If you run an online store, ChatGPT product ads change the job. It goes from “write one ad for each item” to “upload your catalogue and let the system build the ads.” In early June 2026, OpenAI added a Feeds tool and a product feed campaign type to ChatGPT Ads Manager. Retailers can now connect a product feed and get an ad made for each eligible product. The ads show an image, title, price, rating and brand below ChatGPT’s answers.
In this guide I explain how the format works and what the feed has to contain. I also cover how it sits next to ChatGPT’s organic shopping results, and what I’d fix in your store before you spend anything.
What Changed in June 2026
ChatGPT ads have moved fast this year. Here’s the short timeline up to this week:
- February 2026: OpenAI began testing ads in the US for logged-in adults on the Free and Go plans. Plus, Pro and Enterprise users don’t see ads.
- Early May 2026: the self-serve Ads Manager beta opened to US businesses, with cost-per-click bidding and no minimum spend. We covered that launch in our guide to ChatGPT’s self-serve Ads Manager, linked at the end of this article.
- 2 June 2026: advertisers spotted a new Feeds section in Ads Manager, with tabs for connecting a feed and viewing upload history.
- From 5 June 2026: conversion-optimised bidding began rolling out to accounts that already send at least one conversion event.
- 6 June 2026: ChatGPT ads went live for users in the UK. OpenAI says more countries are coming in the following weeks.
- This week: OpenAI started emailing advertisers about the product feed beta. It also published a help article on creating campaigns from product feeds.
Two warnings matter. First, this is a beta, so the screens and rules will change. Second, UK users seeing ads doesn’t mean every UK business can buy them yet. Self-serve buying started with US businesses. Check what your own Ads Manager account can target before you plan a UK campaign. See also our guide to ChatGPT ads in Europe.
How ChatGPT Product Ads Work
The format will feel familiar if you’ve run Google Shopping or Performance Max. You don’t write keywords or build an ad for every product. You give structured product data, and the system decides which product to show for the conversation in front of it.
From Feed to Ad
OpenAI’s help documentation describes these steps:
- You create a feed under Tools > Feeds in Ads Manager.
- You import products. You can upload a CSV or TXT file, point to a hosted HTTPS URL, or connect over SFTP for automatic uploads.
- OpenAI processes the file. Depending on its size, this takes from a few minutes to a few hours. Errors show under Upload History.
- You create a campaign and choose Product feed as the campaign type.
- Inside an ad group, you pick the feed and use filters to decide which products are included.
- You set up an ad template, preview it, and submit the campaign for review.
The ad itself is built from your feed. OpenAI lists product images, titles, star ratings, prices, sale prices and brand as things a product ad can show. Right now, the template lets you choose which image field to use. It takes titles and descriptions straight from the feed.
Where the Ads Appear
Ads appear below ChatGPT’s response and are labelled as sponsored. OpenAI has said ads don’t change what ChatGPT says in its answer. The ad sits next to the answer, not inside it.
That has a practical result. You can’t buy your way into the recommendation itself. Say someone asks for “the best trail running shoes for wide feet.” Your ad might show below the answer. But the answer is still built from what ChatGPT finds on the web.
Targeting Is Based on the Conversation
There are no keyword lists to build. The system matches products to what the conversation is about, much as Google’s automated shopping campaigns match products to searches. That’s why your feed data matters so much. Titles, descriptions and attributes are the main signals the system can use to decide if your product fits.
Scale and Onboarding
Reports from the beta say advertisers can run up to one million SKUs (individual products or variants). New e-commerce partners first send a sample of around 100 products. The full catalogue only goes in once that sample passes. If your sample fails, you fix the data before you can run feed ads. So a clean sample is the first gate, and it’s where I’d put my effort first.
Feed Requirements as Documented
OpenAI points advertisers to a feed specification in its commerce documentation. If you already send a feed to Google Merchant Center, much of the work is done, because the core fields overlap. You’ll still need to map field names and check formats. Don’t assume a Google file will upload cleanly.
The Core Fields
For product discovery, the specification treats these as the basics:
- A stable, unique item ID for each product or variant. Don’t change IDs between uploads, or reporting and history break.
