Walmart has ended its OpenAI Instant Checkout pilot and moved to embed its own shopping assistant, Sparky, inside ChatGPT and Google Gemini.
The US-based retailer began offering approximately 200.000 products through OpenAI's Instant Checkout from November 2025, allowing users to complete purchases without leaving ChatGPT. However, the experiment produced conversion rates that were roughly three times lower than those achieved when users clicked through to Walmart's own website. A senior company official described the in-chat purchasing experience as unsatisfying and confirmed the retailer is moving away from the model entirely.
The results point to a broader tension in agentic commerce: embedding transactional capability inside AI chat interfaces does not automatically replicate the performance of owned retail environments. For a retailer operating at Walmart's scale, even modest conversion gaps carry material implications for revenue.
Sparky fills the gap
Rather than abandon AI platform integrations, Walmart is repositioning its strategy. The company will embed Sparky, its proprietary shopping assistant, into ChatGPT and Google Gemini. Under this model, users authenticate with their Walmart account and synchronise shopping carts, with purchases completed within Walmart's own system rather than through the AI platform's native checkout layer.
Sparky has been deployed across Walmart's website and mobile applications since 2025. The ChatGPT integration was made available to paid subscribers in March 2026, with a rollout to free-tier users expected later in the spring. A corresponding integration with Google Gemini is planned for the following month. Walmart is also in discussions with Anthropic regarding a potential deployment on Claude.
The shift coincides with OpenAI's own decision to phase out Instant Checkout in favour of merchant-controlled, app-based checkout, a move that effectively transfers transactional responsibility back to retailers.
Implications for agentic commerce
The Walmart case illustrates an early and high-profile test of whether conversational AI can serve as an autonomous retail channel. The outcome suggests that, at present, the frictionless promise of in-chat checkout does not translate into purchasing behaviour at scale. Retailers that maintain control over the checkout flow, product data, and cart synchronisation appear better positioned to protect conversion performance.
For the broader payments and commerce ecosystem, the episode raises questions about the near-term role of agentic AI in retail and the extent to which AI platforms can act as credible intermediaries between merchants and consumers. It also reinforces the strategic value of proprietary retail infrastructure, particularly as major AI providers seek to position their platforms as commerce channels.
Whether Walmart's experience reflects a limitation specific to the Instant Checkout implementation or signals a wider structural challenge for in-chat commerce is likely to become clearer as other retailers, including those already preferring app-based models, publish their own results.
