A guest asks the website chatbot: "Is breakfast included?" The assistant answers confidently: "Yes, the breakfast buffet is always included." Shame that at your hotel breakfast is included only in some rates. On arrival, the guest disputes it. And they're right: your hotel told them so.
Generative AI systems sometimes produce plausible but wrong information (so-called "hallucinations"). At a hotel, where prices, services and conditions are promises made to guests, it's a serious risk. Let's look at how to stop an AI assistant inventing information about your hotel.
Why AI invents things
- missing information: if it doesn't know the answer, it may "fill in" with what's typical for a hotel;
- contradictory information: if sources say different things, it picks at random;
- ambiguous questions: it interprets and answers confidently even when it shouldn't;
- weak instructions: nobody told it what it must not do.
7 rules for a reliable assistant
1. A complete, official knowledge base
All your hotel information in one verified document: rooms, amenities, times, rules, conditions, distances, what you don't offer. It's the same base you use for your website, OTAs and FAQs. See what information to publish to help AI assistants understand your hotel.
2. Clear instructions on limits
Write explicitly in the assistant's instructions:
- answer only with information from the knowledge base;
- if you can't find the answer, say so and offer contact with reception;
- don't state prices or availability unless they come from the booking system;
- don't promise discounts, upgrades or exceptions;
- don't give medical, legal or safety advice.
3. Prices and availability only from connected sources
Prices change daily: an assistant must never "remember" them. Either it reads them in real time from the booking engine, or it links to the booking engine.
4. "I don't know" is a valid answer
An assistant that says "I don't have that information — let me put you in touch with reception" is far more reliable than one that always answers.
5. Handover to a person
Every conversation must be easy to pass to a human (WhatsApp, email, phone), especially for quotes, complaints and special requests.
6. Test before launch (and after every change)
Prepare a list of 50–100 real questions, including "trick" ones:
- "Do you have a spa?" (if you don't);
- "How much is a room in August?";
- "Can I bring a 40 kg dog?";
- "Will you give me a discount if I book now?"
Check every answer. Fix the knowledge base or instructions wherever it goes wrong.
7. Ongoing checks
Periodically reread real conversations. Every error you find is information to add or a rule to sharpen.
For AI-generated copy
The same rules apply to website copy, quotes, review replies and newsletters: AI prepares the draft, a person checks facts, numbers and promises. See how to use AI to prepare quotes to check before sending.
What if external AI (ChatGPT, Gemini) says wrong things about your hotel?
There you don't control the instructions, but you can improve the sources: website, Google profile, OTAs, consistent information. See AI describes your hotel incorrectly: what can you do?.
Frequently asked questions
Can AI errors be eliminated completely?
Not 100%, but they can be greatly reduced with a complete knowledge base, clear limits, connected data and human checks.
Who's responsible if the chatbot gives wrong information?
In the guest's eyes, the hotel. That's why the assistant must be designed cautiously, and contractual information (prices, conditions) must come from official sources.
Chatbot or no chatbot?
No chatbot is better than one that invents things. But a well-designed assistant can be very useful. See an AI assistant on your hotel website: when it helps and when it confuses.
Want a reliable AI assistant for your hotel?
We design AI assistants and concierges built on verified knowledge bases and clear limits, like the one we use at our own hotels. With our free marketing audit we assess whether and how to introduce one at your hotel safely.
