When AI Hallucinates Hotel Domains
Large Language Models frequently invent direct booking links — and that's a real risk for hoteliers.
AI and SEO expert at the forefront of AI Search. He analyses models daily and runs hospitality-focused experiments on a database of over 1M prompts, citations and mentions.
Introduction
As ChatGPT becomes a key channel for hotel discovery, something strange is happening: “hallucinated” domains. Large Language Models (LLMs) generate plausible—but not always factual—text.
The result? We’re now seeing hundreds of cases where AI tools recommend websites… that don’t even exist.
Why AIs hallucinate domain names—and what to do about it
At HotelRank.ai, we analyze thousands of responses generated by AIs like ChatGPT, Perplexity, or Gemini. And the findings are surprising: these models hallucinate not just pages—but entire domain names.
This isn’t new. AIs have always “hallucinated.”
On September 5, 2025, OpenAI published a research paper explaining why ChatGPT hallucinates—and showing that GPT-5 hallucinates significantly less.
In short: the more a model “thinks,” the less it hallucinates.
So I ran an analysis on 5-star hotel recommendations in various destinations: Mexico City, London, Bangkok, Courchevel, Manila, Shanghai, and Bordeaux.
Here are a few real examples from Saint-Tropez, involving 5-star hotels:
- bastide-st-tropez.com
- kubehotel-st-tropez.com
- bastidesttropez.com
These domains do not exist.
Yet AI models suggest them as direct links to real hotels like La Bastide de Saint-Tropez or Kube Hotel Saint-Tropez.
A fairly common phenomenon: 2 to 3% of links are fake
Across more than 10,000 AI-generated hotel links analyzed by HotelRank.ai, we found that between 2% and 3% of them are incorrect or completely fictional.
Most common errors:
- Domain doesn’t exist (never registered)
- Misspelled brand name
- Wrong TLD (.fr instead of .com, or vice versa)
- Merging of two existing brands
- Awkward translation of the hotel name
Some cities go completely off the rails: Bordeaux
In Bordeaux, more than 20% of domains are unverified (!)
The AI struggles to guess the correct site for hotels like Palais Gallien, Hôtel des Quinconces, or even the Intercontinental.
And it’s not just the wine talking—it’s the same story in Batumi (Georgia) and Megève.
Hallucinated domains % by city
Hotelrank.ai reserach data
Consequences for hoteliers
These hallucinations have real business impact:
- The displayed link is invalid
- The user clicks, then bounces back
- The booking may go to an OTA or competitor
- The brand perception suffers (even more for luxury hotels)
Should you buy hallucinated domains?
In traditional SEO, there’s a tactic called domain drop catching—buying expired domains to capture “SEO juice,” traffic, and authority.
But here, we’re talking about something new:
→ Domains that were never created, but are treated as real by AI.
This opens the door for bad actors to:
- Buy these hallucinated domains and rank for AI queries
- Redirect to competitors or affiliate pages
- Create clickbait traps or even malicious content
And it costs… less than €10 per year.
A new layer of visibility to monitor
AI hallucinations aren’t just a technical glitch.
They shape the digital image of a hotel, and its ability to exist in emerging discovery channels.
When a chatbot, a voice assistant, or an AI-first interface shows just one link… it’s a shame if it’s the wrong one.
How to respond?
At HotelRank.ai, we actively monitor these hallucinations—model by model—and aim to build a hotel-level database of risky or non-existent domains.
We help hoteliers:
- Identify errors around their brand
- Track how AI suggestions evolve over time
- Compare visibility against OTAs
- Consider securing key domains before someone else does
Nicolas Sitter
AI and SEO expert at the forefront of AI Search. He analyses models daily and runs hospitality-focused experiments on a database of over 1M prompts, citations and mentions.
Co-founder of Hotelrank