---
title: "We tested what AI knows about 20 independent hotels"
description: "We examined twenty independent hotels to find out what AI actually knows about them. Four sites cannot be reached at all. Not one of the other sixteen exposes availability."
url: https://www.weftd.com/resources/use-cases/what-ai-knows-about-20-independent-hotels
locale: en
alternates:
  en: https://www.weftd.com/resources/use-cases/what-ai-knows-about-20-independent-hotels
  fr: https://www.weftd.com/fr/resources/use-cases/ce-que-lia-sait-de-20-hotels-independants
---

# We tested what AI knows about 20 independent hotels

We examined twenty independent hotels to find out what AI actually knows about them. Four sites cannot be reached at all. Not one of the other sixteen exposes availability.

_Charles Cousyn · 2026-09-04 · Hospitality, AI Presence, Distribution_

A traveler asks ChatGPT, Claude or Gemini: "find me a characterful hotel in Annecy for two nights in October, quiet, with a proper breakfast." AI does not return ten links. It puts forward a handful of properties. Your hotel is in that shortlist, or it does not exist for that traveler. We wanted to know what AI can actually learn about an independent hotel today. We examined twenty of them, across five cities.

## The short answer

AI does not read your website the way a guest does. It does not see your photographs, your layout, or the care you put into your home page. It looks for usable facts: rates, availability, terms, amenities, languages spoken. On the sites we examined, that information is rarely available in a form anything can act on. AI then falls back on whatever third-party sources it can reach: OTA listings, metasearch, media, guides. Your hotel ends up living in someone else's account of it rather than your own.

## Summary

- Of our twenty hotels, four sites could not be read at all by our measurement: an expired security certificate, domains that no longer respond, a parked domain.
- Of the sixteen that load, three expose a room rate and none expose availability. The problem is not a closed door. It is an empty room.
- In the Lighthouse study, ChatGPT surfaces only 13% of the hotels in Paris and 10% in Tokyo, and 82% of the sources it draws on are OTAs, metasearch or media.

## What AI looks for, and what it finds

Here is the part almost nobody has in mind: what a visitor sees on your website and what AI reads from it are two different things. A visitor looks at photographs, senses an atmosphere, infers a standard of service. AI looks for answers to precise questions, and those answers have to be stated somewhere. They cannot be inferred from a photograph or a paragraph of prose.

| What the traveler asks | Where the answer lives today | What AI can do with it |
|---|---|---|
| Is a room still free on 12 October? | In your booking engine, often on another domain | Nothing, it does not follow it there |
| What does a double with breakfast cost? | On your OTA listing, sometimes nowhere on your own site | It quotes the OTA's rate, not yours |
| Do you accept dogs? | One sentence inside a "practical information" page | It may find it, if it reads the right page |
| What is the difference between the Cosy and the Deluxe room? | In your front desk's head | Nothing, it is written nowhere |
| Does anyone at reception speak Spanish? | Nowhere | Nothing |

The last two questions are the revealing ones. They are about what actually separates one hotel from another, and your team answers them on the phone every day. The problem is not that the information is missing. It is that it sits nowhere AI can reach.

## Our survey: twenty independent hotels, five cities

> **Our method.** We extracted from OpenStreetMap the hotels of Paris, Lyon, Marseille, Annecy and Biarritz that declare a website. After excluding chains and brands, 530 properties remained. We selected twenty of them at a fixed interval down the list sorted by domain name, with per-city quotas. On 4 September 2026 we opened each site in a browser, retrieved the code served by the server without running the page's JavaScript, which approximates what a crawler that does not render the page receives, then analysed the structured data, the rates present in the code and the third-party domains called. We also read each domain's robots.txt file.
>
> **Stated limits.** Twenty properties illustrate; they do not measure a national market. The code served is not everything an agent capable of running JavaScript would obtain. And a robots.txt file declares rules, it says nothing about whether a site can actually be used.

**First finding: some sites cannot even be reached.** Of the twenty, sixteen load normally. The four others could not be read by our measurement, and we confirmed this in a browser rather than with our own tooling alone: an expired security certificate on a four-star property in Paris, two domains that do not respond even with the www prefix, and a parked domain returning an empty seven-hundred-byte page. A human visitor can sometimes click through the browser warning. An agent cannot. It stops there.

