Hospitality Technology

ChatGPT Quoted Hotel Prices in 95 of 99 Test Queries — but Named the Date Only 18 Times

Lumina Frontier put 99 hotel price questions to ChatGPT on 31 October. Prices came back almost every time. The date behind them came back rarely, and booking sites far outranked hotel direct pages.

When a guest asks ChatGPT what your hotel costs, whose price do they hear?
When a guest asks ChatGPT what your hotel costs, whose price do they hear? — Rodrigo_SalomonHC / pixabay

A test of 99 ChatGPT hotel price queries returned a price in 95 answers, but only 18 specified which night that price bought. The operator behind the test, Lumina Frontier, ran the queries on 31 October with web search enabled and a US location, and saved both the answers and the links behind them.

Marco Di Cesare, Founder and CEO of Lumina Frontier, a Zurich company building an AI marketing team for hotels, published the wording and results in a study of hotel prices in ChatGPT. The study grew out of an October incident. A reader sent Lumina the answer ChatGPT had given him about the Venetian in Las Vegas. The model had used Lumina's guide, and it flagged the figures with a caveat: "historical quotes, not current offers". It also wanted the prices compared on the same suite, on the same night, with the cancellation terms, and with an answer to a very Las Vegas question. Was the resort fee included?

Whose price does the guest hear?

The distribution of cited sources favors intermediaries. Of the 99 answers, 80 cited a booking site or Google Hotels. Lumina could identify the hotel's own site, or its group's site, in 11 answers. An answer could cite both.

The pattern shifts when the guest asks a different question. When the query was whether to book direct, Lumina found an official hotel or group site in 61 of 100 answers. The phrasing of the question, not the hotel's pricing strategy, changes which source the model reaches for.

For a revenue manager, the missing date matters more than another decimal place. On 15 November, Lumina could give two correct answers to the question "What does a room at Ashford Castle cost?" The castle's own booking engine offered a Corrib room for two at €527 on Tuesday 9 February and €1,250 on Tuesday 15 June. Both dates fall in 2027. Both prices were correct when checked. The person at reservations who picks up the phone after a guest has heard only the first number has a problem no rate parity clause will fix.

What can the hotel say that a booking site cannot?

The useful detail hid in the offer structure. At Ashford Castle, the winter offer on that February Tuesday cost €527, while the same Corrib room on the bed-and-breakfast rate cost €585. The cheaper offer included breakfast and a welcome drink. By Friday, the winter offer no longer started that night. A visitor who knew only the room name and the first price had little chance of working that out.

The official pages ChatGPT cited elsewhere showed what hotels can do. Lumina's Di Cesare pointed to Marina Bay Sands' booking terms, the Broadmoor's policies, and an offer at Resorts World Las Vegas that explicitly removed the resort fee. Those pages answered questions a guest would otherwise have to call and ask.

What should revenue and marketing publish together?

Di Cesare's prescription is narrow and operational. Start with the offer reservations has to explain most often. Have revenue pick a real room and two dates that show why its price changes. Marketing puts both on the offer page, with:

  • the total for two
  • what breakfast costs
  • when the guest can cancel
  • the date the figures were checked
  • a link to live availability for the guest ready to book

Then give that page to whoever answers the phone on Friday. If they still have to explain why Tuesday's rate has disappeared, that is the sentence Di Cesare asks them to write.

The study does not claim to cover all models or all markets. It tested one model, one day, one location setting, and questions Lumina wrote itself, apart from the reader's Venetian query. Whether other AI assistants show the same tilt toward booking sites remains untested. What the data does establish is that the price a guest hears before calling is often undated, often sourced from an intermediary, and sometimes months stale by design.

ai-searchhotel-pricingrevenue-managementdirect-bookingshotel-marketing

More from Marcus Bennett

Marcus Bennett

Show full bio

Market editor covering technology and product news in Restaurant and hotel operations at The Pass Brief.

798 articles

Pairings

« Previous articleNext article »