ChatGPT Ads hotel feeds: how the property feed beta works

ChatGPT Ads hotel property feeds let hotels and travel advertisers send their properties to Ads Manager. Each row describes one property. OpenAI calls it a limited beta, open only to approved pilot advertisers. The feed has its own format, separate from the product feed used by stores. This guide covers who can use it, how the file works and the checks that stop rows being rejected.

Every platform fact links to OpenAI’s documentation. Where we give our own view, we say so.

What a ChatGPT Ads hotel feed is

A hotel feed describes properties, not rooms, offers or dated stays. Each row is one hotel, with its name, description, landing page, images and location. Optional columns add a brand, a static starting price, a star rating, a category and guest ratings.

Hotel feed	Product feed
One row per property	One row per product or variant
The hotel_property_v1 format	OpenAI’s product feed specification
Full snapshots only	Full uploads, plus the Delta API for quick changes
Limited beta, approved pilot advertisers only	Available to advertisers with a feed

Don’t use the product feed schema for a hotel feed. OpenAI says the two formats are separate, and the field definitions may change during the beta.

Who can use it

OpenAI says hotel property feeds are available only to approved pilot advertisers, and that this isn’t an open beta. You need an ad account approved for the beta and permission to manage its feeds. Setting up SFTP also needs permission to manage feed credentials.

If the Hotels option doesn’t appear in your account, OpenAI says to contact your OpenAI account team. Our view: travel brands should prepare the file anyway. A clean property list with good images and pages is useful for every channel, and you’ll be ready if access opens.

How to create a hotel feed in Ads Manager

OpenAI lists five steps.

1. Open Ads Manager and select the approved ad account.
2. Open Feeds, select Create Feed and choose Hotels as the inventory type.
3. Name the feed and choose how to send the file: Upload CSV or TXT, Hosted URL or SFTP connection.
4. Create the feed and finish the setup for the method you chose.
5. Check Upload History for processing status and row diagnostics, then fix rejected rows and upload a corrected full snapshot.

Keep one feed and refresh it with new snapshots. Don’t create a new feed for every update, and keep each hotel_id the same between uploads.

Through the Ads API, you create the feed with the hotel inventory type and schema profile, then set up SFTP. Hosted URLs and browser uploads are Ads Manager workflows only.

The file format

Send a full snapshot as a UTF-8 CSV or tab-delimited TXT file. Most values are ordinary columns, but some must be JSON inside a cell, and some use flattened column names.

Value	How it’s written
Scalar fields	Ordinary columns, such as hotel_id, name or base_price
Images	A JSON array in the images cell
Location	Flattened columns, such as location.address.city
Neighborhoods	A JSON array in the location.neighborhoods cell
Guest ratings	Numbered, flattened columns: guest_rating[0].score and so on
Ads metadata	A JSON object in the ads_metadata cell

Don’t add top-level columns called location or guest_ratings. In CSV, wrap any cell containing commas, quotation marks or line breaks in double quotes, and double any quotes inside it.

Required fields

Every row needs these fields, with valid values, even rows marked as archived.

Column	Rules
hotel_id	A stable, case-sensitive ID, up to 100 characters
name	The display name, up to 150 characters
description	A static description, up to 5,000 characters
url	The property’s landing page, as a full web address
images	1 to 20 images, as a JSON array. The first valid image is the main one
location.address.city	The city, up to 100 characters
location.address.country_code	Two-letter country code, such as US
location.coordinates.latitude	A number from -90 to 90
location.coordinates.longitude	A number from -180 to 180

The country code describes where the property is, not where you want to show ads. Ad locations are set on the campaign. Landing pages and images must be publicly accessible.

Optional fields worth adding

Column	What it adds	Rules
brand	The hotel or chain brand	Up to 70 characters, separate from your advertiser name
base_price	A static starting price	An amount and currency code, such as 149.00 USD
star_rating	The hotel’s star rating	1.0 to 5.0 in half steps
category	One hotel category	Up to 200 characters
location.neighborhoods	Up to 20 neighborhood names	A JSON array
guest_rating[n].*	Up to 20 rating groups	Each needs system, score, maximum score and reviewer count
ads_metadata	Your own labels	Up to 50 entries, as a JSON object
is_ads_eligible	Whether to process the property for ads	true or false

The base price isn’t a quote. OpenAI says it’s a static starting price, not a price for particular dates or a statement of availability. We’d set it to a price you offer today, because the ad policy bans unfounded pricing claims.

Set is_ads_eligible to true for every property you want to advertise. OpenAI notes that eligibility doesn’t guarantee delivery: properties must still pass review and serving requirements.

Images

The images cell holds a JSON array of up to 20 image objects. Each needs a url and can carry up to 20 tags. Invalid objects are dropped if at least one valid image remains, and a row with no valid image is rejected. Duplicate image addresses are removed.

We’d put your strongest exterior or room photo first, since the first valid image is the main one. OpenAI’s general creative guidance asks for simple, relevant images that match the message.

Guest ratings

Guest ratings are optional, and you can include up to 20 groups. Each group uses four numbered columns, starting at zero: rating_system, score, max_score and number_of_reviewers. Every field in a group must be filled, and a fully blank group is ignored.

The maximum score can be up to 100. A score of 8.8 out of 10 from 425 reviewers is written as four columns in the guest_rating[0] group.

Preparing the data

This is how we’d prepare a group with many properties. Most of the data already exists in a booking engine, a property management system or a channel manager. The work is mapping it to OpenAI’s columns.

Feed column	Usual source
hotel_id	Your internal property code, which should never change
name, description, brand	Your website or content system
url	The property’s page on your own site, not a third-party listing
images	Your photo library, on public addresses
location fields	Your property records, with coordinates checked on a map
guest ratings	The review systems you already display

Check the coordinates before upload. A pin in the wrong city sends the right ad to the wrong conversation.

What’s not supported in the beta

OpenAI lists what the beta leaves out. Columns outside the format aren’t read at all.

Not supported	What to know
Rooms, offers and dated stays	One row describes one property
Dynamic rates and date availability	base_price is static
Partial updates and the Delta API	Send a full snapshot every time
XML files	CSV or tab-delimited TXT only
Columns such as room_id, checkin_date, sale_price, tax or phone	They aren’t ingested
Top-level custom_label columns	Put labels inside ads_metadata instead

Checking the upload

Check Upload History after every snapshot. A completed transfer doesn’t mean every property was accepted. File transfer and processing are separate. Check Upload History in Ads Manager for processing results and row diagnostics, then fix rejected rows before the next full snapshot.

For SFTP, put the file directly in the root folder, not a subfolder. Changing SFTP credentials replaces the old method, so update any automated uploads when you do.

An upload checklist

• Every required field filled on every row, including all four required location columns.
• images, location.neighborhoods and ads_metadata written as valid JSON.
• Guest ratings as complete numbered groups, not a single JSON cell.
• Public landing pages and images, with the sample addresses replaced by your own.
• is_ads_eligible set to true for properties you want to advertise.
• base_price treated as a static starting price.
• hotel_id values unchanged since the last upload.

Where hotel feeds fit in a travel campaign

Hotel feeds suit groups with many properties; single hotels can use standard ads. That’s our view, and here’s why. Travel and experiences are among the consumer categories OpenAI says ads focus on in the early period. Hotel feeds suit groups and chains with many properties, where writing an ad for each one doesn’t scale.

Single properties can run standard ads now, with context hints that describe the stay. Our context hints guide explains how to write them. Our ad examples include a city break, and our context hint examples include a boutique hotel. Retail catalogs use the product feed instead. Once a feed is live, our reporting guide covers how to read product-level results.

Questions