Google Shopping, ChatGPT Checkout, and the New Product-Feed SEO

OpenAI built a way to buy things directly inside ChatGPT, launched it with real retail partners, and within six months was already reworking it after adoption came in weaker than expected. That’s not a failure story exactly — it’s a genuinely useful data point about where AI shopping actually is right now, as opposed to where the announcements made it sound like it was headed.

What actually launched, and what happened next

OpenAI launched Instant Checkout in ChatGPT on September 29, 2025, starting with Etsy sellers and expanding to Shopify merchants, built on an open-sourced standard OpenAI calls the Agentic Commerce Protocol. The pitch was straightforward: find a product through conversation with ChatGPT, and complete the purchase without leaving the chat.

By March 2026, CNBC reported OpenAI had begun reworking Instant Checkout after weaker-than-hoped adoption. One specific data point from that reporting: Walmart said purchases completed directly within the chat interface converted at roughly one-third the rate of simply clicking through to Walmart.com and buying there normally. That’s a meaningful signal — it suggests that even when the friction of leaving a chat is removed, buyers still convert better once they’re actually on a real, familiar storefront they trust, at least for now.

Removing a step in the purchase funnel doesn’t automatically improve conversion if the step you removed was also providing reassurance. In-chat checkout data so far suggests that trade-off hasn’t been fully solved yet.

Google’s parallel, and more established, push

Google announced expanded AI Mode shopping features at I/O in May 2025, including virtual try-on and agentic checkout and price tracking, built on Gemini combined with Google’s existing Shopping Graph — described as covering more than 50 billion product listings. Virtual try-on, which lets someone upload a photo and see how a specific clothing item would look on them, launched with real retail partners including Anthropocene, Everlane, H&M, and LOFT.

Google’s approach has a structural advantage OpenAI’s doesn’t: the Shopping Graph and Merchant Center feed infrastructure already existed, built over years of standard Google Shopping operation. AI Mode shopping features are layered on top of an already-mature product data pipeline, rather than being built from scratch alongside a brand-new checkout mechanism.

What this means for product feed requirements

Here’s where I want to be careful about the limits of what’s actually confirmed: there is no separate, officially documented “AI shopping schema” distinct from what Merchant Center already requires. Standard product feed fundamentals — accurate GTIN, current price, real-time availability — combined with Schema.org Product and Offer JSON-LD markup as a secondary verification signal, remain the practical foundation. I’d flag this specifically as informed industry practice based on how these systems appear to work, not a directly confirmed official Google requirements page naming AI Mode specifically — I couldn’t verify a distinct AI-specific schema requirement through primary Google documentation while researching this.

What that means practically: if your Merchant Center feed and product schema are already solid — accurate pricing that matches what’s on the actual page, real-time stock status, complete GTIN data — you’re already positioned for these AI shopping surfaces as well as you can be without a distinct, separate implementation. There isn’t a parallel checklist to build.

Why the trust gap is probably the real bottleneck

The Walmart conversion data is a useful clue for thinking about why in-chat checkout hasn’t taken off as fast as expected. Buying something typically involves a series of small trust checks a shopper runs almost unconsciously — recognizing the retailer’s checkout page, seeing a familiar payment flow, having a saved address auto-populate correctly. An in-chat purchase flow, however smooth technically, skips past visual cues a shopper has spent years learning to trust. That’s not necessarily a permanent problem — trust in new purchase flows has shifted before, mobile checkout itself faced similar skepticism a decade ago — but it suggests the gap is more about behavioral trust-building than a technical limitation OpenAI or Google can simply engineer away in a single update.

What’s actually worth doing right now

• Don’t over-invest in a separate “AI checkout” integration yet if you’re a smaller merchant — the early adoption data (at least from Walmart’s experience with OpenAI’s implementation) suggests this channel is still finding its footing, not a proven high-conversion surface to prioritize engineering effort toward.
• Do keep your core Merchant Center feed genuinely accurate and current. This is the actual foundation both Google’s more mature AI Mode shopping and any future OpenAI commerce iteration will depend on — it’s not wasted effort even if the AI-specific surface changes shape.
• Watch Google’s AI Mode shopping features more closely than OpenAI’s checkout, at least for now — Google’s version is built on a much more established data and merchant infrastructure, and the early evidence suggests that maturity matters for adoption.
• Reassess in another six to twelve months rather than assuming either direction is settled. OpenAI’s own reworking of Instant Checkout within six months of launch is a clear signal that this space is still actively being figured out, not a stable target to optimize for definitively today.

The honest state of AI shopping right now is that it’s genuinely being built, genuinely has real retail partners and real transaction volume, and is also genuinely still working out basic questions like why in-chat conversion trails simple click-through. That’s a normal place for a new commerce channel to be — but it’s a different picture than the “AI agents are about to do all your shopping” framing that accompanied both companies’ initial announcements.