Introducing The OpenAI Product Feed Spec

Four weeks before answer engines freeze training data for Black Friday sales, OpenAI has released its Product Feed Specification the structured data model for merchants who want to sell products directly in ChatGPT with Instant Checkout. This is part of the larger infrastructure — Agentic Commerce Protocol — that will eventually power tomorrow’s agentic commerce, or agent-to-agent and consumer agent-to-business agent interactions.

Last week’s announcements are gaining momentum quickly — for example, Walmart announced a partnership with OpenAI to make its entire product catalog (except fresh food items) available for instant checkout in ChatGPT. Further, as announced at Dreamforce this week, Salesforce’s Agentforce Commerce will integrate with OpenAI, connecting orders, payments, fulfillment, and customer relationships to Instant Checkout.

Getting Started With OpenAI’s Product Feed Specification (The Easy Stuff)

Structured data managed in modern product information management solutions just became a requirement for brand manufacturers and retailers that want to surface and sell products on ChatGPT. To activate products on ChatGPT, merchants must be registered, unblock OpenAI’s crawlers, deliver the data feed in .CSV, .TSV, .XML, or .JSON formats, and send a full initial feed for OpenAI’s indexing team’s validation. Per agreed-upon industry standards for product information, the feed has required and optional data attributes, product variants and dimensions, rich-media assets, geotagging, and related products — no surprises here.

The AI Digital Shelf Is Born

Now, the fun stuff! Digital shelves that are surfaced on answer engines just became a lot more dynamic. OpenAI is introducing agentic commerce controls — two new flags at the product level (“enable_search” and “enable_checkout”). These allow merchants to toggle whether a product can be surfaced in ChatGPT search and whether to allow direct purchases for a product within ChatGPT. Depending on their dynamic commerce strategy, merchants might want consumers to transact within ChatGPT for one product family and instead be redirected to a brand, retailer, or distributor website for a different product family. The Product Feed Spec also introduces several foundational components to seed how AI shelves will operate:

  • Dynamic product recommendations. According to Forrester’s February 2025 Consumer Pulse Survey, 72% of US and 68% of UK online adults say it’s important for a retailer or brand to offer dynamic product images that can be zoomed, rotated, and viewed from all sides on their website. By supplying rich product data attributes through the feed and rich media, including 3D models, merchants can now use ChatGPT’s conversational interface to match user queries to products with a higher level of precision. Through the combination of the Product Feed Spec and its existing intent-based conversational interface, shopping is about to become a lot more dynamic. What is the fidelity of your product information to help you gain first-mover advantage over brands that will submit basic product feeds this holiday season?
  • Agentic time to market. OpenAI is accepting product data feed updates every 15 minutes from merchants, which will further raise the bar for consumer expectations and compress time to market for businesses selling products. Agentic time-to-market scenarios could include drop-shipped products with fluctuating inventories, event tickets with low inventory, or even daily deal promotions in competitive segments such as consumer electronics. Which products in your catalog could benefit from faster refresh rates, and how quickly can you get market insights to make those decisions?
  • A trusted agentic commerce community. Per Forrester’s March 2025 Consumer Pulse Survey, only 24% of US online adults trust AI agents to act on their behalf to make routine purchases. By requiring merchants to share product performance signals in the Product Feed Spec (e.g., return rates), OpenAI is connecting product data to brand perception and attempting to seed an agentic commerce community that is built on trust. How does your brand show up on answer engines today, and are you positioned to sell products on OpenAI this holiday season?

Transitioning From Traditional To Agentic Product Syndication? We Can Help!

Agentic commerce is quickly bringing us to another inflection point in digital commerce. In these moments, leaders must balance practicality and business exposure with market pressure to act. Syndicating product information and content for AI systems is different than traditional search engines. It requires clear and structured product relationships, comprehensive descriptions, up-to-date inventory, and corollary reviews for content authority. Please get in touch with us — Forrester’s commerce team — for an inquiry or guidance session to further design and syndicate your product information if and as you assess selling through agentic commerce this holiday season!