If you want an AI assistant to sell your product this quarter, the thing in your way is the product feed you have been deprioritizing for a decade. Screenshots surfaced this month of ChatGPT ads running full product carousels, several items at a time instead of one. Digiday reports that the format pulls product information directly out of retailer feeds, that OpenAI’s systems decide which products get grouped together, and that OpenAI has not announced any of it. The timing lines up with Q4.

Both protocols run on your product feed

OpenAI’s Agentic Commerce Protocol is closed and built around the transaction. Google’s Universal Commerce Protocol is an open standard it launched in January, co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and endorsed by more than twenty others including Stripe, Visa, Mastercard, Best Buy and The Home Depot. Gianluca Fiorelli and Alex Moss put the difference plainly in their conversation on agentic commerce: ACP is not platform agnostic, and UCP is trying to cover the whole journey from discovery to loyalty.

You do not have to pick a winner, because the entry requirement is the same either way. Google’s own UCP documentation says merchants integrate through Merchant Center: an active account, products eligible for checkout, and inventory Google can actually read. OpenAI’s carousels run off your product feed.

Google says product feeds work and llms.txt does not

Google’s AI optimization guide names Merchant Center feeds and Business Profiles as the levers for showing up in AI answers with products and local details. The same document says you can ignore llms.txt files and other special markup, chunking your content, rewriting text specifically for AI systems, and chasing inauthentic mentions. Google also links out to UCP from that page, so the agentic path and the “don’t bother” list are sitting in one guide, a few screens apart.

Consultants quote you the first half and skip the second. Cleaning up a feed is tedious and nobody can sell it as a methodology.

You have four things to fix

• The product data on your page matches the product data in your feed, including price and availability.
• Feed attributes are complete and correct: title, price, availability, GTIN, brand, condition, shipping.
• Product pages are server rendered, so the main content does not wait on JavaScript to exist.
• Merchant Center is in good standing with no open policy violations.

That is a week of unglamorous work for most catalogs, and it is closer to the technical 10% of your program than the 90% that decides whether anyone recommends you at all.

A clean feed will not get you recommended

None of this makes an assistant name you. That comes from the same place it always did: content good enough to end someone’s search, and enough coverage and mentions across the web that the platforms keep seeing your brand in the context you want. Skip that and clean feeds will get you nothing, because nothing will be pointed at your products.

The feed decides what happens after the recommendation. Winning the mention and then being unbuyable is a worse result than it sounds, because you paid for the demand and handed the checkout to whoever did the boring part.

For a decade the cost of skipping that boring part was some lost Shopping traffic. Now it is the checkout.