Ryan Yu

Case study - September 2026

Why OpenAI's Instant Checkout failed

1. What Instant Checkout was

On September 29, 2025, OpenAI launched Instant Checkout: a way to buy a product without leaving a ChatGPT conversation [1]. It ran on the Agentic Commerce Protocol (ACP), co-built with Stripe and open-sourced on launch day [1][4]. Live for U.S. Etsy sellers on day one, single-item purchases only, with "over a million Shopify merchants" - Glossier, SKIMS, Spanx and Vuori named - promised as coming soon [1]. The merchant stayed the merchant of record: ChatGPT passed order details to the merchant's backend, and the merchant processed payment through its existing provider and owned fulfillment, returns, and support [1]. The money flow was a fee on completed purchases, free for users [1][3].

Wall Street's first reading was that this mattered: Etsy closed up nearly 16% on the announcement, Shopify up more than 6% [2]. One estimate at the time put ChatGPT's shopping surface at roughly 75 million product-related conversations a week [6], against an audience OpenAI framed at 700 million weekly users [1].

Five months later it was dead. The Information reported the scale-back on March 5, 2026 [13], OpenAI confirmed the pivot to press on March 6 [12], and the formal obituary ran as a blog post on March 24, 2026: merchants would use their own checkout experiences, and OpenAI would focus on product discovery [16][17].

2. The five months, month by month

WhenWhat happened
Sep 16, 2025Google announces AP2, its competing agent-payments protocol, 60+ backers [5]
Sep 29, 2025Instant Checkout launches: Etsy live, "1M+ Shopify merchants coming soon," single-item only, ACP open-sourced [1]. Etsy +16%, Shopify +6% same day [2]
Oct 2025OpenAI introduces ChatGPT apps at its developer conference - a second, merchant-controlled commerce surface that will matter later [17]
Oct 14, 2025Walmart announces it will sell through ChatGPT via Instant Checkout [7]
Oct 28, 2025PayPal adopts ACP, commits to wallet support and merchant-network integration starting 2026 [8]
Nov 2025Walmart goes live with ~200,000 products purchasable in chat [15]
Dec 2025Retail apps keep arriving (DoorDash and others) - on the app model, not Instant Checkout [18]
Jan 21-22, 2026The 4% merchant fee surfaces, effective Jan 26; Google and Microsoft charge nothing [9][10]
Feb 2026Forrester counts roughly 30 Shopify merchants actually live [11]. OpenAI launches advertising in ChatGPT [20]
Mar 3, 2026Finkelstein: only "about a dozen" of Shopify's millions of merchants ever went live, and "the holdup was on the AI firms' side" [14]
Mar 5, 2026The Information: OpenAI is scaling back; staff concluded users research products in ChatGPT but don't buy there [13]
Mar 6, 2026OpenAI confirms: "Instant Checkout is moving to Apps, where purchases can happen more seamlessly" [12]
Mar 18, 2026Wired publishes Walmart's numbers: in-chat conversion ran at one-third of click-out conversion. Danker calls the experience "unsatisfying" and Instant Checkout "a flop" [15]
Mar 24, 2026OpenAI's formal pivot post: merchants use their own checkout; ACP is repurposed for product discovery; Target, Sephora, Nordstrom, Lowe's, Best Buy, Home Depot, Wayfair signed on for discovery feeds. Same day: Walmart's own app (account linking, loyalty, Walmart payments) launches in ChatGPT, and Shopify announces its "Agentic Plan" [16][17]
May 2026Postscript: OpenAI opens its ads manager to all U.S. advertisers. The monetization model that survived is advertising, not transactions [20]

The count to hold onto: against a promised million-plus merchants, the number ever live on Instant Checkout was somewhere in the tens - "about a dozen" per Shopify's president [14], "roughly 30" per Forrester [11].

3. The merchant-side economics

This is the section most post-mortems skip, and it's where the failure becomes visible.

What merchants were offered. Access to 700 million weekly users at the exact moment of stated intent [1]. The fee was undisclosed at launch - OpenAI would only say "a small fee" on completed purchases [3]. The number surfaced in January 2026, when The Information reported (and Shopify confirmed) that Shopify merchants would pay OpenAI 4% per ChatGPT checkout sale starting January 26 [9]. For context: Amazon's referral fee runs around 15% before the ad spend sellers need to stay visible; ChatGPT's 4% plus Shopify's standard ~2.9% processing came to roughly 7% all-in, with no advertising requirement [10]. Google and Microsoft charged merchants nothing extra at all for their chatbot checkouts [9]. On price alone, the offer was generous.

