AI Hotel Landscape
How leading AI assistants recommend hotels — 616 prompts across 56 destinations, refreshed every Monday.
The 616 prompts we ask ChatGPT every week
56 destinations × 11 templates. Each prompt has structured dimensions: city, country, region, persona (couples / families / solo / business), budget (luxury / mid / budget), and location zoom (wide city vs neighborhood vs landmark). Control prompts are unmodified baselines.
Every destination below gets the same 11 question types — just with the city name swapped in. That's how 56 destinations × 11 templates = 616 prompts per platform per week.
- luxury hotels in <city>
- family friendly hotels in <city>
- hotels in <city> with rooftop pool
- hotels near <neighborhood>, <city>
- boutique hotels in <neighborhood>, <city>
- best hotels in <city> for solo travelers
- best hotels in <city>
- affordable hotels in <city> under $200
- best hotels in <city> for couples
- best hotels in <city> for business travelers
- best hotels in <neighborhood>, <city>
- hotels near <city> National Park
africa4 destinations · 44 prompts
americas14 destinations · 154 prompts
Each destination is asked the 11 prompt patterns above (11 prompts × 14 cities = 154 prompts in this region).
asia16 destinations · 176 prompts
europe17 destinations · 187 prompts
oceania5 destinations · 55 prompts
Each destination is asked the 11 prompt patterns above (11 prompts × 5 cities = 55 prompts in this region).
Named hotels in Bali — ChatGPT
ChatGPT's top picks specifically in Bali, ID. Pick a different destination above, or clear to see worldwide.
Most-mentioned hotel by ChatGPT this week.
When ChatGPT hands out a hotel link, where does it go?Worldwide
Every URL the AI returned for a named hotel, classified into direct (the hotel's own site), OTA (Booking / Expedia / …), chain page (marriott.com / hilton.com / …), or other. Chain folds into direct in the headline number because chain pages are still brand-controlled.
Sponsored placementsWorldwide
OpenAI is rolling out paid sponsor placements in the US, Australia, New Zealand, and Canada — testing may expand. We detect real single-advertiser ad units (organic shopping cards excluded) and pull the brand.
| # | Advertiser | Captures |
|---|---|---|
| 1 | Expedia | 292 |
| 2 | IHG Hotels & Resorts | 98 |
| 3 | Marriott | 26 |
| 4 | Hilton | 23 |
| 5 | GetYourGuide | 18 |
| 6 | Gate 1 Travel | 7 |
| 7 | Super.com | 6 |
| 8 | Choice Hotels | 2 |
| 9 | Priceline | 2 |
| 10 | Preply | 1 |
Where ChatGPT pulls hotel information from
Every URL ChatGPT cited classified into 12 buckets: OTAs, editorial, chain sites, direct hotel pages, social, meta-search, AI tools, community, government, directory consortia, and other.
Other enprimeurclub.com 51% · theblackstonehotel.com 26% · luxstay.world 23% | 23.1% | 52 |
Independent Hotel panpacific.com 59% · hotel.hardrock.com 21% · lepremierhotel.com 20% | 20.9% | 47 |
Chain marriott.com 60% · hyatt.com 21% · all.accor.com 18% | 12.9% | 29 |
Review Site tripadvisor.com 62% · oyster.com 35% · tripadvisor.ca 3% | 11.6% | 26 |
OTA & Aggregator expedia.com 55% · booking.com 24% · hotels.com 21% | 8.9% | 20 |
Community hotelierschoice.com 50% · travelmyth.com 35% · budgetyourtrip.com 15% | 7.1% | 16 |
Editorial thehotelguru.com 39% · timeout.com 32% · vogue.com 29% | 6.2% | 14 |
Meta-search google.com 62% · kayak.com 24% · momondo.com 14% | 4.9% | 11 |
Top hotel parent groups by ChatGPT mentionsWorldwide
Each ChatGPT mention matched against Google Places, then rolled up to parent group (Marriott = Ritz-Carlton + Westin + Sheraton + EDITION + …, etc.). WoW delta vs last week. Detection via 175+ brand-domain rules.
