AI is not one channel, it is six
Each AI surface decides which hotels to recommend based on a different lever. Measuring "AI traffic" as a single channel hides that Google, ChatGPT and Perplexity are rewarding different things, and often different hotels.
- Only 18% of the hotels named by ChatGPT and Google AI Mode overlap, in an analysis of 695 real searches.
- 71% of AI answers about travel lean on first-hand reviews, with Perplexity at 100%.
- Almost half of the hotels AI recommends do not appear among Google's organic results for that same search.
- Each surface rewards a different lever: a complete listing, editorial placement, recent reviews or structured data.
- A sixth surface is missing from the published frameworks: voice.
The framework, and where it comes from
I started thinking about this while reading Gourmet Marketing, a hotel marketing agency that insists on testing each assistant separately: "ask ChatGPT, Claude, Perplexity and Gemini the questions your ideal guest would ask", in their own words. Cross that advice with two studies of real data and you get a more useful framework than talking about "AI traffic" in general: do not split by assistant brand, split by the lever that wins the placement in each one. The question that really matters is not "do I show up in AI", it is "what do I have to change so they name me here".
| Surface | What gets you named |
|---|---|
| Google AI Overviews / AI Mode | Prominence in Maps and a complete Google listing |
| ChatGPT | Editorial placement, appearing in third-party roundups |
| Perplexity | Depth and recency of reviews |
| OTAs and review platforms | The source the AI uses to describe you |
| Agentic booking | Structured rate and policy data |
| Voice | Structured data for voice search and assistant integration |
Why it matters: the engines do not agree with each other
A Tharro study of 695 hotel searches in the Algarve, Mallorca and Rhodes puts numbers on it instead of guesses:
That 18% overlap is the proof that measuring "AI" as a single thing is a measurable error, not a harmless simplification. The 66% versus 27% ratio confirms something intuitive that is worth backing with data: being in Google's top 3 almost doubles your chance of being named by AI compared with being outside the top 20. And even so, almost half of the hotels AI recommends did not appear among Google's organic results for that search: ranking helps, but it does not come close to explaining the whole recommendation.
The real gap: reviews
A Feefo analysis of real travel queries, published in August 2025, found that 71% of AI answers leaned on first-hand reviews. By engine: Perplexity 100%, ChatGPT 58%, Gemini 56%.
Most hotels measure AI clicks and mentions, but do not watch the volume or recency of their reviews as a lever for AI visibility. It is the connection almost nobody is making yet: a hotel that Perplexity does not name probably has a reviews problem before it has a content problem.
The missing surface: voice
The frameworks published in 2026 are written from text chat and leave voice out: home speakers, voice booking assistants, conversational concierge inside the hotel itself. It is not a minor oversight. Voice has its own retrieval rules (specific structured data, integration with assistant skills, voice search results) and in hospitality it is still a real touchpoint, not a curiosity from 2018.
Treating it as a sixth surface, with its own lever, is more honest than folding it into "agentic booking" or ignoring it.
What cannot be concluded from this: neither the Tharro study nor the Feefo one publishes its full sampling methodology, so these figures are a snapshot of a specific moment and market, not a general law. They serve to illustrate the size of the problem, not to set a universal percentage.
What to do with this
Before investing in "showing up in AI" as if it were one thing, decide which surface your customer is on and optimise that specific lever. If you sell through agencies and press, editorial placement weighs more than your Google listing. If your customer compares several options in depth, recent reviews weigh more than anything else. Measuring the whole as "AI traffic" is only good for the headline of a meeting, not for deciding where to work first.
Questions
Why is measuring AI traffic as a single channel useless for hotels?
Because each AI surface decides who to recommend based on a different lever: Google AI Overviews looks at your listing and your reviews, ChatGPT leans on third-party editorial placement, Perplexity weighs the depth and recency of reviews (it leans on reviews in 100% of its travel answers), and agentic booking needs structured rate data. Optimising for one of those things does not move the others.
Do Google AI Overviews and ChatGPT recommend the same hotels?
Not necessarily. An analysis of 695 hotel searches in the Algarve, Mallorca and Rhodes found only an 18% overlap between the hotels named by ChatGPT and by Google AI Mode. Measuring AI as one thing hides that each engine is recommending different hotels.
How much do reviews weigh in what AI recommends for hotels?
A lot. In real travel queries analysed by Feefo, 71% of AI answers leaned on first-hand reviews, with Perplexity leaning on reviews in 100% of cases, ChatGPT in 58% and Gemini in 56%.
Is voice missing from this framework of AI surfaces for hotels?
Yes. The frameworks published in 2026 focus on text chat and leave voice out: smart speakers, voice booking assistants and conversational concierge. It is a surface with its own rules, especially in hospitality, and it is better treated as a sixth lever than folded into the other five.
Sources: Gourmet Marketing · Tharro, hotel AI visibility study · Hotel News Resource, coverage of the Tharro study · Feefo, how reviews shape AI travel recommendations.
Joan Sanz. Fifteen years in travel and hotel tech. I read these frameworks and test them before repeating them. More at /about.
Do you know which surface your customer is on?
I can help you find out and decide where to work first.
Message me on LinkedIn