What merchants were asked to give up. The checkout itself. Shopify president Harley Finkelstein put it plainly at a Morgan Stanley conference on March 3, 2026: the moat "is not just the payment. That is the checkout itself, that is subscriptions, that is the inventory, that is shipping taxes, all the different merchandising options that come with any type of transaction" [12]. Instant Checkout asked merchants to compress all of that into OpenAI's template - no loyalty integration, no coupons, no promotions, no customer data collection, no store pickup, no real-time inventory. A C-suite executive at a large specialty retailer had listed exactly those features in October 2025 as the bar for "ready for prime time." They never arrived [18].

What merchants actually wanted. Discovery, not delegated conversion. Circana's chief retail adviser Marshal Cohen: "Why would I want to give someone else control of my customer base, and then I become a commodity-based business?" [18]. Etsy's spokesperson said sales volume through Instant Checkout was low but ChatGPT was valuable as a discovery channel - and that Etsy was building its own ChatGPT app precisely because it gives Etsy "more control over the user experience" than the Instant Checkout template [18]. Instacart, Expedia, Lowe's, and CarMax all landed in the same place: happy to be discovered in chat, unwilling to close the sale there [18].

The incentive mismatch in one line: OpenAI needed merchants to surrender the highest-value surface in retail. Merchants wanted OpenAI as a free top-of-funnel. The 4% fee priced a transaction nobody on the merchant side was eager to have.

4. Why it died: three readings

The official reading. "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we're allowing merchants to use their own checkout experiences while we focus our efforts on product discovery" [16]. Read: a merchant-choice story, no failure admitted.

The reported reading. Four findings across The Information, Wired, CNBC, and Modern Retail:

  1. Users researched in ChatGPT and bought elsewhere [13].
  2. The one merchant with real data, Walmart, saw conversion inside chat at one-third of click-out conversion [15].
  3. Onboarding merchants was arduous and error-prone; item data was often stale; multi-item carts, promo codes, loyalty, and accurate tax and shipping never shipped [11][14][18].
  4. Almost no merchants ever went live - tens, not millions [11][14].

The structural reading. The reporting is right and still misses the order of operations. Three layers:

Layer 1: the checkout mostly didn't exist. A buy button only rendered if the merchant had enrolled, the account was in the rollout, and the query was phrased the right way. Against every product page on the internet, tens of enrolled merchants is a rounding error - which means for nearly every buyer with money in hand, the funnel ended at the gate before trust was ever asked for. The industry argument about whether consumers will trust an agent with their card assumes buyers reach a checkout. Most never did.

Layer 2: where it did exist, it fought how people actually shop. Single-item carts only. Walmart's Danker: customers "fear that when checkout happens automatically after every single item, they're going to receive five boxes when they actually just want it all in one" [15]. No loyalty, no bundles, no promotions, no way to see the thing Walmart's own site would have shown you. The design asked buyers to do less than a real cart in exchange for doing it inside a chat window - and the exchange rate was a third of the conversion [15].

Layer 3: trust context doesn't travel. The buyer arrives at an in-chat checkout as a stranger: no account, no saved cart, no loyalty identity, no visible returns story. The survivors of the shakeout all patched this the same way - by handing the money moment back to whoever already owns the trust. Walmart's replacement for Instant Checkout is Sparky, its own assistant with account linking, loyalty, and Walmart payments, running inside ChatGPT [15][17]. Etsy's fix is its own app, for control [18]. Amazon never had the problem: its agent shows a buy button in chat but closes in Amazon's own cart, on Amazon's own identity, card, and returns. ChatGPT's discovery traffic was real - Walmart says ChatGPT brings in about twice the rate of new customers as search engines [15] - but discovery was all merchants ever wanted from it.

5. What it proves about a consumer trust layer

Instant Checkout is the cleanest natural experiment agentic commerce has produced: maximum distribution (700M weekly users), working rails (ACP/Stripe did their job), genuine purchase intent - and a five-month collapse. Strip the execution excuses and what's left is a spec:

  1. 1.

    Availability is the zeroth job.

    Before consent, before reversibility: a buy button has to exist at the moment of intent, for this buyer, on this account, for this merchant. A trust surface that only appears for enrolled merchants is a trust surface most buyers never meet. Any trust layer gated per-merchant repeats Instant Checkout's five months.

  2. 2.

    The trust context has to travel with the purchase.

    Amazon works because one company owns identity, card, fulfillment, and returns, so its agent inherits trust instead of borrowing it. The open-web version of that has to be carried across surfaces - identity, consent scopes, payment credential, and reversibility terms that show up with the buyer instead of being rebuilt at every merchant.

  3. 3.

    The three jobs a checkout page does invisibly have to be re-created deliberately.

    A normal checkout page delivers consent (I know what I'm agreeing to), transparency (I can see and check what's happening), and reversibility (if it's wrong, I can undo it). A chat thread collapses all three into one line of text and a yes/no prompt.

  4. 4.

    Merchants will not outsource the money moment.

