In short: OpenAI rolled back in-chat purchasing after Walmart found ChatGPT checkout converted at a third the rate of its own website. The AI is still deciding what your customer buys — it's just not the one collecting payment anymore. That split matters more for brands in Singapore and Malaysia than the checkout headline does.

Last month we wrote about ChatGPT's shopping research feature and how it was compressing the browsing-and-comparing journey into a single AI conversation. The natural next question was checkout — if the AI can research and recommend, why wouldn't it also complete the purchase? For a few months in late 2025, OpenAI tried exactly that. By early 2026, it had walked much of it back. What happened in between tells you more about where this is actually going than the original launch did.

What "Instant Checkout" Was, and Why It Didn't Stick

In September 2025, OpenAI launched Instant Checkout inside ChatGPT, built on the Agentic Commerce Protocol it developed with Stripe. Walmart was one of the retailers who went all in, listing around 200,000 products so shoppers could complete a purchase without ever leaving the chat window.

The results were the problem. Walmart's own EVP of product and design reported that purchases made inside ChatGPT converted at roughly a third the rate of clicking through to walmart.com — a gap large enough that Walmart called the experience "unsatisfying" and pulled back. The reason is structural, not cosmetic: checkout is where loyalty points, saved cards, bundle logic, delivery options, and return policies all live. Stripping that out into a generic AI-native checkout removed the exact things that make a shopper trust the final click.

OpenAI has since retired the in-chat purchase flow in favour of something closer to what actually works: discovery happens in the AI conversation, checkout happens on the merchant's own site or app.

The Model That's Actually Winning: Discovery In AI, Buy On Site

This is the detail brands should sit with. The AI didn't lose influence over the purchase — it lost the payment step. ChatGPT, Perplexity, and Google AI Overviews are still the layer where a growing share of shoppers decide what to buy and who to buy it from, before a single product page loads. What changed is where the transaction itself completes.

That's actually good news for brands with a well-built site. It means the advantage isn't in building your own AI checkout integration — it's in making sure your product content is what gets surfaced and trusted during the AI's research phase, so that when the AI hands the shopper back to a website, it's yours. This is the same ground we covered in our piece on ChatGPT's shopping research feature: the comparison already happened before the click. The click just needs somewhere good to land.

The Payment Networks Aren't Waiting Around

While OpenAI recalibrated the consumer-facing experience, the infrastructure for AI agents to pay on a customer's behalf kept moving. Mastercard and Visa both launched agentic payment protocols within a day of each other in April 2025 — Mastercard's Agent Pay and Visa's Intelligent Commerce, with Visa's Trusted Agent Protocol following that October. Mastercard completed its first live agentic payment transaction on 29 September 2025, and by June 2026 had extended Agent Pay to support machine-to-machine, always-on payments across cards, accounts, and stablecoins. Visa now has more than 100 partners building on Intelligent Commerce, with over 30 actively integrating in its sandbox.

Consumer comfort is catching up too: roughly 70% of shoppers say they're at least somewhat comfortable with an AI agent making a purchase on their behalf, though only about 36% are ready to hand over full control today. The infrastructure is being built for a future where agent-to-agent transactions are routine — B2B procurement agents negotiating directly with supplier agents is already being piloted. That future just arrived slower, and through a narrower door, than the initial checkout launches suggested.

What This Means for Brands in Singapore and Malaysia

None of this changes the fundamentals we've written about before — it sharpens them. The AI research-and-recommend phase is where the real battle for a customer's decision now happens, and that's exactly the layer most SG/MY brands still haven't optimised for.

Don't build for AI-native checkout yet. The infrastructure (ACP, Agent Pay, Trusted Agent Protocol) is real, but the consumer-facing version is still being figured out by companies with far bigger budgets than most SMEs in this region. Wait for the standard to settle.

Do build for AI-native discovery. Structured product content, clear specs and pricing, genuine comparison content — the GEO fundamentals — determine whether your brand is even in the conversation when an AI is doing the research on a shopper's behalf.

Make the handoff count. If the AI's research phase sends a pre-qualified, high-intent shopper to your site, your product page needs to close that visit, not lose it. Site speed, clear pricing, and trust signals matter more here, not less — this is the moment Walmart's checkout experiment showed AI can't yet handle for you.

Check where you currently stand. The YEA AI Site Audit scores exactly this — how your site reads to AI platforms doing the discovery and research that now precedes almost every purchase decision.

Conclusion

The lesson from Instant Checkout isn't that agentic commerce failed — it's that the industry moved fast, hit a real limit, and course-corrected within months. That's normal for infrastructure this new. What hasn't changed, and won't anytime soon, is that AI platforms are increasingly where the decision gets made, even if the transaction itself still happens on your website. Brands that treat the discovery layer as seriously as their storefront are the ones who'll be ready whenever agentic checkout does mature. If you want to know where your brand currently stands in that layer, YEA Business can help you find out.