OpenAI shipped a significant update to ChatGPT shopping ads on May 12: instead of building campaigns manually, retailers can now connect their product catalog and let the platform auto-generate ads from the data already in their feed. If you sell products online and you have a Google Shopping feed, that file is now potentially a ChatGPT ad campaign.
What changed on May 12
The first version of ChatGPT shopping ads (launched three weeks ago) required advertisers to set up campaigns manually — write copy, upload creative, manage bids. Self-serve, but still demanding active campaign work.
The new product feed format changes that entirely. You submit a structured product file (the same format you already maintain for Google Shopping or Meta catalog ads), set filters for which products are eligible, and OpenAI's system generates ads from your product names, attributes, and images. The platform supports up to one million SKUs per advertiser. Ads appear below ChatGPT responses, labeled as sponsored.
OpenAI is onboarding new partners with a 100-product sample before scaling to the full catalog. The company is working with ad tech partners Criteo and StackAdapt. If you're already running retargeting through Criteo, the pipeline to ChatGPT may already be closer than you expect.
Why this is different from the self-serve launch
Manual self-serve ads require marketing judgment: pick which products to feature, write copy, set bids, track performance. That format works for a controlled test of a new channel with a specific budget.
Product feed automation works more like Google Shopping or Meta's catalog ads. You define eligibility rules, the platform decides which products to surface for which conversational queries, and the creative is generated from your catalog data. You're not writing ads. You're supplying inventory.
That's a fundamentally different operating model. The entry barrier for scale drops. And it means the quality of your product data becomes the ceiling on ad performance — not the quality of your creative team.
Your feed quality is your creative quality
Here's the problem most e-commerce teams will run into: product catalogs built for Google Shopping aren't optimized for AI-generated copy inside a conversational interface.
A Google Shopping feed needs the right title structure, clean images, accurate pricing. But Shopping titles are written to match keyword intent: "Men's Running Shoes Nike Air Max 270 Black Size 10." ChatGPT is deciding whether your product is the right answer to a question like "what's a good running shoe for flat feet?" — and generating ad copy from whatever's in your feed.
If your descriptions are thin (SKU numbers and dimensions only), the AI has almost nothing to work with. If your title is variant-stuffed, the generated copy will read like a SKU list. If your images are pure white-background product shots without any context, you're starting behind in a format that appears alongside natural language responses.
The brands that will get the most from this are the ones whose catalog data reads like a product page written for a human: descriptive titles, benefit-forward descriptions, category-specific attributes, images that show the product being used.
The reach question
ChatGPT is not yet a shopping destination the way Google or Amazon are. But the intent signal inside ChatGPT is genuinely different — someone asking "what blender should I buy for making smoothies?" is explicitly in purchase mode. That's a different context than a Meta feed scroll or a Google search triggered by a broad match keyword.
OpenAI has not published click or conversion benchmarks for this format yet. Criteo and StackAdapt are in early testing. This is not a proven channel with established CPMs and ROAS norms.
If you're spending $10,000 a month on Google Shopping, you're not reallocating budget here yet. This is test-and-learn territory right now, not a primary performance driver.
Two scenarios where moving now makes sense
If your catalog is already in good shape and you're running through Criteo or StackAdapt, the integration cost is low. Request access, submit a 100-product sample, and see what the platform generates. You'll learn something about your feed quality at minimum.
If your product data is weak — thin descriptions, generic titles, missing attributes — don't connect a bad feed to a new AI engine and expect good results. Fix the data first, then request access.
For everyone else: watch the early case studies from Q2 2026. The brands in Criteo's beta will start publishing performance numbers by summer. That's when this moves from an interesting beta to a line item worth budgeting.
Before adding a new channel, it's worth knowing exactly what your current accounts are actually doing. Our free audit tool at gromerce.com/audit gives you a full account review in under five minutes.
Your product catalog was already your ad engine for Google and Meta. OpenAI just added a third platform to that pipeline.
Related articles: ai-traffic-ecommerce-adobe-2026 · chatgpt-ads-self-serve-what-it-means · pmax-hybrid-strategy-ecommerce-2026
Sources: Search Engine Land, Digiday, Modern Retail, OpenAI, May 2026
