What ChatGPT Says About Your Hotel: A Field Test
We asked ChatGPT, Claude, Perplexity, and Gemini about the same boutique hotel. The four answers diverged in instructive ways, and the fixes are concrete
There is a small experiment every hotelier should run this week. Open four browser tabs. ChatGPT, Claude, Perplexity, Gemini. Type the same question into each: "Tell me about [your hotel]. What is it known for, what kind of guest is it good for, and what is the best room there?" Read the four answers side by side.
We have run this test on dozens of independent hotels in the last few months. The results are consistently strange, often wrong in interesting ways, and almost always a useful diagnostic for what the AI ecosystem actually knows about a property. To make this concrete without naming a real hotel, let us walk through a fictional case: The Avery Charleston, a 28-room boutique on Meeting Street.
The four answers
ChatGPT opened with a confident summary. The Avery is described as a "design-forward boutique hotel in the historic French Quarter of Charleston, known for its rooftop bar and award-winning Southern restaurant." This is roughly half right. The Avery is indeed on Meeting Street, but it is not in the French Quarter, it does not have a rooftop bar, and the restaurant won an award in 2019 that ChatGPT presents as current. The best room, according to ChatGPT, is "the Magnolia Suite with views of the harbor." The Avery has no Magnolia Suite. The harbor is not visible from the property.
Claude was more cautious. It correctly identified the property as a small boutique hotel in downtown Charleston, declined to describe specific room types it could not verify, and noted that for current pricing and availability, the guest should check the hotel's website. It mentioned the restaurant by name correctly. It did not invent amenities. The answer was thinner but more honest.
Perplexity produced the most up-to-date answer because it ran a live web search. It pulled the current website description, two recent TripAdvisor reviews, and a 2025 Conde Nast Traveler mention. The room types were accurate. The price range was within $30 of the actual ADR. It also surfaced a one-star review that the GM had not seen, in which a guest complained about a noisy ice machine on the third floor.
Gemini confidently confused The Avery Charleston with The Avery Tucson, a different small hotel with a similar name. The entire answer was about the wrong property. This is a failure mode that happens more often than you would think, particularly for hotels with common single-word names.
Why the answers diverge
Four assistants, four different stories. The reasons are worth understanding because they tell you what to fix.
ChatGPT relied heavily on its training data, which is months or years old depending on the topic, and confidently filled in gaps with plausible-sounding details. Hallucinated room names, invented bars, outdated awards. This is not a bug. It is the predictable behavior of a model that has been trained to give helpful answers even when its information is thin.
Claude leaned conservative. When it could not verify a fact, it declined to make one up. That makes for a less impressive answer but a more accurate one.
Perplexity is doing live retrieval, which means its answer reflects whatever the web says about you right now. That is good when your website is sharp and your reviews are recent. It is bad when there is a stale Tripadvisor post or a months-old news article that no longer reflects the property.
Gemini's confusion with a similarly-named hotel is what happens when an AI cannot disambiguate between entities. If your property shares a name with another hotel, a restaurant, a neighborhood, or a person, you are at high risk for this failure mode.
The specific failure modes
If you run this test on your own property, you will likely see some combination of the following:
- Stale prices. An assistant cites a rate from 2022.
- Wrong room counts. "A 64-room boutique" when you have 38 rooms, or vice versa.
- Made-up amenities. A spa you do not have, a rooftop you closed, a restaurant under different management.
- Confused identity. Conflation with a similarly named property in another city.
- Outdated leadership or ownership. The general manager from three years ago is still listed as the face of the property.
- Hallucinated awards. A "James Beard nominated chef" who was nothing of the sort.
- Misattributed reviews. Praise or criticism from another hotel showing up in your summary.
Each of these is a small problem on its own. In aggregate, they shape the answer your next guest gets when they ask an AI assistant whether they should book with you.
The painful part is that you cannot control what the AI says about you by writing better marketing copy. You can only control it by giving the AI better source data.
How to fix what AI says about your hotel
The good news is that the fixes are concrete and most of them are free or cheap. Here is the order we would recommend.
1. Fix your Google Business Profile first
This is the single most-read source for AI assistants doing live retrieval. Your hours, address, photos, room types, and amenities should be current and accurate. If your last photo upload was 2022, that is what an AI will describe. If your category is "lodging" and not "hotel" or "boutique hotel," you will get filtered out of many recommendation queries.
2. Claim your hotel on Roomza
Roomza is the structured-data layer that exposes your property to AI assistants in a machine-readable format. Claiming your hotel takes about ten minutes. Once claimed, you control the room-level data, the amenity list, the descriptions, and the brief.
3. Write a concierge brief
This is the field on your Roomza profile that explicitly tells AI assistants what to say about your hotel. Two paragraphs, plain English, written in the voice you would use to brief a new concierge on their first day. Who this hotel is for. What makes it different. What to recommend it for. What to warn guests about.
A good concierge brief reads like a knowledgeable friend describing the property. A bad one reads like a brochure. Aim for the friend.
4. Expose structured data via MCP
Roomza ships an MCP server for every hotel on the platform. This means Claude, ChatGPT (via plugins and the upcoming MCP-compatible features), and any other compatible AI assistant can query your hotel directly for real, structured, current information. Your room types, your amenities, your availability windows, your rate ranges.
This is what separates a hotel that gets recommended for "a quiet boutique in Charleston with a clawfoot tub" from a hotel that does not. The assistant can ask a specific question and get a specific answer.
5. Watch for hallucinated facts in reviews
Periodically run the four-tab test described at the top of this post. Look specifically for facts the AI invents. If ChatGPT keeps claiming you have a rooftop bar you do not have, that hallucination is being seeded by something, often an old article, a confused review, or a similar-sounding competitor. Find the source and address it. Sometimes a single update to your website description or a polite correction request to a travel publication is enough to retrain the next generation of answers.
You do not need an AI SEO agency
There is a small cottage industry forming around "AI search optimization" for hotels. Some of it is useful. Much of it is the same playbook that traditional SEO agencies have been selling for years, repackaged with new acronyms.
The honest answer is simpler. AI assistants give better answers when they have better source data. The job of a hotel in 2026 is to make sure the source data is correct, current, and structured. Roomza exists to be that layer, the place where your data is canonical, machine-readable, and shaped by you instead of inferred by a model.
Run the test. See what the four assistants say about your hotel. Then fix what they got wrong. The next guest who asks will get a better answer, and you will have spent an afternoon doing what most of your competitors are still ignoring.
