Status, checked 5 October 2026: OpenAI scaled Instant Checkout back on 24 March 2026. Merchants now use their own checkout (on their site, through an in-app browser or in their own ChatGPT app), and the Agentic Commerce Protocol now mainly powers product discovery feeds.
Quick Answer: ChatGPT Instant Checkout let shoppers buy inside the chat from 29 September 2025 until OpenAI scaled it back on 24 March 2026. Today the sale closes on the merchant's own checkout, and the Agentic Commerce Protocol carries your product feed into ChatGPT's shopping results instead.
Instant Checkout is the part of ChatGPT shopping that most brand teams heard of and few actually used. The launch post named the merchants, the protocol and the payment partner; six months later OpenAI said the first version "did not offer the level of flexibility that we aspire to provide" and moved its effort to product discovery.
This page covers what changed, what the protocol still does, and the feed contract that now decides whether your products appear at all. The crawler-and-schema route is covered in how products get into ChatGPT shopping, so it is not repeated here.
What is the status of ChatGPT Instant Checkout now
Instant Checkout is no longer OpenAI's focus. In its 24 March 2026 product discovery post, OpenAI said it is "allowing merchants to use their own checkout experiences while we focus our efforts on product discovery", and CNBC had reported the pivot on 20 March.
| Question | Status as of 5 October 2026 |
|---|---|
| In-chat checkout | Scaled back on 24 March 2026; no longer the default path to purchase |
| Where the sale closes | The merchant's own checkout: its website, an in-app browser, or its own ChatGPT app |
| What ACP does now | Carries product feeds and promotions so catalogues are "fully represented" in ChatGPT's shopping results |
| Shopify merchants | Product data flows in through Shopify Catalog with no extra work from the merchant, per OpenAI |
| Retailers named on ACP for discovery | Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot and Wayfair |
| Native retailer experiences | ChatGPT apps, such as Walmart's in-ChatGPT app with account linking, loyalty and Walmart payments |
| Who sees the new shopping results | Free, Go, Plus and Pro users, from the March rollout |
What this means for a brand
The work moved from enabling a Buy button to being represented accurately in discovery. A clean feed, current prices and stock, and pages that win the shortlist now matter more than any checkout integration.
The rest of this page keeps the launch-era detail because the protocol, the feed fields and the tie-break rules still describe how ChatGPT reads your catalogue. Read checkout-specific parts as history.
What Instant Checkout changed for brands at launch
At launch, the brand stayed merchant of record, paid a fee on completed orders, and got no ranking benefit for enabling checkout. Those three facts are stated in OpenAI's launch post, and they decide how much of your effort belongs in the feed versus the content.
Orders, payments and fulfilment ran on the merchant's existing systems. ChatGPT passed the order to your backend, you accepted or declined it, and you kept returns, support and the customer relationship.
At a glance
| Question | What OpenAI's launch post and docs said |
|---|---|
| Launch date | 29 September 2025, US ChatGPT Plus, Pro and Free users |
| Merchants at launch | US Etsy sellers live; over a million Shopify merchants "coming soon", with Glossier, SKIMS, Spanx and Vuori named |
| Scope at launch | Single-item purchases; multi-item carts and more regions promised next |
| Payment partner | Stripe co-developed the protocol; other processors join through a Shared Payment Token or the Delegated Payments spec |
| Fee | A "small fee" on completed purchases; free for users, no effect on prices |
| Effect on ranking | None. Product results are organic and unsponsored, and Instant Checkout items "are not preferred" |
| Feed access | Approved partners only, via the merchant application form |
Did checkout change which products ChatGPT shows
No, with one narrow exception. OpenAI stated that product results are ranked on relevance to the user and that Instant Checkout items were not preferred in those results.
The exception is a tie between merchants selling the same product. There, ChatGPT weighs availability, price, quality, whether the merchant is the primary seller, and whether Instant Checkout is enabled. If you are the brand rather than a reseller, being the primary seller already counts in your favour; checkout is a tie-breaker among resellers.
What is the Agentic Commerce Protocol
The Agentic Commerce Protocol (ACP) is an open standard, co-developed by OpenAI and Stripe, for sharing catalogue data and orders between merchants and AI agents. It launched as the way ChatGPT sent an order to a merchant's backend, and since March 2026 OpenAI has extended it to carry feeds and promotions for discovery.
The developer docs frame it more narrowly for brands: ACP is "the connective layer between merchants and ChatGPT users" that lets ChatGPT "ingest structured catalog data, understand merchant inventory, and surface relevant products in context", per OpenAI's commerce documentation. Discovery comes first; checkout is a separately enabled integration on top.
The three parts a brand should know
- Product feed. A structured catalogue you push to OpenAI, either as a daily file upload over SFTP or through the API. This is the part every brand touches.