    Every survivor kept or regained control of checkout. The design assumption has to be merchant-retained checkout with an agent-readable consent surface on top - not an agent that replaces the merchant's surface.

  5. 5.

    Transactions lost to advertising as the business model.

    OpenAI's post-Instant-Checkout revenue move was ads, not take-rate [20]. Anyone building the trust layer should note who ended up paying for the discovery surface.

JOB 0AVAILABILITYa buy button exists at the moment of intent
JOB 1CONSENTI know what I am agreeing to
JOB 2TRANSPARENCYI can see and check what is happening
JOB 3REVERSIBILITYif it is wrong, I can undo it

Glossary

ACP (Agentic Commerce Protocol).
The open standard, co-built by OpenAI and Stripe, that let ChatGPT and a merchant's systems talk to each other to complete a purchase. Open-sourced Sept 29, 2025 [1].
Merchant of record.
The party legally responsible for the sale - payment processing, fulfillment, returns, support. Under ACP this stayed with the merchant, which was a selling point [1].
AP2 (Agent Payments Protocol).
Google's competing open standard for agent-initiated payments, announced Sept 16, 2025 with 60+ backers [5].
Instant Checkout.
OpenAI's killed feature: buy a single item inside a ChatGPT conversation, from an enrolled merchant.
ChatGPT apps.
Merchant-built experiences that run inside ChatGPT (announced Oct 2025). The survivors' model: the merchant owns the surface, ChatGPT supplies the audience [17].
Sparky.
Walmart's own AI shopping assistant, now embedded in ChatGPT and Gemini - Walmart's replacement for selling through Instant Checkout [15].

Footnote

  • The two live-merchant counts ("about a dozen" [14] vs "roughly 30" [11]) come from different observers a few weeks apart and count slightly different things. The case study's point only needs the shared fact: tens, against a promised million-plus.

Sources

  1. OpenAI, "Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol," Sep 29, 2025 - openai.com
  2. CNBC, "Etsy pops 16% as OpenAI announces ChatGPT Instant Checkout for the shopping site," Sep 29, 2025 - cnbc.com
  3. Reuters, "OpenAI partners with Etsy, Shopify on ChatGPT payment feature," Sep 29, 2025 - reuters.com
  4. TechCrunch, "OpenAI takes on Google, Amazon with new agentic shopping system," Sep 29, 2025 - techcrunch.com
  5. Google Cloud Blog, "Announcing Agent Payments Protocol (AP2)," Sep 16, 2025 - cloud.google.com (backer count via TechCrunch, Sep 16, 2025 - techcrunch.com)
  6. Marketplace Pulse, "ChatGPT's Instant Checkout is an Instant Threat," Oct 2, 2025 - marketplacepulse.com
  7. Walmart corporate, "Walmart Partners with OpenAI to Create AI-First Shopping Experiences," Oct 14, 2025 - corporate.walmart.com
  8. PayPal newsroom, "OpenAI and PayPal Team Up to Power Instant Checkout and Agentic Commerce in ChatGPT," Oct 28, 2025 - newsroom.paypal-corp.com
  9. PYMNTS, "Shopify Merchants to Pay 4% Fee on ChatGPT Checkout Sales," Jan 21, 2026 - pymnts.com
  10. Marketplace Pulse, "ChatGPT's 4% Fee Confirms Marketplace Economics," Jan 22, 2026 - marketplacepulse.com
  11. CNBC, "OpenAI's first try at agentic shopping stumbled. It's trying again," Mar 20, 2026 - cnbc.com
  12. Digital Commerce 360, "OpenAI shifts checkout plans in its agentic commerce strategy," Mar 6, 2026 - digitalcommerce360.com
  13. The Information, "OpenAI Scales Back Shopping Plans for ChatGPT," Mar 5, 2026 - theinformation.com
  14. Forbes, "Why OpenAI's Checkout Retreat Spells Trouble For Its Commerce Strategy," Mar 10, 2026 - forbes.com
  15. Wired, "Why Walmart and OpenAI Are Shaking Up Their Agentic Shopping Deal," Mar 18, 2026 - wired.com
  16. OpenAI, "Powering Product Discovery in ChatGPT," Mar 24, 2026 - openai.com
  17. CNBC, "OpenAI revamps shopping experience in ChatGPT after struggling with Instant Checkout offering," Mar 24, 2026 - cnbc.com
  18. Modern Retail, "What went wrong with ChatGPT's Instant Checkout," Mar 27, 2026 - modernretail.co
  19. Search Engine Land, "Walmart: ChatGPT checkout converted 3x worse than click-out," Mar 19, 2026 - searchengineland.com
  20. Marketplace Pulse, "Rufus, ChatGPT, and the Agentic Ad Convergence," May 7, 2026 - marketplacepulse.com

This case study is the autopsy behind the claim on the research page: The Agentic Commerce & Payments Landscape.

Back to the research Back to the thesis