That leaves sixteen sites on which to look at what is actually available.

| What we looked for | How many hotels |
|---|---|
| The property is identified as a hotel in structured data | 3 / 16 |
| A room rate is present in the code served | 3 / 16 |
| Availability is exposed, on any date at all | 0 / 16 |
| The booking path leaves for a third-party domain | 9 / 16 |

Seven further sites do declare structured data, but only generic types: web page, web site, organisation. Useful to a search engine, silent on the fact that this is a hotel, on its rooms and on its services. Six declare none at all.

The most striking line is availability. Not one of the sixteen sites that load exposes it, on any date. The problem is not a closed door. It is an empty room.

## And AI crawlers?

A robots.txt file is how a website declares to crawlers what they are allowed to consult. Across our twenty domains, fourteen were readable.

**Not one of those fourteen mentions the AI crawlers we looked for**, neither GPTBot, nor ChatGPT-User, nor OAI-SearchBot, nor ClaudeBot, nor PerplexityBot, nor Google-Extended. No explicit permission, no explicit block. Two files do block BUbiNG, an academic research crawler.

What that shows is not that these hotels chose to open their presence to AI. It shows that, for most of them, the question has not been dealt with yet. There is no strategy of closing the door. There is no source prepared to be used either.

## The most telling case

The best-structured site in our sample belongs to a hotel in Biarritz. It declares its property type, its postal address, its geographic coordinates, the languages spoken, and eight room categories. That is exactly the kind of information AI can work with.

But it declares no rate at all, neither in its text nor in its structured data. AI therefore knows this hotel has eight room categories. It cannot answer a single question about their price.

The site is perfectly readable. It is simply not queryable.

## What AI recommends today

The pattern reaches far beyond our sample. Lighthouse submitted 4,545 prompts to ChatGPT across nine global destinations and five traveler profiles. The result: 49,707 hotel mentions, but only 2,721 distinct properties ([Lighthouse, Luminate 2026 study](https://www.mylighthouse.com/resources/blog/ai-recommendations-as-new-battleground-for-hotel-distribution)).

| What the study measures | Result |
|---|---|
| Share of a market's hotels that ChatGPT actually surfaces | 13% in Paris, 10% in Tokyo |
| Share of sources coming from OTAs, metasearch or media | 82% |
| Share of all mentions captured by the 100 most-recommended hotels worldwide | Over 13% |

One qualification worth stating, because it cuts against the easy story: Paris is the only market in the study where independents beat chains in share of mentions, even if the gap is narrow. Independence is not a sentence. But at 13% coverage, the vast majority of Paris hotels never appear at all.

The issue, then, is not simply being present on the web. It is being usable at the moment AI assembles its answer.

Our survey does not prove that AI favours OTAs because hotel websites are poorly structured. It shows something simpler: the sites we examined hold very little of the information AI needs to answer a precise question. And the Lighthouse study shows that third-party sources occupy a very large share of ChatGPT's hospitality answers. The two meet in the middle.

## Commission first, then the narrative

Hospitality already knows what intermediation costs. Across Europe, 71.2% of hotels' online revenue came through indirect channels in 2023, and Booking.com alone accounted for 43% of online revenue ([D-EDGE](https://www.d-edge.com/hotel-distribution-report-2024-have-direct-bookings-reached-a-peak/)). Among independent hotels, 63.4% of bookings run through OTAs worldwide, with commissions of 15% to 30% and a cancellation rate of 21.8% against 10.6% on direct bookings ([Cloudbeds](https://www.cloudbeds.com/hospitality-industry-report/)).

France, where our sample was drawn, shows the same shape at national scale: 11,555 of its 15,155 hotels are independent in the sense used by the national statistics office, meaning not affiliated to an integrated chain, or 76% of the market, for 214.5 million overnight stays including 77 million from international guests ([INSEE, 2024](https://www.insee.fr/fr/statistiques/2015412)). This is not a niche segment. It is the core of the market, and the most exposed part of it.

A second layer of intermediation can now form on top of the first. If the sources most easily used by AI remain those of the platforms, then platforms no longer serve only to book: they become sources of knowledge. They no longer capture only the reservation. They capture the narrative.

