ChatGPT apps let a brand put its own data and services inside ChatGPT conversations, and as of 2026 they are the main branded channel inside ChatGPT, because OpenAI is retiring custom GPTs on December 11, 2026. Apps now ship as plugins in a directory, and ChatGPT can suggest them when they're relevant. This post explains what changed, when an app is worth building, how to plan one and how it fits with organic AI visibility.
OpenAI has changed this area several times in a year. Check its current help center and developer docs before you commit to a build.
How we got here
The branded-channel options inside ChatGPT have moved quickly:
| When | What happened (per OpenAI) |
|---|---|
| October 2025 | OpenAI introduced apps in ChatGPT and a preview of the Apps SDK, built on the Model Context Protocol (MCP). Users can discover apps when ChatGPT suggests one or by calling them by name. |
| December 2025 | OpenAI opened app submissions for review, with approved apps listed in an in-product directory. |
| July 9, 2026 | The app directory was migrated to a plugin directory, with existing apps packaged into plugins. |
| December 11, 2026 | Scheduled retirement date for custom GPTs, per OpenAI's migration FAQ. |
Custom GPTs: what retirement means
If your brand published a custom GPT, plan its migration now. According to OpenAI's custom GPT retirement and migration FAQ:
- Custom GPTs are scheduled to retire on December 11, 2026; Enterprise workspaces with an approved deferral have until February 11, 2027.
- In the migration, the GPT's instructions become a skill in a new plugin and knowledge files become reference files.
- Custom actions don't transfer, and the selected model doesn't carry over.
- Sharing settings don't carry over. A migrated personal plugin starts private, and making it public requires a separate plugin submission.
- After migration the original GPT becomes read-only until retirement.
For most brands, the honest assessment is that a GPT built on instructions and a few PDFs was a thin channel. If yours had real usage, migrate it. If it relied on custom actions to call your systems, rebuild that part as an app.
Apps, skills and plugins
OpenAI's help center describes the current model this way:
- Apps connect ChatGPT to external data and actions. Built with the Apps SDK, an app is an MCP server that exposes tools, optionally with an interactive interface rendered in the chat.
- Skills package instructions and reference material for a workflow.
- Plugins bundle skills, apps and app templates so a workflow is easy to enable. The directory lists plugins.
Whether a user can install or invoke a plugin depends on their plan, workspace settings, role, region and the capabilities of its apps, according to OpenAI.
When a ChatGPT app is worth building
An app earns its place when it does something useful with your live data or services that ChatGPT can't do from the open web. Use this test:
| Strong case | Weak case |
|---|---|
| Search live inventory, availability or listings | Restating product pages |
| Book, reserve or schedule | Brand storytelling |
| Answer account-specific questions after sign-in | General FAQs already on your site |
| Calculators or configurators with your pricing rules | A chatbot persona |
| Query a proprietary dataset or tool | A downloadable guide |
If your idea is in the right column, the better investment is making your site answer-first so ChatGPT cites it in ordinary answers. See how to get your brand mentioned in ChatGPT.
What submission requires
OpenAI's current plugin guidelines and submission docs describe a review process. Highlights as of October 2026:
- Verified individual or business identity in the OpenAI Platform dashboard
- Directory metadata: a display name of up to 30 characters, descriptions, category and HTTPS links to your website, support, privacy policy and terms (required for MCP submissions)
- Five positive and three negative test cases, plus release notes
- For apps with MCP connections: a demo video, reviewer test credentials if sign-in is needed, and explicit read-only, open-world and destructive hints on every tool, with a justification for each
- Only one active review per plugin; rejections come with a reason by email
OpenAI's app developer terms say it doesn't guarantee any placement, ranking or promotion in the directory or in ChatGPT's suggestions.
Designing an app ChatGPT will actually use
Because ChatGPT decides when to suggest or call your app, tool design works like AEO for software. The model reads your tool names and descriptions to decide whether you're relevant.
- Name tools for the job:
search_available_rentals, notquery_v2. - Describe when to use each tool in plain language, including what it can't do.
- Return structured, factual results with clear fields (price, availability, link), so answers quote them accurately.
- Make the interface optional. OpenAI's docs say tools should still work without the UI.
- Handle the empty case: "no availability on those dates" is a useful answer.
- Link back to the relevant page on your site for next steps.
For agent readiness beyond ChatGPT, see AI agents and agentic browsing.
Example: a vacation rental company plans an app
This is an illustrative plan for a fictional company.
Goal: let travelers check real availability and prices from inside a ChatGPT trip-planning conversation.
Tools:
search_rentals(location, dates, guests, pet_friendly)returns up to ten properties with nightly price, rating and linkget_property_details(property_id)returns amenities, house rules and cancellation policycheck_availability(property_id, dates)returns a yes or no with total price
What's deliberately left out: booking and payment inside ChatGPT for version one, to keep review simpler and risk lower; the app links to the booking page instead.
Success measures: app invocations, click-throughs to booking pages tagged with UTM parameters, and bookings from those sessions. See attribution for AI search.
Organic work in parallel: the company keeps improving its destination guides and property pages, because most trip-planning conversations will never invoke an app.
Apps versus organic visibility
Organic visibility reaches every ChatGPT user who asks a relevant question; an app reaches users who invoke it or accept ChatGPT's suggestion. For most brands, organic visibility comes first and an app is a second step for brands with a real interactive use case. See which AI platform to prioritize.
The bottom line
ChatGPT's branded channel is now the app, packaged as a plugin, and custom GPTs are on their way out on December 11, 2026. Migrate any GPT with real usage, rebuild custom actions as an app, and only build an app when it does something useful with your live data or services. Design tools so ChatGPT can tell when you're relevant, measure invocations through to outcomes, and keep investing in the organic answers where most users will meet you first.
