ChatGPT agent prepares a Gatwick hotel booking
Peter Hartree ran ChatGPT agent on a Gatwick hotel search three times. It chose his preferred property and reached card verification, then handed control back without completing payment.
wow.pjh.is
Original prompt
Shared by the source author. See source notes for context.
Please book a Gatwick airport hotel for October 14th.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Peter Hartree
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Three reported Gatwick runs chose the owner’s preferred hotel and reached card verification; completed payment is not established.
- Human involvement
- Hartree defined preferences and handled the first login. The agent requested a card-security-code handoff; checking the chosen listing and alternatives was his planned review before payment.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Set Booking.com filters, ranking rules and a human confirmation checkpoint.
- Let the agent compare Gatwick hotels and navigate to the preferred option.
- Take over when card verification is required and inspect hotel facts before proceeding.
Peter Hartree tested whether ChatGPT agent could select the airport hotel he would choose himself. He supplied detailed Booking.com instructions covering review thresholds, facilities, transfer times, price comparisons and a final confirmation checkpoint.
Across three Gatwick runs, he reports that the agent selected his preferred hotel and reached the screen asking for the card security code. It then requested human takeover. Each run took about five to ten minutes. His account therefore supports hotel selection and checkout preparation, rather than an unattended completed booking.
The rules limited website choice and specified the exact filter names, reducing repeated exploration of unfamiliar interfaces. On an additional Reykjavik trial, an air-conditioning requirement excluded the available hotels with gyms. Hartree removed that requirement and reports much better results.
The account also records inaccuracies: a hotel was described with the wrong star rating, and a terminal walk was understated. Hartree planned to inspect the selected listing, alternatives and search results before entering payment information. Earlier broad searches had taken longer and sometimes crashed. The case shows a useful bounded research and checkout workflow, with the owner still verifying hotel facts and completing the consequential payment step.
Prompt
Original prompt
Source: Blog · Peter Hartree
Frequently asked questions
Was the hotel booked without human involvement?
The reported runs reached the card-security-code screen and requested takeover; the account does not establish unattended payment.
What errors occurred?
The author reports an incorrect star rating and an understated walk to the terminal in separate runs.
Why did Reykjavik initially fail?
The air-conditioning filter excluded hotels that had gyms; removing it improved the results.
More details
About this account
The agent filters and compares Booking.com listings and navigates to the CVC takeover screen.
Firsthand primary account opened and read; original publication date checked in primary metadata or visible timestamp. Outcomes are author-reported, not independently reproduced. Checkout reached; no completed booking or payment claimed. Original prompt excerpt reproduced exactly from the source; excerpt only, not the complete prompt.
Credit: Peter Hartree
Source published
Added to Agents In Use
Source checked
Tools and topics
ChatGPT agent browser · Booking.com
Travel · Reported deployment
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.