The law, for its part, has moved the other way. Booking.com has been subject to the obligations of the Digital Markets Act since 14 November 2024, which among other things lets a hotel offer better conditions on its own channels ([European Commission](https://digital-markets-act.ec.europa.eu/booking-must-comply-all-relevant-obligations-under-digital-markets-act-2024-11-14_en)). France had already made price-parity clauses unenforceable back in 2015 ([Légifrance](https://www.legifrance.gouv.fr/loda/id/JORFTEXT000030978561)). The direct channel exists. What remains is a discovery problem: being found before the platform is.

## The groups have already moved

This is not a forecast. OpenAI opened ChatGPT to third-party applications, and the major groups are already exposing their knowledge inside it. IHG's application covers more than 7,000 hotels, with natural-language search, availability, rates and amenities ([Hotel Dive](https://www.hoteldive.com/news/ihg-hotels-resorts-launches-chatgpt-app/821979/)). Accor's is offered in more than twenty languages and routes travelers to the group's direct booking ([Accor](https://group.accor.com/en/news-stories/accor-leading-hospitality-ai)). Marriott launched Ask Bonvoy, Hilton an AI trip planner ([Hotel Dive](https://www.hoteldive.com/news/marriott-debuts-ai-powered-hotel-search-tool-ask-bonvoy/823130/)).

Look at what these experiences have in common. They do not merely present a brand: they expose structured, current knowledge to AI, rates and availability included. That is exactly what none of the twenty sites in our survey does.

The groups can fund those integrations, those data infrastructures and those proprietary applications. An independent hotel will generally not build that layer on its own. So the gap is not opening on the quality of the website. It is opening on the preparation of the information.

None of this is marginal on the traveler's side: 45% of travelers already plan their trip with AI, with wide variation by market, up to 88% in China and 82% in India ([Allianz Partners / Ipsos, 2026](https://www.allianz-partners.com/content/dam/onemarketing/awp/azpartnerscom/en_global/reports/travel-confidence-index-2026/Allianz_Partners_Global_Travel_Confidence_Index_Report.pdf)).

## Citable is not the same as usable

Two channels now coexist, and they do not replace one another.

| | Citation channel | Agentic channel |
|---|---|---|
| What you optimise | Your pages, your search ranking | Your knowledge, structured |
| What AI does | Finds you, summarises you, sometimes cites you | Queries your source and answers with your data |
| Who controls the information | Search engines and AI | You |
| What you get | Visibility | The ability to answer and convert |

SEO and GEO remain indispensable. They improve your odds of appearing. But they optimise what outside systems decide to use: you control neither the final answer, nor the other sources drawn on, nor the way your property is described. Your website can be perfectly visible without being usable by an agent. One makes you citable. The other makes you activatable.

That gap is exactly what [Weftd sets out to close for hospitality](/solutions/hotel): building an official source that is structured and queryable by AI, alongside the website and the search ranking you already have.

## Where to start

There is no technical project to launch before doing this groundwork. Start with four things.

1. **Check that your website loads.** It sounds obvious. One site in five in our sample was unreachable to our survey, usually without anyone having noticed.
2. **Write down what only your team knows.** The real difference between your room categories, your pet policy, your late arrival times, the languages spoken, what your breakfast is actually worth.
3. **Gather that knowledge in one official place.** Not another page on the site: a single, current source you know to be authoritative.
4. **Make it queryable.** A source AI can consult directly, in the traveler's language, without going through a third-party listing.

The point is not to produce more content. It is to formalise what your team already knows, keep it current, and hold an official source AI can genuinely use.

The channel already exists. The only open question is who represents your hotel in it: you, or somebody else.

[See how Weftd applies to hospitality](/solutions/hotel)

The hospitality white paper, which documents these mechanisms in full, is available from that same page.

## FAQ

**Should hotels block AI crawlers in their robots.txt file?**

For an independent hotel there is no general reason to. Blocking means disappearing from a channel where your competitors already show up. Our survey cannot tell you that all twenty hotels are open: only fourteen robots.txt files were readable, and none of those fourteen explicitly blocked the AI crawlers we looked for. The real question is elsewhere: what does AI find once it reaches your site?

**Is structured data, schema.org, enough on its own?**

It helps, and it belongs to the citation channel. But it is not in itself a source AI can query. Our best example makes the point: one hotel declares eight room categories in structured data and not a single price. AI knows the rooms exist. It cannot answer "what does the Deluxe cost?".

