Walmart let ChatGPT handle checkout. Conversion dropped 3x. Here's the funnel lesson.

For a few months in late 2025, Walmart let OpenAI process purchases directly inside ChatGPT — no click-through, no landing page, just a conversation that ended in a completed order. It converted at roughly a third of the rate of sending that same shopper to Walmart.com instead, and OpenAI quietly killed the feature five months after launch. At the same time, AI-referred traffic that does land on a merchant's own site is now converting better than traditional search traffic. The lesson isn't "ignore AI shopping." It's "stop trying to sell inside the chatbot, and put the CRO budget where the AI is actually sending people."

The Walmart experiment: what happened, and why OpenAI killed it five months later

Walmart made roughly 200,000 products available through OpenAI's Instant Checkout feature starting in November 2025, per Search Engine Land. OpenAI had launched Instant Checkout on September 29, 2025, and pulled it back on March 4, 2026 — a five-month run. Walmart's EVP of product and design, Daniel Danker, was quoted by MarTech saying purchases completed directly inside ChatGPT converted at roughly one-third the rate of transactions where users clicked through to Walmart.com instead.

OpenAI's own stated reason for pulling back, per The Keyword: "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we're allowing merchants to use their own checkout experiences while we focus our efforts on product discovery." That's a polite way of saying the in-chat purchase flow wasn't working well enough to keep pushing.

Why in-chat checkout failed: trust, tax, fraud and the missing "last look"

Forbes retail analyst Jason Goldberg's reporting on the retreat named the concrete technical gaps: no built-in sales-tax collection, no fraud-prevention mechanism, and real-time inventory sync problems once multiple merchants were involved at scale. But the deeper issue is behavioral, not just technical. As one analysis from Digital Applied put it, "a roughly 3x conversion penalty against a plain click-through is not a bug you patch, it is a signal about where trust, familiarity and clean data actually live." Buyers still want the "last look" — the actual product page, the actual checkout they recognize, the actual confirmation screen — before they hand over payment details, even when an AI assistant has already done the recommending.

The twist nobody expected: AI traffic now converts better than Google — on your own site

Here's the part that should reshape where CRO attention goes next. TechCrunch, citing Adobe's analysis of more than 1 trillion visits across 130+ top North American retailers, reported that AI-driven traffic to US retail sites rose 393% year-over-year in Q1 2026 (holiday season 2025 saw it up 693% YoY). More strikingly, that traffic converted 42% better than non-AI traffic in March 2026 — a complete reversal from March 2025, when AI-referred traffic converted 38% worse. AI-referred shoppers also spent 48% more time on product pages and viewed 13% more pages per visit than non-AI visitors.

Put the two data points together and the picture is coherent rather than contradictory: AI is bad at closing a sale inside its own chat window, but it's gotten good at sending qualified, higher-intent visitors to a merchant's actual site — visitors who then convert unusually well once they're there.

"Discover in AI, buy on site": the new division of labor

Post-pivot, ChatGPT now routes product recommendations to retailer pages via direct links instead of trying to process the purchase itself — AI owns discovery, retailers own the transaction, as Digital Applied's coverage of the shift frames it. Three open standards are emerging to formalize that handoff: UCP (Universal Commerce Protocol, backed by Google, Shopify and 20-plus partners, covering discovery through post-purchase), ACP (Agentic Commerce Protocol, from OpenAI and Stripe, using Stripe's Shared Payment Token so the agent never sees raw payment credentials), and AP2 (Agent Payments Protocol, initiated by Google and now governed by the FIDO Alliance, using cryptographically signed "mandates" to prove who authorized a purchase). None of these protocols try to keep the transaction inside the chat window — they're all built around a clean handoff to the merchant's own environment.

What this means for your landing page and checkout right now

If AI-referred visitors are arriving in greater numbers and converting better once they land, the highest-leverage CRO work isn't a chatbot integration project — it's making sure the page that receives that traffic doesn't waste the advantage. That means the same fundamentals matter more, not less: page speed on the specific product or landing page an AI agent links to, pricing and availability that's accurate and matches what the AI just told the shopper, and a checkout with as few extra steps as possible between arrival and purchase, since this visitor has already done most of their research inside the conversation.

This is core conversion optimization work, just pointed at a newer traffic source. If your landing pages and checkout flow already handle paid-traffic visitors well, they're probably most of the way to handling AI-referred visitors well too — but it's worth auditing the specific pages your product feed exposes to AI agents and shopping surfaces, since that's increasingly where the qualified traffic is arriving.

The protocols to watch, and how to get handoff-ready

• Keep product data clean and current — price, availability and specs need to match what an AI agent surfaced, or the visitor arrives already suspicious.
• Treat the AI-referral landing page as a first-class page, not an afterthought — fast load, clear value prop, obvious next step, same as your best paid-traffic landing pages.
• Watch UCP, ACP and AP2 adoption among the platforms and payment processors you already use — this is the plumbing that will eventually standardize agent-to-merchant handoffs.
• Don't build a custom in-chat checkout integration speculatively — the market leader tried it with 200,000 products and pulled back after five months. Let the standards mature first.
• Track AI-referral traffic as its own segment in analytics now, even informally, so you can see whether the 42% conversion lift Adobe reported shows up in your own numbers.

Frequently asked questions

Why did Walmart's ChatGPT checkout convert worse than its own website?

Walmart's EVP of product and design said purchases completed directly inside ChatGPT converted at roughly one-third the rate of transactions where users clicked through to Walmart.com. Reported technical gaps included no built-in sales-tax collection, no fraud-prevention mechanism, and real-time inventory sync problems across merchants at scale.

Is AI-referred traffic worth optimizing for if in-chat checkout failed?

Yes. Adobe's analysis of over 1 trillion visits found AI-referred traffic to US retail sites was up 393% year-over-year in Q1 2026 and converted 42% better than non-AI traffic once it landed on the retailer's own site, a full reversal from converting 38% worse a year earlier.

What is the new division of labor between AI chatbots and merchant websites?

Industry commentary describes it as "discover in AI, buy on site." ChatGPT now routes product recommendations to retailer pages via direct links rather than processing the purchase itself, meaning the AI drives qualified intent to your site but the actual conversion still happens on your landing page and checkout.

What should I actually do differently because of this shift?

Treat the landing page a chatbot link drops a visitor onto as a first-class conversion surface: fast load, clear pricing and availability, minimal friction between arrival and purchase, and a checkout that doesn't ask the visitor to re-establish trust from scratch. That handoff moment, not the chatbot conversation itself, is now the highest-leverage point in the funnel.

The takeaway

The most-watched AI commerce experiment of the last year failed at the one job it was built for — closing the sale — and succeeded, unintentionally, at proving something more useful: the handoff to a merchant's own site is still where trust and conversion actually happen. AI is getting very good at creating qualified intent. Your landing page and checkout are still what turns that intent into revenue, and right now that page is getting more traffic, and converting it better, than it has in years.