On March 24, OpenAI announced its most ambitious shopping play yet: richer, more visual product discovery inside ChatGPT, powered by an expanded Agentic Commerce Protocol. Target, Sephora, Nordstrom, Best Buy, The Home Depot, Wayfair, and Lowe's are already on board. Shopify's entire merchant base is integrated by default. The press release reads like the future arriving on schedule. But buried in the same announcement is a quieter, more revealing story—one that suggests the entire industry may be thinking about AI-native commerce backwards.
Starting in November 2025, Walmart offered roughly 200,000 products through OpenAI's Instant Checkout feature, allowing users to complete purchases entirely inside ChatGPT without ever visiting Walmart.com. It was the showcase implementation of agentic commerce—the idea that AI assistants wouldn't just recommend products but close the sale.
It didn't work.
Daniel Danker, Walmart's EVP of AI Acceleration, Product, and Design, confirmed that in-chat purchases converted at one-third the rate of transactions where users clicked through to Walmart's website. He called the experience "unsatisfying."
That's diplomatic language for: we tested the vision everyone is excited about, and the data said no.
Embedded in OpenAI's own March 24 announcement is a line that most coverage will skip over: "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."
Read that again. OpenAI is walking back its checkout ambitions. Instant Checkout—the feature that was supposed to make ChatGPT a one-stop shop—is being phased out. Merchants are being told to handle their own checkout. The company that wants to mediate every interaction between humans and information is admitting it cannot mediate the moment someone pulls out a credit card.
This isn't a failure of technology. It's a revelation about user psychology.
The conversion gap makes perfect sense once you think about what happens in a shopper's brain during the final stages of a purchase. Discovery is exploratory, conversational, low-stakes. You're browsing. Comparing. Imagining. AI is exceptionally good at this. But checkout is a commitment. It requires trust—trust in the payment infrastructure, trust that the return policy will be honored, trust that the product arriving at your door matches what you saw.
Walmart has spent decades building that trust. Their website has saved payment methods, addresses, order history, loyalty programs, Walmart+ perks. A ChatGPT text box has none of that. When Instant Checkout asked users to enter credit card information into a chat interface, it was asking them to abandon a trusted context for an unfamiliar one. No loyalty points. No saved addresses. No order history to reference. Just a text conversation and a payment form.
Walmart's response is instructive. Rather than abandoning ChatGPT entirely, they're embedding their own experience inside it. A new in-ChatGPT app takes users from AI-powered discovery into a tailored Walmart environment that supports account linking, loyalty programs, and Walmart's own payment infrastructure. Their chatbot, Sparky, will live inside ChatGPT.
The architecture is significant: ChatGPT handles the top of funnel. Walmart handles the bottom. The AI figures out what you want. The retailer handles the transaction. Neither tries to do the other's job.
Danker framed the pivot this way on LinkedIn: "We've spent decades building great retail experiences. Now, we're bringing these experiences into entirely new places." A similar Walmart integration is reportedly coming to Google Gemini next month.
If you're running Google Ads, Shopping campaigns, or Performance Max, the Walmart-OpenAI story carries a clear signal: AI is restructuring the funnel, not replacing it.
OpenAI's ACP is a direct competitor to Google Shopping for the "what should I buy?" moment. When ChatGPT can show you visually browsable product grids, side-by-side comparisons, and conversational refinement, the traditional Google Shopping carousel looks like a relic. Target, Sephora, and Nordstrom aren't integrating with ACP for fun. They see the traffic shifting.
Walmart's 3x conversion gap proves that the last mile of commerce is not a technology problem—it's a trust problem. For advertisers, this means your landing page experience, your checkout flow, and your post-click infrastructure matter more than ever. AI can drive intent. It cannot close it. If your site can't convert the high-intent traffic that AI surfaces, you're paying for discovery you can't capitalize on.
The vision of AI agents autonomously browsing, comparing, and purchasing on behalf of users remains distant. If Walmart—with its massive brand recognition, competitive pricing, and sophisticated logistics—couldn't make AI-native checkout work, smaller merchants don't have a chance. The realistic near-term is AI-assisted discovery feeding into merchant-controlled checkout. That's less revolutionary than the pitch decks suggest, but it's what the data supports.
OpenAI's ACP requires merchants to share product feeds and promotions to appear in ChatGPT. Shopify merchants are already integrated via Shopify Catalog. This is the Shopping feed all over again—except now the feed powers AI recommendations, not just ad placements. If your product data is sparse, outdated, or poorly structured, you're invisible in the AI discovery layer. Treat your product feed with the same rigor you treat your ad copy.
Google should be paying close attention. OpenAI isn't trying to sell ads (yet). It's inserting itself into the consideration phase—the exact phase that Google's search ads have dominated for two decades. If a meaningful share of "what should I buy?" queries migrates from Google to ChatGPT, the impact on Shopping and Search revenue could be substantial.
But Walmart's data also suggests a ceiling. If AI assistants can't close transactions, they become referral engines. And Google is very good at being the place people go when they're ready to buy something specific. The threat to Google isn't that ChatGPT replaces it. The threat is that ChatGPT intercepts the undecided shopper before they ever reach Google.
That's a battle over the top of the funnel, not the bottom. And it's a battle Google is already fighting with AI Overviews, Shopping Graph, and its own conversational search experience.
The contours of AI-native commerce are becoming clearer, and they look different from the original pitch:
After managing $350M+ in Google Ads spend, here's my honest read on where this leaves us:
The hype about agentic commerce has been running ahead of the data. Walmart, one of the most sophisticated retailers on the planet, tested the thesis with 200,000 products and five months of data, and the answer was clear: users don't want to buy inside a chatbot. They want a chatbot to help them decide what to buy, and then they want to buy it from someone they trust.
For advertisers, the immediate playbook hasn't changed as dramatically as the conference keynotes suggest. Your landing pages still matter. Your checkout flow still matters. Your conversion tracking still matters. What is changing is where the intent forms. And that shift—from Google search to conversational AI—is real, gradual, and worth preparing for.
The advertisers who will navigate this well are the ones who obsess over two things: making their products discoverable in AI systems (through rich product data and feed optimization) and making their owned properties convert the traffic that AI delivers. The middle part—the checkout inside the chatbot—was a detour. Walmart just proved it.
Sources: OpenAI — Powering Product Discovery in ChatGPT (March 24, 2026) · Search Engine Land — Walmart: ChatGPT checkout converted 3x worse than website (March 19, 2026) · Daniel Danker, LinkedIn (March 24, 2026)
Want expert eyes on your account? Fill out the form and we'll send you a personalized audit with actionable recommendations.
We've received your request. Expect a personalized audit within 48 hours.