- Title and description in plain text. Keep titles short and descriptive, and keep descriptions free of HTML.
- Product URL that lands on the exact product page.
- Brand and seller name.
- Main image URL: a direct, public image file such as JPEG or PNG.
- Price with its currency, and availability using the allowed values (in stock, out of stock, pre-order and so on).
Variants need grouping fields, so that sizes and colours sit under one parent product. Optional fields add detail. These include GTIN (barcode number), sale price, colour, size, material, condition, review count and star rating. I’d always include ratings and sale prices, because both can appear in the ad itself.
Ads Eligibility and Custom Labels
The feed has an ads eligibility setting for each product. It lets you keep some items out of ads without deleting them from the file. Check how your feed handles products where the value is left blank. The defaults are documented, so don’t guess.
There’s also an optional ads metadata field for your own grouping values, such as a margin tier or product line. These work like Google’s custom labels. They let you filter ad groups by your own business logic, not only by category or brand.
Format Details That Trip People Up
- Price format: the specification expects a plain amount with a three-letter ISO currency code (for example,
79.99 USD). Don’t use currency symbols or thousands separators. - Sale price: it must be lower than the regular price and in the same currency, or it’s ignored.
- GTIN: use real barcodes only and keep leading zeros. Leave the field empty if a product has no GTIN. Never make one up.
- Availability: a missing or unknown value can get the row rejected.
- Freshness: update the feed whenever prices or stock change. For any catalogue that changes weekly, a hosted URL or scheduled SFTP upload is safer than manual CSV uploads.
The full specification is longer than this, and it’s a beta document. Before you build your export, read OpenAI’s own guide to creating campaigns from product feeds.
How Paid Product Ads Relate to Organic Shopping in ChatGPT
This is where I see most store owners get confused. So let’s separate the two clearly.
Organic Shopping Results
ChatGPT has shown product results in its answers since 2025. These are product cards with images, prices and links to merchants. OpenAI has said these results aren’t ads. The products are chosen from structured product data, merchant information, prices, descriptions and reviews gathered from across the web. They also come from merchants who share product data with OpenAI directly.
To be considered organically, your product pages need to be reachable by OpenAI’s search crawler. Your product data also has to be clear and consistent. If you’ve blocked OpenAI’s crawlers in robots.txt by mistake, fix that first. Our guide to robots.txt and AI crawlers explains which bots do what.
Paid Product Feed Ads
The product feed you upload in Ads Manager powers the sponsored spots below the answer. OpenAI’s beta documentation says products uploaded this way are eligible for ads only. They won’t appear in organic ChatGPT conversations, though it suggests this may change.
In other words, uploading your catalogue to Ads Manager won’t get you into the organic product cards today. The Ads Manager screens do include a setting for organic product listings. That shows where OpenAI is likely heading: one feed doing both jobs. For now, treat them as two separate tracks.
Why the Two Still Overlap in Practice
Even if the pipelines are separate, the inputs are the same. You need accurate titles, clear attributes, real reviews and product pages that match the feed. A store that has fixed its data for paid ads is in a far better position organically. The reverse is also true.
What E-commerce Stores Should Prepare
Most of the work here isn’t in Ads Manager. It’s in your catalogue and on your product pages.
1. Clean the Feed Before You Upload It
Questions in a chat are long and specific. Someone might ask for “a waterproof hiking jacket under £150 that packs small.” A title like “Men’s Jacket – Blue” gives the system nothing to match. Weak titles like this are the most common feed problem I come across.
- Write titles with brand, product type and the details people ask about: “Brand Men’s Packable Waterproof Hiking Jacket, Navy, Medium”.
- Fill in colour, size, material, gender and age group wherever they apply.
- Remove HTML, sales text and ALL CAPS from descriptions. Describe what the product is, who it’s for and how it’s different.
- Remove discontinued products instead of leaving them as out of stock forever.
2. Make Product Pages Match the Feed
The ad links to your product URL. A mismatch costs you twice: an annoyed shopper and a wasted click. Check that the price, sale price, stock status and variant on the landing page match the feed. Make sure the URL opens on the right variant. It shouldn’t be a general page where the shopper has to pick the size and colour again.
