ChatGPT Projects vs Custom GPTs for Teams (2026): Which to Use Now That OpenAI Is Retiring GPTs

TL;DR: Projects vs custom GPTs

As of September 2026, a team should put new work in shared ChatGPT Projects, not custom GPTs. OpenAI says it plans to retire custom GPTs and move them to plugins: its published Enterprise timeline ends new GPT creation on September 25, 2026 and stops GPTs from running on December 11, 2026, and OpenAI says other plans may follow. Projects already hold 40 files on Business, share with up to 100 people, and keep chats, files, and instructions together. Every limit below comes from OpenAI's help center, read September 12, 2026.

Every article ranking for this comparison was written before OpenAI's retirement notice went up. Several still say projects can't be shared. That was true in early 2025. It hasn't been for almost a year.

The news that changes the answer: OpenAI is retiring custom GPTs

OpenAI's help center now opens every GPT article with the same notice: "We're planning to retire custom GPTs," with plugins as the recommended replacement. Here's the timeline OpenAI published for Enterprise workspaces:

• September 17, 2026: target date for the migration tool and an in-product banner.
• September 25, 2026 (planned): creation of new custom GPTs ends.
• December 11, 2026: scheduled retirement. OpenAI's words: "Custom GPTs stop running."

OpenAI says the transition "affects all ChatGPT plans" and that "the other plans may follow the same transition timeline," with notices in the product. So if you're on Business, don't treat December as a guaranteed extension. Treat it as the plan.

Personal plans were already cut off earlier. New GPT creation and publishing aren't available on Free, Go, Plus, or Pro. Existing GPTs still work there. Only Business, Enterprise, and Edu workspaces can still build new ones, and not for long.

ChatGPT Projects vs custom GPTs, side by side

When a team should use a project

Almost always, now. A project is the right home when the work has a history you want to keep: a client account, a hiring round, your sales proposals, a monthly board update. Teammates can open each other's chats, branch one to try a different approach, and pick up where someone left off.

On Business, Enterprise, and Edu you can invite individuals, groups, or the entire workspace by link. Link joiners get chat access by default; you upgrade the two or three people who should edit instructions and files. That split matters. Chat access lets people use the project. Edit access lets them change what it knows.

For a team, I build one project per recurring job, not one per person. "Client proposals" with your pricing sheet, three winning proposals, and a tone guide will outperform any GPT you were going to build for the same job. If you've set up the Claude version of this, it works almost the same way; I compared the mechanics on how to use Claude Projects.

When a custom GPT still makes sense (for about three months)

Honestly, only one case: you already have a GPT that people use every day and it works. Keep using it. Don't rebuild it in a panic this week, and don't add features to it either.

What I wouldn't do is start a new one. A GPT built in September is a GPT you'll migrate in October. If the job genuinely needs packaging, like the same instructions plus the same connected app for 40 people, that's what OpenAI is steering everyone toward with plugins, which bundle reusable instructions (skills) and apps together.

What breaks when you migrate a GPT to a plugin

OpenAI's retirement FAQ is unusually specific about what doesn't make the trip. Plan around these:

1. Custom actions don't transfer. If your GPT calls your own API, someone has to rebuild that with a supported connector or a custom MCP server. This is the expensive one.
2. The selected model doesn't carry over. Test the plugin, because it may answer differently.
3. Conversation starters and previous chats may not copy. Screenshot anything people rely on.
4. Drafts can't migrate. A GPT has to be published first. It doesn't have to be public, just published.
5. The original goes read-only. After migration, the old GPT keeps working until retirement, but you can't edit it and its creator can't delete it.
6. The replacement starts private. People who used the GPT don't automatically get the plugin. A link redirect doesn't grant access, either.

What does transfer: OpenAI says the GPT's instructions become a skill inside the new plugin, and connected apps come along as apps. Only the GPT's creator or a workspace admin can run the migration. Permission to use someone else's GPT doesn't let you migrate it.

What breaks in Projects (the limits nobody mentions)

Projects win this comparison, but they're not magic. The fine print from OpenAI's Projects article:

• Shared projects lock to project-only memory. They can't see anyone's personal memories or custom instructions, and you can't switch them back.
• ChatGPT Work isn't available in a project set to project-only memory. If a job needs Work, run it outside the shared project. What Work does is covered on ChatGPT Work for business.
• One delete hits everyone. Deleting a file in a shared project removes it for all members. Deleting the project deletes everyone's chats in it.
• Google Drive doesn't sync inside a project. ChatGPT can still search Drive, but the content isn't indexed ahead of time.
• Groups don't auto-update. People added to a group after you shared the project don't get notified or see it in their sidebar.
• 40 files is the ceiling on Business. Combine small documents into one before you hit it.

My migration plan for a small team

1. Inventory this week. List every GPT your team uses: name, creator, who uses it, and whether it has custom actions.
2. Copy the brains out. Paste each GPT's instructions into a document and download its knowledge files. OpenAI itself suggests recording instructions, reference material, integrations, and a few familiar prompts to test with.
3. Publish drafts you want to keep before September 25, 2026, since drafts can't migrate.
4. Rebuild simple GPTs as shared projects now. Instructions go in project settings, files go in the project. Test with the familiar prompts.
5. Migrate only the GPTs that need packaging once the tool appears in your workspace, and scope the custom-action rebuilds separately.
6. Tell people where to go. Pin the new project, share the link, and stop using the old GPT.

If you're setting up a fresh workspace, skip GPTs entirely and follow the order in my ChatGPT Business setup guide. Whether ChatGPT is even the right assistant for your team is a separate question, and ChatGPT for small business covers it. If you'd rather have the projects built and the old GPTs moved for you, that's part of the ChatGPT Business setup install.