ChatGPT Instant Checkout Is Gone: How ChatGPT Shopping Picks Products Now

In November 2025 Walmart put about 200,000 products into ChatGPT's Instant Checkout. A shopper could ask for a gift idea, pick one of the results and pay without leaving the conversation.

Four months later Daniel Danker, Walmart's executive vice president of product and design, described how it went. Purchases made inside ChatGPT converted at a third of the rate of shoppers who clicked through to Walmart's own site. He called the experience "unsatisfying".

On March 24, 2026 OpenAI retired Instant Checkout, saying its first version "did not offer the level of flexibility that we aspire to provide". ChatGPT kept the shopping and handed the checkout back to the stores.

For anyone selling online that is the more useful arrangement. The buyer still pays on your site, at your price, under your returns policy, as your customer. What moved is the moment of choice: more and more buyers decide what to buy inside a conversation, before they reach any store, and at that moment ChatGPT knows about your product only what your data tells it. Since July 2026, that data has mostly come from product feeds.

What happened to Instant Checkout

OpenAI launched Instant Checkout on September 29, 2025, with US Etsy sellers first and Shopify merchants to follow. It ran on the Agentic Commerce Protocol, built with Stripe, and it was the most visible early bet on agentic commerce. The merchant stayed the seller: ChatGPT passed a payment token, and the merchant charged it, shipped the order and handled returns as usual.

Six months in, the selection was still narrow, and the largest retailer in the program was measuring a threefold drop in conversion. The March update turned ChatGPT shopping into discovery: more visual product results, side-by-side comparisons, and the same protocol reused to bring merchants' product data in.

Retailers that still want the purchase to happen inside ChatGPT now build their own app there. Walmart's assistant runs in ChatGPT with the shopper's Walmart account, loyalty program and Walmart's own payments.

Look at the last two rows: the checkout left in March, and in July the feed became what decides.

How ChatGPT shopping works now

A shopper asks something that sounds like buying — running shoes for flat feet, a gift under fifty dollars, a replacement filter for a particular vacuum. ChatGPT answers with product results carrying images, prices and reviews, lets the shopper compare and narrow them down in conversation, and links to the stores that sell each item. After a click it can also show other merchants offering the same product, at the same price or a different one.

OpenAI's help center says those product results are selected independently, are not ads, and are not influenced by OpenAI's partnerships.

Advertising sits next to them now. Since early 2026 OpenAI has tested ads for logged-in adults on the Free and Go plans in the US, shown as labeled units below the answer, and it states that ads do not change what ChatGPT answers. So there are two ways into the conversation, and only one of them can be bought.

Why product feeds took over in July 2026

ChatGPT can learn about a product in two ways. It can search the web and read the pages it finds, or it can pull from structured product data that merchants send it. For most of its first year, the web did most of the work.

That changed in a single day. Profound, a company that tracks how AI assistants mention brands, followed about 1.75 million shopping prompts around July 10, 2026. Before that date, under a tenth of ChatGPT's product recommendations came from merchant feeds. Right after it, three in five did, and by early September the share was about two thirds.

Read the middle row first: the ten most-cited stores doubled their share in one day.

Recommendations concentrated at the same time. The ten most-cited stores went from about a fifth of all mentions to about two fifths, the number of different merchants cited fell by more than a fifth, and about two thirds of the brands Profound tracks lost at least a third of their shopping visibility.

Profound's numbers come from prompts its own monitoring runs, not from real shoppers, so they describe how ChatGPT answers rather than what people buy. The direction is still hard to misread. Stores that send a feed moved up, and stores that relied on being crawled lost ground overnight.

Is traffic from ChatGPT worth the work?

The buyers who arrive from AI assistants are unusually decided. Adobe, which measures visits to US retail sites, found that AI-referred traffic more than doubled year over year in May 2026 and converted about half as well again as traffic from other sources. Shopify reported that AI-driven traffic and orders to its merchants' stores tripled year over year in the second quarter of 2026.

That matches what Walmart saw from the other side. A conversation is good at producing a buyer who has already chosen. A store's own checkout — with the account, the saved card and the returns policy the buyer already knows — is good at finishing the sale.

What a small store can do about ChatGPT shopping

Get into the feed. On Shopify, the Agentic Storefronts channel in the admin connects your catalog to ChatGPT, Microsoft Copilot, Google's AI Mode and Gemini, with no fees beyond standard payment processing. Merchants on other platforms can share a product feed with OpenAI, or list their catalog for AI channels through Shopify's Agentic Plan.

Make each product answer the question a buyer would ask. A feed carries only what you put into it: the exact product type, brand and model, size, material, compatibility, the barcode number where one exists, and a title that says what the item is before anything else — the direction marketplace title rules are moving too. A buyer who asks for "a folding phone stand that charges" gets products whose data mentions charging.

Keep price and stock true. The feed is how ChatGPT shows your price and says whether the item is available. A buyer who clicks through to a different price or a sold-out page leaves, and ChatGPT has usually already shown them another merchant with the same product.

Let OpenAI's search crawler in. OpenAI's documentation says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers". Check your robots.txt and any bot protection your host switches on for you. Blocking GPTBot, which collects training data, is a separate choice and does not remove you from search.

Put shipping, returns and reviews in plain text. A comparison in a chat quotes what it can read: delivery time, return window, what reviewers praise and what they complain about. Terms that live in an image or load behind a script do not make it into the answer.

Ask your buyers' questions and watch who gets named. Put your category's real questions to ChatGPT every few weeks. When a competitor is recommended and you are not, their product data usually says something yours leaves out, and it is worth checking who the assistants cite in your category across the other engines too.

The last two are the kind of work every store agrees with and nobody does on a schedule. SellerClaw is a team of AI agents that connects to the stores you already sell on, plus your suppliers and ad accounts; you talk to it in a chat, and there is nothing to install. Ask it whether ChatGPT, Perplexity, Gemini and Claude recommend you for your buyers' questions, and it puts those questions to the engines and reports who they named instead. Ask it to fix what that exposed, and it rewrites thin titles and descriptions and fills missing product attributes across the stores you connected.

It does not send a product feed to OpenAI for you. That connection runs through your store platform or OpenAI's own program, and the quality of what travels through it is the part you can improve.

ChatGPT gave the checkout back and kept the decision. A store that wants to be part of that decision needs product data a machine can read and a price that is still true when the buyer clicks. Connect one store to SellerClaw and ask what ChatGPT recommends in your best-selling category — the first month of a plan costs about a dollar.