Pages also need to load quickly and show their key content without relying on scripts that might fail. Shoppers coming from a chat have already had a short answer. They won’t wait for a slow page.
3. Get Reviews Into the Data
Star ratings can appear in the ad, and OpenAI names reviews as an input to organic shopping results. If your reviews live only inside a widget that isn’t in your feed or your markup, both systems can’t see them.
- Include review count and average rating in the feed, for each product, not for the whole store.
- Only send ratings that have real reviews behind them. A 5.0 from a single review helps nobody.
- Keep collecting reviews after purchase. The number of reviews matters as much as the average.
4. Add and Test Product Schema
Product structured data won’t place your ads. But it helps every system that reads your pages understand them, including AI crawlers. Mark up the name, brand, image, GTIN, offers (price, currency, availability) and aggregate rating. Make sure the markup matches both the visible page and your feed. Our schema markup guide walks through Product and Offer markup in detail.
5. Install Conversion Tracking First
I wouldn’t launch feed campaigns without measurement in place. ChatGPT Ads supports a JavaScript pixel and a server-side Conversions API. The newer conversion-optimised bidding needs at least one conversion event already coming in. Running both the pixel and the API gives you the most complete data. Deduplicate them (remove double counts) with a shared order ID.
How to Measure ChatGPT Product Ads
Measuring a new channel takes discipline. The numbers in Ads Manager are a starting point, not the whole picture.
Inside Ads Manager
Product feed campaigns have a Products tab that reports results product by product. It’s a reporting view, so products only show up after a campaign has delivered impressions. OpenAI notes that reporting can take a few hours to appear once delivery starts. An empty tab on day one doesn’t mean your upload failed.
Reporting is combined: impressions, clicks, spend and conversions, with no user-level data. Use it to spot which products and product groups earn clicks, and which get impressions but no engagement. In my experience, the second group usually points to a weak title, image or price.
In Your Own Analytics
Add UTM parameters to your landing page URLs, or campaign-level tracking where Ads Manager supports it. That keeps paid ChatGPT clicks apart from organic ChatGPT referrals. If you skip this, both blur together under the same referrer. Our guide to tracking ChatGPT and AI visits in GA4 shows how to set up a channel group for AI referrals. You can add a separate paid ChatGPT rule next to it.
Mark purchases and add-to-basket events as key events in GA4. Then you can compare ChatGPT traffic with Google Shopping and paid social on the same terms.
Metrics Worth Watching
- Click-through rate by product group: shows which parts of the catalogue suit discovery in chat.
- Cost per click versus your Shopping CPC: a quick check on whether the channel is priced sensibly for you.
- Conversion rate and revenue per click: the real test. Clicks from a chat may be better or worse than search clicks. Only your own data will tell you.
- Landing page bounce and price mismatches: if people land and leave, check that the page matches the ad.
- Organic ChatGPT referrals over the same period: to check that paid isn’t just replacing visits you’d have got anyway.
Give a test enough time and budget to reach meaningful numbers before you judge it. Keep the product set small at first, so you can tell what’s working.
A Sensible Way to Start
I wouldn’t upload your whole catalogue on day one. Start with a few dozen to a few hundred products that have strong margins, healthy stock, good images and real reviews. Products people tend to research before buying are natural picks. Think outdoor gear, home appliances and skincare, where shoppers already ask ChatGPT for comparisons.
Use the ads metadata field to tag products by margin or priority. Then you can split ad groups and bids without rebuilding the feed. Expand once your measurement works and the first products show what sells.
Conclusion
ChatGPT product ads give retailers a Shopping-style format inside AI chats, powered by a feed most stores already have in some form. The ads sit below answers, not inside them. For now, the Ads Manager feed is separate from ChatGPT’s organic product results. Both still reward the same things: complete, accurate product data, product pages that match it, real reviews and clean markup.
My suggested next step for ChatGPT product ads: check your current Merchant Center feed against OpenAI’s specification. Fix titles and attributes for your top products. Get the pixel or Conversions API sending purchase events. If you haven’t set up an Ads Manager account yet, start with our walkthrough of how ChatGPT’s self-serve ads work.
