2026 turned the ChatGPT shopping story into a roller coaster. Instant Checkout launched in September 2025 with Etsy sellers and a promise of buying inside the chat. By early March 2026, OpenAI pulled it due to:
- low user adoption,
- lack of real-time inventory synchronization,
- merchants’ unwilling to cede ownership of the transaction, and
- poor conversion metrics.
Fewer than 30 Shopify merchants ever went live. Suddenly the talk shifted to discovery, merchant apps, and sending people off to finish the buy somewhere else.
From my perspective working on feed formats and validation complexity daily at Feedonomics, one rule stays constant if you want to leverage ChatGPT’s shopping channel, where the LLM will converse with shoppers about a purchase decision, then refer them to the direct-to-consumer e-commerce site’s product detail page for add-to-cart and purchase: You still need a structured product feed that gets refreshed on a regular schedule. Without that feed your products stay invisible in ChatGPT shopping.
OpenAI’s own docs spell it out. Merchants need a secure CSV or JSON feed. It needs identifiers, descriptions, pricing, inventory levels, media links, and fulfillment options. The required fields decide whether price and availability show up correctly. Recommended ones like extra images or review counts help with ranking. You send a sample first so they can check the parsing. Then you keep sending daily snapshots.
Their Get Started - Agentic Commerce Protocol page lays out the order. Pick your delivery method: full file upload once a day or API updates throughout the day. Check every required field on every record, especially price and inventory availability to avoid oversell situations. Upload it. Then keep it current by overwriting the same file or pushing upserts. No shortcuts exist here. The feed comes first.
Stable product IDs. Current prices. Availability. Working image URLs. Brand names. Category paths. Shipping details where they apply. For data governance, flag any missing required field as a hard stop.
Then lock in the delivery. Most teams do a full daily snapshot plus API pushes for price and stock changes. Validate before every push. Watch for ingestion errors. Handle removals by dropping the record or flipping the eligibility flag.
Here’s where it gets complex fast. A set-it-and-forget-it feed or manual revision cycle won’t cut it if you want real revenue performance gains. You have to test title and description variations at scale. You need ongoing AI data enrichment that fills missing attributes and builds AEO question/answer pairs so the LLM surface can push shoppers further down the funnel. Specs change. Ranking signals shift. Doing all of this by hand on a large catalog turns into a full-time job that still lags behind.
And yes, it’s time for a shameless plug, only because I’ve seen how it delivers double-digit performance gains for our customers: Feedonomics.com runs a managed service that takes all that complexity off your plate, covering A/B testing, AI data enrichment including those AEO question/answer pairs, continuous data governance, and the daily optimization work that keeps the feed competitive. You can see how that played out in the Euro Car Parts Success Story.
Feed quality is the lever you control for visibility in ChatGPT shopping. And while the rest of the 2026 AI noise will keep shifting, the power of good product feed management will not.