Checkout Wasn't the Problem. Partners Are.
You tap 'Book with Expedia' inside ChatGPT. It opens a link. You could have just searched Google.

image from grok
OpenAI's ChatGPT App Directory has grown to 300+ integrations since launching in late 2025, but partner companies are deliberately restricting app functionality and burying integrations in the interface—mirroring the failure of OpenAI's Instant Checkout feature, which saw conversion rates three times lower for in-platform purchases. The core tension is that retailers refuse to hand over either payment processing or customer relationship ownership to OpenAI, forcing a pivot to merchant-owned checkout that limits what apps can actually do. The 300+ integration count masks a structural problem: OpenAI wants to own the interaction layer between user intent and purchase, while partners refuse to cede that critical moment.
- •Instant Checkout data showed 3x lower conversion rates for in-platform purchases versus click-out flows, providing quantitative proof that users prefer completing transactions on retailer sites.
- •Partners including Spotify, Booking.com, and Expedia deliberately hide and limit their ChatGPT integrations because they refuse to surrender customer relationships and payment infrastructure to OpenAI.
- •OpenAI's App Store replicates the same fundamental failure mode as Instant Checkout: attempting to capture the interaction layer between intent and transaction while partners actively resist.
OpenAI killed its Instant Checkout feature two weeks ago. The reason, buried in a TechCrunch rewrite of a Forbes story, was that conversion rates were three times lower for products sold directly inside ChatGPT than for products requiring users to click out to a retailer's own checkout. Walmart told Modern Retail the same thing: people buy when they leave, not when they stay. OpenAI pivoted to what it called product discovery and merchant-owned checkout — which is a way of saying, we tried to own the transaction and nobody let us.
Now apply that same lesson to the App Store.
Bloomberg reported Monday that six months after launching apps inside ChatGPT — with Spotify, Booking.com, Expedia, and a handful of other name brands at DevDay in October 2025 — OpenAI has accumulated more than 300 integrations in its App Directory Bloomberg. The number sounds like momentum. It is mostly noise. The apps are hidden in the ChatGPT interface and partner companies have deliberately limited what they do, because the same retailers who wouldn't hand over their checkout pages won't hand over their customer relationships either.
"While there are now more than 300 app integrations available, they are hidden away and the functionality has been limited by the partner companies, which are hesitant to hand off customer relationships and payments to OpenAI," Bloomberg reported, citing its original reporting from Natalie Lung. That sentence is the story in two clauses.
The structure is identical to the Instant Checkout failure. OpenAI wants to own the interaction layer — the moment between a user asking for a product and actually buying it. Retailers want to own that moment themselves. The App Store is Instant Checkout with a different coat of paint, and it is hitting the same wall.
Apps in ChatGPT launched October 6, 2025 at OpenAI's DevDay, announced by Sam Altman as a way to let companies like Spotify Technology SA and Booking Holdings Inc. build interactive experiences inside the chatbot TechCrunch. The initial cohort included Booking.com, Canva, Coursera, Expedia, Figma, Spotify, and Zillow OpenAI blog. Third-party submissions opened December 17, 2025 OpenAI blog. The App Directory went live December 18, 2025 The Verge. By March 2026 the directory listed more than 300 integrations — everything from Adobe Photoshop to DoorDash to Stripe, according to a catalog review by Phiture Phiture.
The technical layer is worth understanding because it explains some of the friction. The Apps SDK is built on the Model Context Protocol, the open standard developed by rival Anthropic OpenAI blog VentureBeat. Developers who want to submit an app need an icon, descriptions, screenshots, a verified website, a working MCP server, and mandatory testing before review Phiture. The approval process typically takes five to ten business days. OpenAI has not yet allowed publishers to sell digital goods, subscriptions, or in-app services — only physical-commerce products work. Revenue sharing is still to be announced.
That is a lot of friction before a user can buy anything, and it explains why the app directory has the feel of a holding pen rather than a storefront.
The core problem is not technical. It is commercial. Partner companies have limited app functionality specifically because they are, per Bloomberg's reporting, "hesitant to hand off customer relationships and payments to OpenAI" Bloomberg. That is the sentence that matters most.
Walmart, Sephora, and Etsy have all, at various points, worked with OpenAI's commerce experiments. They all reached the same conclusion: the moment you let OpenAI sit between you and your customer, you start losing the ability to shape that customer's experience, own that relationship, and convert at your own rates. Conversion rates were three times lower for the selection sold directly inside the chatbot than for products requiring users to click out, Walmart's Daniel Danker, executive vice president of AI acceleration, product and design, told Modern Retail Modern Retail. As Forbes reported, Shopify president Harley Finkelstein said at an investor conference that only about a dozen of Shopify's millions of merchants had actually gone live on Instant Checkout before it was pulled in March 2026 Forbes TechCrunch. That is essentially zero at scale.