- Checkout. When a shopper tapped Buy, ChatGPT sent the order details to your backend to accept or decline, charge and fulfil. The spec is still published, but OpenAI now steers merchants to their own checkout.
- Delegated payment. A Stripe merchant can enable agentic payments "in as little as one line of code"; a merchant on another processor uses Stripe's Shared Payment Token API or the Delegated Payments spec, without switching processor.
Who was named, and who was not
OpenAI's post names Etsy, Shopify, Stripe, Glossier, SKIMS, Spanx and Vuori. It does not name a payment network or any retailer outside those. For a D2C brand on Shopify the platform is still the integration path: OpenAI's March post says Shopify product data reaches ChatGPT through Shopify Catalog without extra work from the merchant.
What a product feed must carry to be eligible
A useful discovery feed needs nine required fields on every row, and a missing or malformed value rejects the row. The list is in OpenAI's file-upload products spec: item_id, title, description, url, brand, seller_name, image_url, availability and price.
The format rules are stricter than most Shopify or Merchant Center exports. Price is written as 79.99 USD, a decimal amount, a space and an ISO 4217 code; availability must be exactly in_stock, out_of_stock, pre_order, backorder or unknown; titles are kept to 150 characters and descriptions to 5,000, in plain text.
The fields that decide the product card
| Field | Rule in the spec | Why it matters to a brand |
|---|---|---|
gtin or mpn |
GTIN of 8, 12, 13 or 14 digits with a valid check digit; MPN only with brand |
Matches your listing to third-party price and review data and dedupes you against resellers |
group_id + variant_dict |
One row per variant, same group_id, listing_has_variations=true |
Size and colour become selectable in chat instead of separate products |
sale_price |
Positive, strictly below price, same currency |
A sale that is not in the feed is not a sale ChatGPT knows about |
review_count + star_rating |
Product reviews only, same population, rating to two decimals | Feeds the likes-and-dislikes summary in the card |
is_eligible_search |
Defaults to true; false also disables checkout |
Your kill switch per item |
is_eligible_checkout + seller_privacy_policy + seller_tos |
All three needed to opt into checkout | The Buy button does not appear without public policy URLs |
How the snapshot pipeline behaves
OpenAI treats each upload as a full snapshot, recommends at least a daily push, and matches products by item_id rather than filename, per the file-upload overview. A product missing from a snapshot is not deleted immediately; its last record is retained for up to 14 days, so a broken export does not wipe you out overnight.
The practical reading is that price and stock in the feed are the price and stock ChatGPT will show. Sale-window dates and availability dates in the feed do not schedule anything; you update the current value when it changes, or the card is wrong.
What changes for a D2C brand's content
The feed decides whether your product can appear; your content still decides whether your brand is the one the answer recommends. Instant Checkout adds a Buy button to a product card, but the card only shows up after ChatGPT has decided which products answer the prompt, and that decision leans on the same category pages, comparisons and reviews that D2C brands already compete on in AI shopping.
Three things do change. Descriptions are read by a model, not a shopper, so OpenAI's best-practices guidance asks for "concise, factual copy" and tells you to omit optional fields that need brittle transforms.
Variants are modelled at row level, so a size guide buried in an image is invisible. And the product URL should carry a feed-specific tracking parameter such as utm_medium=feed, or you will never separate ChatGPT orders from the rest.
Product pages still have to earn the citation
The product card is not the only place a brand appears. When a shopper asks "best running shoes under $100", ChatGPT builds the shortlist "from across the web" before it shows any card, so the product schema on your page and the pages that get your brand cited by ChatGPT for the category are still the front door.
Write the buying-decision content as if the feed did not exist: the specification table, the comparison against the two rivals the model already names, the answer-first FAQ on fit and returns. That is the structure answer engines extract, and it is what puts a brand into the shortlist that the card then closes.
How to check whether your products surface in ChatGPT
Run your twenty highest-intent buying prompts on live ChatGPT with search on and record three columns per prompt. Did a product card appear, was your brand in it, and did the price match your feed? A product that is never shown cannot be bought anywhere, so this read comes before any integration work.
Repeat the read on the live ChatGPT surface rather than a vendor API, and re-run every prompt when a major model ships, since product cards can change with the model.
What to do this quarter
Fix the feed, because discovery is now where OpenAI is investing and checkout happens on your own site. Validate a 100-item sample against the nine required fields, add GTINs where you own them, and confirm that price and availability match your site on the same day.
Then build the content that gets you shortlisted, and measure both on the same prompt set.
If your products never surface, the feed or its access is the gap; if they surface with wrong prices, the feed is the fix; if they surface correctly but a rival is recommended, the page is the fix. Ecommerce AEO is that second problem, and the wider answer engine optimization method is how you solve it.
