ChatGPT Apps Overview
Around 800 million people use ChatGPT every week. Most of them still think of it as a chatbot, a place to ask questions and get answers. That picture is changing.
Since October 2025, OpenAI has been building something closer to an application platform. Developers can now publish interactive apps that run directly inside ChatGPT conversations, with real interfaces, real integrations, and real actions. The format is called ChatGPT Apps, and it represents a meaningful shift in what the platform can do.
This article explains what ChatGPT Apps are, how the underlying technology works, and why the format is different from what OpenAI has offered developers before.
What ChatGPT Apps actually are
A ChatGPT App is an interactive application embedded inside the ChatGPT interface. It can display visual components - maps, playlists, forms, product listings, and connect to external services, all within a single conversation.
The definition matters because it separates ChatGPT Apps from ChatGPT itself. The core product is a conversational AI. Apps are third-party software that extend what that conversation can do. When you ask ChatGPT to help you find an apartment and the Zillow app surfaces with an interactive map of listings filtered by your budget, you are using a ChatGPT App. The conversation started with ChatGPT; the tool doing the work is an app built by Zillow's developers.
OpenAI introduced the format in October 2025 and opened public submissions in December 2025. The app directory, accessible at chatgpt.com/apps, is now available to all logged-in users on Free, Go, Plus, and Pro plans.
How ChatGPT Apps are built
Apps are built with the Apps SDK, OpenAI's development toolkit released in preview alongside the format announcement. The SDK is built on top of the Model Context Protocol (MCP), an open standard originally developed by Anthropic in late 2024.
OpenAI chose to build its app ecosystem on a standard that competing AI platforms - including Claude, Cursor, and VS Code Copilot already support. In principle, an app built for ChatGPT using MCP can be adapted to work across other AI platforms that support the same protocol, which reduces the risk of building for a single ecosystem.
The SDK gives developers control over both the backend logic and the frontend interface. Apps can display visual widgets inline in the chat, in picture-in-picture mode, or in fullscreen. They can request OAuth authentication so users connect their accounts securely. They can call external APIs and return structured data to ChatGPT, which incorporates that data into the ongoing conversation.
For companies that already have web or mobile products built with reusable APIs, much of the existing backend can carry over. The development effort focuses on the conversational layer, how the app responds to natural language and integrates with the chat context ChatGPT passes in.
How users find and use ChatGPT Apps
There are three ways an app reaches a user inside ChatGPT.
The first is the app directory. Users browse to chatgpt.com/apps or open the tools menu inside ChatGPT and search for an app by name or category. Once connected, apps stay available across future conversations.
The second is direct mention. Users call an app by name using an @ mention - for example, "@Canva, turn this outline into a slide deck" and the app activates within the chat. This works for users who already know the app exists.
The third is indirect prompting. ChatGPT watches the conversation for intent signals and proactively suggests relevant apps when they could help. Someone planning a trip might see a Booking.com suggestion appear before they've asked for it. OpenAI has described this feature as experimental and driven by signals including conversational context, usage patterns, and user preferences.
When a user connects to an app for the first time, ChatGPT displays what data will be shared with the third party and provides the app's privacy policy for review. All apps go through a review process before publication that checks functionality, safety compliance, and privacy policy completeness.
Who is already building ChatGPT Apps
The early set of published apps includes companies across several industries. Spotify, Apple Music, and DoorDash handle media and delivery. Zillow, Expedia, and Booking.com cover real estate and travel. Canva, Dropbox, and Figma work on the creative and productivity side. Coursera and Tripadvisor round out the initial list.
These are not small experiments. These companies committed engineering resources to build and maintain a new surface because being discoverable inside ChatGPT conversations, at the moment users are actively planning or deciding, carries real distribution value.
New apps are being added regularly as more developers work through the submission and review process.
What ChatGPT Apps can and cannot do
They can guide users through multi-step processes, complete purchases, retrieve personalized data, display interactive content, and integrate with external platforms. Apps can also be localized, developers can control which regions their app appears in and adapt content accordingly.
Monetization remains in early development. External checkout routing users to the developer's website to complete a purchase is the primary available path. In-app checkout with saved payment methods is in limited rollout for select partners. OpenAI has indicated that digital goods and additional commerce options are coming, but has not published a timeline.
Why the format is worth watching
The practical shift that ChatGPT Apps represent is this: for the first time, a service can reach a user inside the moment they are actively working through a problem, without requiring them to open a separate app, search for a product, or click through to a website.
That is a different kind of distribution. Whether it becomes a dominant channel depends on how the ecosystem develops, how users adapt their behavior, and how well the apps being built actually solve problems worth solving. The early signs 800 million weekly users, recognizable launch partners, an open standard as the technical foundation suggest the format has enough momentum to matter.