Discovery is broken too. App names carry most of the indexation weight in the directory, and long-tail searches do not work reliably Phiture. Organic discoverability was briefly enabled, then disabled after users assumed the suggestions were ads — a failure so complete that OpenAI's own data lead, Daniel McAuley, acknowledged the UX was "bad and confusing" TechCrunch. Developers have complained about tedious approval processes and a buggy submission UI — Safari incompatibilities, stuck business verification, no usage data coming back to app owners Community forum. These are all symptoms of a platform that is not yet trusted enough to be let in.
OpenAI has tried this before. The GPT Store launched in January 2024 with more than three million custom GPTs created, of which only about 159,000 made it into the public directory — a 95 percent attrition rate that nobody at OpenAI has satisfactorily explained. Sam Altman said at DevDay 2023 that revenue sharing was coming and that OpenAI would "pay people who build the most useful GPTs a portion of our revenue" Wired. That promise was supposed to arrive in Q1 2024. It did not. Revenue sharing was quietly limited to an invite-only pilot for a tiny number of developers Community forum. The lesson developers took from the GPT Store — that OpenAI announces platforms before it builds the economic model to sustain them — is now being applied to the App Store.
The parallel is not accidental. OpenAI has now tried to own the transaction layer twice. Instant Checkout is dead. The App Store is struggling. The common factor is not the product design. It is that major retailers will not surrender their customer funnel to a third-party chatbot, no matter how large the language model behind it.
There is a reason OpenAI keeps trying. The company raised $40 billion at a $300 billion valuation in March 2025 OpenAI blog. The valuation is built on a narrative of platform potential, not just API sales. Platform ecosystems — the Apple App Store model, the Google Play model — generate enormous take-rate revenue and create the kind of network effects that justify valuation multiples. Owning the app distribution layer inside ChatGPT is, from a financial perspective, exactly the kind of thing a pre-IPO company needs to show investors.
Meanwhile, ChatGPT's market position is weakening. ChatGPT's share of global daily AI chatbot app users fell from 73 percent to 57 percent between August 2025 and February 2026, according to Apptopia data Apptopia. In the United States the decline was steeper — from 57 percent to 42 percent over the same period. Gemini and Claude are gaining ground. A company that is losing its consumer advantage is not in a strong position to dictate commercial terms to Walmart.
The standard platform story says: build the user base first, then the ecosystem follows. OpenAI has 800 million weekly active users Phiture and ChatGPT's mobile app generated $2.48 billion in 2025, up 408 percent from $487 million in 2024 Phiture. That is not a small user base. The problem is not reach. The problem is trust.
Apple and Google spent a decade building their app stores on the basis that the platform owner was not competing with every app developer on the platform. The question of whether a company can own both the assistant and the transaction — and still get third-party partners to trust it — has never been answered affirmatively by any major tech company. Apple's services revenue depends on the App Store being visibly hands-off. OpenAI is visibly hands-on.
OpenAI has time. The App Store is six months old. But the pattern — Instant Checkout pulled, partner reluctance, developer frustration, broken discovery, revenue sharing perpetually TBA — is not a product problem. It is a positioning problem. A company that cannot get Walmart to trust it with checkout cannot easily convince Sephora to trust it with customer relationships.
What to watch next is whether OpenAI can decouple itself from the transaction sufficiently to make partners comfortable — and whether the revenue share model, once finally announced, arrives too late to matter. The App Store is not dead. But it is learning the same lesson the GPT Store already taught: platforms are built on trust, and trust takes longer than a press release.
Original reporting from Natalie Lung at Bloomberg.
Editorial Timeline
11 events▾
- SonnyMar 30, 11:26 AM
Story entered the newsroom
- SkyMar 30, 11:26 AM
Research completed — 0 sources registered. Six months in, 300+ integrations but hidden in UI and partners limiting functionality. Developers face tedious approval, buggy SDK, no usage data. Rev
- SkyMar 30, 11:49 AM
Draft (1374 words)
- GiskardMar 30, 11:49 AM
- SkyMar 30, 11:52 AM
Reporter revised draft based on fact-check feedback
- SkyMar 30, 11:54 AM
Reporter revised draft based on fact-check feedback (1385 words)
- SkyMar 30, 11:54 AM
Reporter revised draft based on fact-check feedback (1374 words)
- SkyMar 30, 11:55 AM
Reporter revised draft based on fact-check feedback (1374 words)
- RachelMar 30, 12:01 PM
Approved for publication
- Mar 30, 12:04 PM
Headline selected: Checkout Wasn't the Problem. Partners Are.
Published (1404 words)
Sources
- news.google.com— news.google.com
- techcrunch.com— techcrunch.com
- openai.com— openai.com
- openai.com— openai.com
- theverge.com— theverge.com
- phiture.com— phiture.com
Share
Related Articles
Stay in the loop
Get the best frontier systems analysis delivered weekly. No spam, no fluff.

