- ChatGPT now processes 50 million product queries daily, making it one of the web’s most active research tools despite struggling to convert searches into purchases.
- Major U.S. retailers including Target, Walmart, and Sephora have integrated with ACP to surface products inside ChatGPT without paying for placement.
- A HubSpot survey found 70% of users who tried ChatGPT for shopping recommendations prefer it over traditional search engines like Google.
ChatGPT is overhauling how its users find and compare products. OpenAI this week rolled out a revamped shopping experience that brings visual product browsing, image-based search, and side-by-side comparison to the chatbot — while quietly retiring its earlier attempt to let users complete purchases without ever leaving the app.
The move reflects a more honest read of how people actually use ChatGPT when shopping. According to CNBC, the platform now handles roughly 50 million product-related queries every day, making it one of the most active product research tools on the web. But converting that research intent into completed purchases inside the chatbot turned out to be a harder problem than OpenAI initially anticipated.
OpenAI’s Instant Checkout Experiment Fell Short — Here’s What Replaced It
The company acknowledged that Instant Checkout, its previous feature that allowed users to buy select items without leaving the conversation, fell short of expectations. In OpenAI’s own words, the initial version “did not offer the level of flexibility” the company had set out to provide. Rather than continuing to iterate on a checkout layer inside ChatGPT, OpenAI decided to hand that responsibility back to the merchants themselves, focusing instead on making the research phase as useful as possible.
The new experience is powered by the Agentic Commerce Protocol (ACP), a framework co-developed with Stripe and released as open source. Through ACP, users can browse products visually, upload a photo to find similar items, and refine results through back-and-forth conversation — all without leaving the chat window. Once a user is ready to buy, they’re routed to the merchant’s own checkout flow.
The gap between research and purchase was always going to be the central challenge. ChatGPT’s 900 million weekly users account for 97% of all AI-driven product research traffic across the web — yet in-app conversions weren’t materializing at the scale OpenAI had projected. The pivot essentially accepts that reality and builds around it rather than against it.
Major Retailers Are Betting on ChatGPT as a Discovery Layer, Not a Storefront
Several of the largest names in U.S. retail have already integrated with ACP for the discovery layer. Target, Sephora, Nordstrom, Lowe’s, Best Buy, Home Depot, and Wayfair are all on board, using the protocol to surface their products inside ChatGPT search results. Walmart is taking a step further: the retailer is launching a dedicated in-ChatGPT app that connects the research experience directly to a tailored Walmart checkout, available now on web browsers with mobile apps to follow.
One aspect of ChatGPT’s shopping results that sets it apart from traditional search engines like Google Shopping is how products are ranked. OpenAI states that results are organic and unsponsored — meaning no retailer can pay to appear higher in the feed. Rankings are based on relevance alone, though roughly 75% of the product data ChatGPT surfaces still comes from Google Shopping, routed through encoded queries behind the scenes.
Consumer interest in using AI for shopping decisions appears genuine and growing. A HubSpot survey found that 55% of consumers have already turned to ChatGPT for product recommendations, and 70% of that group said they prefer it over traditional search. The new features roll out this week across all user tiers — free, Go, Plus, and Pro — meaning the full base of ChatGPT’s 900 million weekly users will have access. Whether that preference eventually translates into a checkout model that works at scale inside the platform remains the open question OpenAI is still working to answer.