklist
Print this, or paste it into your sheet as the first tab. The entire thing should take about 45 minutes once you’ve done it twice.
- Open a clean session in ChatGPT, Gemini, and Perplexity (logged out / incognito, personalization off where possible).
- Run the same twelve prompts in the same order in each tool. Do not improvise.
- Screenshot the answers, especially Perplexity’s citations.
- Score each prompt 0-3 using the rubric. Log mentioned / position / accuracy / sentiment.
- Update the Misinformation Log with any new wrong facts and the date.
- Total your AI Visibility index and drop it in your trend tab.
- Pick one to three fixes for the month - source corrections first, then positioning, then structured data.
- Diarize next month. Same date, same prompts. That’s the whole magic.
Where this fits in the bigger picture
This audit is the diagnostic, not the cure. It tells you exactly where you stand and what’s broken, which is genuinely most of the battle - you cannot fix invisible problems. The actual fixing happens across your website content, your structured data, your reviews, and your presence on the third-party sources the models trust. If you want the conceptual map of how answer engines, generative engines, and classic search relate, AEO vs GEO vs SEO for hotels lays it out, and getting your hotel cited in Google’s AI Overviews handles the Google side specifically.
And to be very clear about the prize here: none of this lets you fire the OTAs or break away from them entirely. No independent hotel can, and anyone selling you that is selling you nonsense. What it does do is help you show up accurately and appealingly at the exact moment a traveler is deciding - which nudges more of them toward booking direct, claws back margin the OTAs would otherwise take, and gives you a healthier channel mix you actually control a little more of.
Do the audit this month. Even if you just run the twelve prompts in ChatGPT alone and eyeball the results, you will learn something uncomfortable and useful in the next twenty minutes.
Want us to run the full audit for you? We do a month-one AI visibility baseline across ChatGPT, Gemini, and Perplexity, hand you the scored sheet, and tell you exactly which sources to fix first. Book a free intro call or see how our AI Visibility (AEO/GEO) service works. Bring your competitors’ names - it’s more fun when we get to compare.