OpenAI opens Apps SDK for apps in ChatGPT, but still hasn't defined how developers will make money

Editor's note: this article mentions 'October 2025' as the original launch month of the Apps SDK, but the official source consulted (OpenAI's announcement) does not confirm that date; it only records an update published on November 13, 2025, without citing the month of the original launch. The other data and quotes in the text (800 million users, pilot partners, monetization timelines, open source nature via MCP) were confirmed in the source.

Since October 2025, OpenAI has let any developer build a native app inside ChatGPT via the Apps SDK. For whoever decides where to invest the next product cycle, the real catch isn't the promised distribution: it's what's still left to settle about monetization.

Since the launch of the Apps SDK, OpenAI has let any developer build a native app inside ChatGPT. For whoever decides where to invest the next product cycle, the real catch isn't the promised distribution: it's what's still left to settle about monetization.

OpenAI published the announcement "Introducing apps in ChatGPT and the new Apps SDK", with an update recorded on November 13, 2025 extending the feature to Business, Enterprise, and Edu accounts. Since that launch, the pitch remains the same: any developer can build an app that runs inside a ChatGPT conversation, with its own interface, its own logic, and a connection to the builder's own backend.

The reason for a founder to pay attention isn't the technical novelty. It's the question OpenAI still hasn't fully answered: is it worth more to build inside ChatGPT's distribution, or to keep your own channel? The announcement gives strong hints on both sides.

What the Apps SDK really is

Technically, the Apps SDK is an extension of the Model Context Protocol (MCP), the open standard that already let ChatGPT talk to external tools. OpenAI took that protocol, which was only about exchanging data and commands, and added an interface layer: now the developer also designs the screen that appears inside the chat, not just the logic behind it.

The detail that changes the platform-dependency math is this: the SDK was released as open source and built on an open standard. In theory, that means an app built for the Apps SDK can run on any other client that adopts the same MCP protocol, not just ChatGPT. That's a real difference from precedents like the 2007 Facebook Platform or Twitter's closed APIs: there, leaving the platform meant rewriting the product from scratch.

The distribution promise has fine print

OpenAI describes the central appeal in a direct line within the announcement itself:

building with the Apps SDK makes it possible to reach over 800 million ChatGPT users at just the right time.

OpenAI, official statement "Introducing apps in ChatGPT"

The problem for whoever decides where to invest is the verb "reach." The app doesn't appear because the user chose it in a store: it appears when ChatGPT decides to suggest it, or when the user already knows the app's exact name and calls on it in the conversation ("Spotify, make me a playlist"). Organic discovery, which in an app store depends on category, search, and ratings, here depends on the model's own judgment of relevance.

The announcement confirms this by framing whoever gets "more prominent" placement as whoever meets "higher standards of design and functionality" set by OpenAI itself. It's the same curation logic as an App Store, except decided by criteria the developer can't audit and that can change without notice.

Who's already placed a bet, and why it matters

The pilot partners, named directly in the announcement, aren't early-stage startups: Booking.com, Canva, Coursera, Figma, Expedia, Spotify, and Zillow. All of them already have an established distribution channel of their own, a recognized brand, and a customer base that doesn't depend on ChatGPT to exist.

That signals something specific about the right risk appetite for this stage of the product. For these companies, the app in ChatGPT is an incremental channel: one more place where an already loyal customer can find the service, without replacing anything that already works. The risk of an OpenAI rule change affects the margin of that extra channel, not the whole business.

For a startup that's still building its brand and customer base, the equation is different. Betting the whole product on a channel whose monetization rules the platform itself admits it hasn't finalized is betting on a promise, not a contract. The announcement is explicit about this: "later this year, we'll share more details on how developers can monetize their apps" is a statement of intent, not a term sheet.

The monetization gap, broken down

OpenAI's text lists what was still open at the time of the announcement:

• App submission and review for publication: promised for "later this year" (2025), still without a fixed date in the statement itself.
• A dedicated directory for users to browse and search for apps: also treated as a next step, not as a feature available at launch.
• Support for the Agentic Commerce Protocol, an open standard cited to enable instant checkout inside ChatGPT, mentioned as something to be detailed "soon."
• More granular controls over which categories of data each app can use to personalize results, also slated for "later this year."

In other words: at the moment the developer gains technical access to the SDK, OpenAI still hadn't published how the money would flow between user, app, and platform. Whoever built in the very first month operated on the assumption that the rule, once it came, would be favorable enough to justify the engineering investment made in the dark.

The counterpoint: why it may be worth the risk anyway

The strongest argument in favor of going in early is the usual one for any new platform: whoever builds the integration first learns the suggestion algorithm's behavior before competitors do, and is already ready by the time the directory and monetization finally leave the drawing board. In attention markets, being on the launch-partner list carries brand value on its own, as shown by the fact that OpenAI picked big, non-anonymous names for the pilot.

The fact that the Apps SDK is open and based on MCP also reduces the worst-case scenario: unlike the App Store, where Apple can simply reject or remove an app, an app built for ChatGPT, even if poorly suited to it, can in theory still be ported to another client that speaks the same protocol. That doesn't eliminate the risk of depending on OpenAI's curation for distribution inside ChatGPT specifically, but it avoids the scenario of rewriting everything from scratch if the relationship with the platform sours.

What to decide today

For whoever already has an established distribution channel and brand, like the pilot partners, the Apps SDK pays off as a low-cost incremental bet: the main product keeps existing outside ChatGPT, and the app inside it is just one more touchpoint.

For whoever is deciding to build the entire product around this distribution, the right question isn't "how many users does ChatGPT have," but "what happens to my business if the monetization rule that doesn't exist yet turns out unfavorable, or if the featured curation simply doesn't pick me." Until OpenAI publishes the monetization terms promised for the end of 2025, any decision to bet everything on this channel is a bet on a promise, not on a finalized business plan.