**Does this replace my booking engine?**

No. Your rates, your availability and your transactions stay in your systems. An AI Presence supplies the knowledge and a usable interface, then routes the traveler to your direct channel. Payment never leaves your engine.

**Does this replace my search ranking?**

No. SEO and GEO work on discovery and citation. An agentic presence works on the direct use of your knowledge. The two add up.

**Can AI already book a room?**

It depends on the assistant and the integration. The major groups already offer experiences inside ChatGPT where a traveler can search for a hotel, consult rates and availability, then continue to their direct booking. For an independent property, the immediate stake is not the transaction inside the chat: it is being described correctly and routed to your own channel rather than to an OTA listing.

**My site ranks well on Google. Isn't that enough?**

Ranking well means a search engine can find you. It does not mean AI can answer a precise question about your hotel. That is exactly what our survey shows: of the sixteen sites that load, only three exposed a room rate in the code served, and none exposed availability.

## Sources

- Weftd survey, 4 September 2026. Twenty independent French hotels selected from 530 properties listed on [OpenStreetMap](https://www.openstreetmap.org) (Paris, Lyon, Marseille, Annecy, Biarritz), chains excluded, fixed-interval selection down the list sorted by domain. Analysis of the code served by each site without running JavaScript, of structured data, rates and third-party domains, plus a reading of the robots.txt files, fourteen of twenty usable. The full method and the per-property detail are available on request.
- [Lighthouse, Luminate 2026 study](https://www.mylighthouse.com/resources/blog/ai-recommendations-as-new-battleground-for-hotel-distribution): 4,545 prompts submitted to ChatGPT across nine destinations and five traveler profiles, 49,707 mentions, 2,721 distinct properties, 13% coverage in Paris and 10% in Tokyo, 82% of sources from OTAs, metasearch and media.
- [D-EDGE, Hotel Distribution Report 2024](https://www.d-edge.com/hotel-distribution-report-2024-have-direct-bookings-reached-a-peak/): 71.2% of online revenue from indirect channels in Europe, Booking.com at 43% of online revenue.
- [Cloudbeds, State of Independent Hotels 2026](https://www.cloudbeds.com/hospitality-industry-report/): 63.4% OTA bookings among independents, cancellations 21.8% against 10.6% direct. [OTA commissions](https://www.cloudbeds.com/online-travel-agencies/commissions): 15% to 30%.
- [INSEE, French hotel inventory 2024](https://www.insee.fr/fr/statistiques/2015412): 15,155 hotels, 11,555 independent, 214.5 million overnight stays including 77 million international.
- [Allianz Partners / Ipsos, Global Travel Confidence Index 2026](https://www.allianz-partners.com/content/dam/onemarketing/awp/azpartnerscom/en_global/reports/travel-confidence-index-2026/Allianz_Partners_Global_Travel_Confidence_Index_Report.pdf): 45% of travelers plan their trip with AI, 88% in China, 82% in India.
- [European Commission, Digital Markets Act](https://digital-markets-act.ec.europa.eu/booking-must-comply-all-relevant-obligations-under-digital-markets-act-2024-11-14_en): obligations applicable to Booking.com since 14 November 2024. [Court of Justice of the European Union, Case C-264/23, 19 September 2024](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A62023CJ0264). [Légifrance, Law of 6 August 2015, article 133](https://www.legifrance.gouv.fr/loda/id/JORFTEXT000030978561).
- [Hotel Dive, IHG](https://www.hoteldive.com/news/ihg-hotels-resorts-launches-chatgpt-app/821979/) · [Hotel Dive, Marriott](https://www.hoteldive.com/news/marriott-debuts-ai-powered-hotel-search-tool-ask-bonvoy/823130/) · [Hotel Dive, Hilton](https://www.hoteldive.com/news/hilton-launches-generative-ai-agent/814424/) · [Accor](https://group.accor.com/en/news-stories/accor-leading-hospitality-ai).

Read next: [Getting cited by AI isn't enough](/resources/insights/getting-cited-by-ai-isnt-enough) and [How an AI conversation turns into a booking or a sale](/resources/insights/ai-conversation-to-booking-or-sale).