| 1 | Lungarno | 4 → 12 | +200% |
| 2 | Leela | 8 → 13 | +63% |
| 3 | Shangri-La | 28 → 42 | +50% |
| 4 | Minor | 26 → 31 | +19% |
| 5 | Peninsula | 28 → 33 | +18% |
| 1 | Capella | 18 → 10 | -44% |
| 2 | Rosewood | 24 → 15 | -38% |
| 3 | Accor | 209 → 141 | -33% |
| 4 | Wyndham | 19 → 14 | -26% |
| 5 | Mandarin Oriental | 55 → 44 | -20% |
| # | Brand | Mentions | Hotels | W/W | |
|---|---|---|---|---|---|
| 1 | Marriott | 327 | 179 | · 0% | |
| 2 | Accor | 141 | 89 | ↓ 32.5% | |
| 3 | Hilton | 131 | 83 | ↓ 2.2% | |
| 4 | Hyatt | 121 | 61 | ↓ 4.0% | |
| 5 | Four Seasons | 98 | 36 | ↑ 5.4% | |
| 6 | IHG | 94 | 69 | ↓ 16.8% | |
| 7 | Mandarin Oriental | 44 | 17 | ↓ 20.0% | |
| 8 | Shangri-La | 42 | 18 | ↑ 50.0% | |
| 9 | Peninsula | 33 | 9 | ↑ 17.9% | |
| 10 | Minor | 31 | 20 | ↑ 19.2% | |
| 11 | Kempinski | 26 | 9 | ↓ 7.1% | |
| 12 | Langham | 24 | 9 | ↓ 4.0% | |
| 13 | Taj | 22 | 10 | · 0% | |
| 14 | Radisson | 21 | 16 | ↑ 5.0% | |
| 15 | Meliá | 20 | 12 | ↑ 11.1% | |
| 16 | Rosewood | 15 | 6 | ↓ 37.5% | |
| 17 | Wyndham | 14 | 11 | ↓ 26.3% | |
| 18 | Oberoi | 14 | 6 | ↑ 7.7% | |
| 19 | Leela | 13 | 2 | ↑ 62.5% | |
| 20 | Lungarno | 12 | 3 | ↑ 200.0% |
Of the hotels named this week: how many resolved to a chain (Marriott / Accor / …), an independent property, a vacation rental, or stayed unmatched.
What every metric means
Plain-English definitions for each number on this page. For deeper visuals + examples, see the annual landscape report linked at the bottom.
- Captures
- Number of AI responses we collected this week. Target = 616 per platform (one per prompt). Less when there are upstream errors.
- Web search
- Did the response trigger a live web fetch? Detected from the underlying response stream’s search-result events, not the unreliable top-level flag. We do NOT force web search — this is organic model behavior.
- Map widget
- Did the response render a hotel-card map (the Google-Maps-style widget ChatGPT shows for travel queries)? Each card is a "map entity" with its own provider.
- Sponsored placements
- Paid sponsor placements. Detected from real single-advertiser ad units in the response stream. Excludes ChatGPT’s organic shopping cards (which are unpaid product carousels). US/AU/NZ/CA only as of May 2026.
- Sources / response
- Distinct URLs the model consulted while answering — the URLs that show up in its retrieval log, regardless of whether it cited them inline. Computed only over web-search responses (otherwise the answer is from training data, no sources to count).
- Citations / response
- Subset of sources that the model rendered as inline footnote pills in the answer. A source becomes a citation when the model explicitly references it.
- Fanouts / response
- Number of sub-queries the model spun up internally to answer the prompt (e.g. "best hotels Paris" might fan out to "hotels Paris Marais", "luxury hotels Paris", "rooftop hotels Paris", …).
- Map entities / response
- Hotel cards inside the map widget. Each carries a provider (Google Places, TripAdvisor, Yelp, Foursquare, SERP) and a place ID where applicable.
- OTA
- Online travel agency (Booking, Expedia, Hotels.com, Agoda, Trip.com, Priceline, etc.) — commission-based booking sites.
- Direct
- A hotel’s own website. Computed at request time by matching the cited domain against Google Places.
- Chain
- A hotel chain’s brand website (hilton.com, marriott.com, ihg.com, …). 175+ chain-domain rules.
- Editorial
- Travel media (Condé Nast Traveler, Time Out, Lonely Planet, Forbes Travel Guide, NYT, etc.) and city-specific travel blogs (santorinidave.com, theurbanlist.com, etc.).
- Directory
- Multi-property hotel consortia / brand collectives (Small Luxury Hotels of the World, Design Hotels, Preferred Hotels, Virtuoso, …). Aggregate hotels under one umbrella but aren’t a single OTA.
- Review
- TripAdvisor, Oyster, Yelp, Trustpilot, Holidaycheck — review-first platforms.