Walmart's real-world experiment with OpenAI's Instant Checkout feature has produced one of the starkest data points yet on the limits of agentic commerce. Starting in November 2025, Walmart listed approximately 200,000 products available for direct purchase inside ChatGPT, allowing users to complete transactions without ever visiting Walmart's own website. The results were sharply negative: conversion rates for in-chat purchases came in at one-third the rate of click-out transactions, where users were redirected to Walmart's site to complete checkout. Daniel Danker, Walmart's EVP of Product and Design, described the in-chat experience as "unsatisfying" and confirmed the retailer is moving away from the model. OpenAI has since phased out Instant Checkout entirely in favor of app-based checkout handled by merchants directly.
The failure exposes a structural problem that has defeated multiple platform checkout experiments over the past two decades. E-commerce transactional flows have been refined across thirty years of consumer behavior data, and trust in a payment interface is not transferable — it is earned through repeated successful transactions and clear merchant accountability. Hacker News commenters provided concrete evidence of one specific technical failure: stale inventory data. Users reported being able to proceed all the way to the payment step on a Walmart item through ChatGPT before receiving an out-of-stock error — while Walmart's own website surfaced the out-of-stock status immediately. Real-time catalog normalization is a decade-scale infrastructure problem. Google built Merchant Center to address it; OpenAI has no equivalent foundation. This same failure mode previously plagued Google Shopping Actions and Facebook Shops native checkout, both of which similarly underestimated the infrastructure prerequisites for owning the transactional layer. The novelty of LLM-powered agents doesn't appear to change those economics. AI can drive discovery. Checkout still wants to happen where fulfillment lives.
Walmart's pivot is instructive for how the industry is likely to resolve the tension between AI discovery and merchant-controlled checkout. Rather than abandoning AI-assisted shopping entirely, Walmart plans to embed its own chatbot, Sparky, inside ChatGPT — keeping checkout within Walmart's own trusted infrastructure while still surfacing inside the AI interface. A similar integration with Google Gemini is planned for the following month. The architecture is structurally identical to the affiliate and click-out model that has powered comparison shopping engines since the early 2000s — the LLM wrapper is new, the underlying dynamic is not. The graveyard of platform checkout experiments — Google Checkout, Buy on Google, Facebook Shops native checkout, Twitter's Buy Now button, and now OpenAI Instant Checkout — suggests this is not a solvable product problem so much as a structural limit on how consumer trust can be delegated across organizational boundaries.