Walmart scrapped OpenAI’s ChatGPT checkout feature on March 20 after conversions crashed to one-third of their standard website performance. After making 200,000 products available through ChatGPT’s “Instant Checkout” last November, Walmart EVP Daniel Danker called the experience “unsatisfying.” The collapse marks the first major public test of “agentic commerce,” where AI agents shop on your behalf. The verdict: AI can help you find products, but it can’t replace the checkout experience.
Why the Single-Item Constraint Killed Conversions
OpenAI’s Instant Checkout forced single-item purchases – every product recommendation triggered its own transaction, shipping calculation, and delivery. Customers didn’t want five separate boxes for five products. They wanted traditional e-commerce: add multiple items to cart, review together, complete one purchase.
The technical gaps ran deeper. As of February 2026, OpenAI hadn’t built state sales tax collection – a basic e-commerce requirement. Wrong items appeared in carts. Payment providers raised fraud concerns. These aren’t bugs to patch. They’re infrastructure gaps exposing how premature this launch was.
Discovery Works, Transactions Don’t
OpenAI’s own data revealed the pattern: users researched products in ChatGPT but didn’t buy there. They returned to familiar platforms when ready to purchase. The discovery layer worked. The transaction layer failed.
Conversational interfaces excel at handling complex questions and surfacing results from massive catalogs. But checkout is different. People need visual context – images, specs, reviews displayed simultaneously. They want to manage carts, compare shipping options, review orders before confirming. Traditional e-commerce UX solves these problems. Conversational interfaces strip away the context users need when committing money.
Walmart’s Sparky Pivot Shows What Actually Works
Walmart already has a solution. Starting March 25, the retailer embeds its own chatbot, Sparky, inside ChatGPT and Google Gemini. Customers’ requests route to Sparky, which searches inventory, presents options, and processes orders. OpenAI gets a referral fee. Walmart owns the customer data, transaction, and relationship.
Sparky users spend 35% more per order than other shoppers, according to Walmart US CEO David Guggina. The hybrid approach works: AI handles discovery, then routes to merchant-controlled checkout. This respects both AI’s strengths and checkout’s requirements.
The Agentic Commerce Reality Check
Despite projections of a $30 billion market by 2028, current data shows the gap between hype and reality. Seventy-three percent of consumers use AI in shopping – for product ideas, review summaries, price comparisons. But only 13% completed a purchase after an AI referral. OpenAI promised integration with 1 million Shopify merchants. Roughly 12 actually integrated. OpenAI is phasing out Instant Checkout entirely.
What Developers Should Learn
The lesson cuts through the agentic commerce hype: AI works as an assistance layer, not a replacement for proven UX patterns. Discovery and transaction are different problems. Conversational interfaces reduce friction in search. Checkout requires visual context, trust signals, and familiar flows. Stripping these away doesn’t modernize commerce – it breaks the purchase experience.
The future Walmart demonstrates is realistic: AI as a discovery tool embedded in conversational platforms, routing to merchant-owned checkout when users are ready to buy. Hybrid approaches that respect what each interface does well will win. Pure conversational checkout that ignores decades of e-commerce UX learning will keep failing.
