ChatGPT Projects vs custom GPTs: what goes where
Use a project for work that keeps changing and that other people add to: it holds the chats, files and instructions for one piece of work, and everyone you share it with builds on the same material. A custom GPT is a reusable assistant with fixed instructions, files and tools, and OpenAI plans to retire custom GPTs on 11 December 2026, moving that job to plugins. Personal plans can no longer create new GPTs, so on those plans the choice today is between a project and a plugin.
The facts below are from OpenAI’s and Anthropic’s help centres, OpenAI’s developer docs and ChatGPT’s pricing page, checked on 30 September 2026. The custom GPT dates are OpenAI’s plan and may move, so check the notice in your account.
What is a ChatGPT project?
A project keeps “related chats, files, and instructions together”, and ChatGPT can draw on all of it when you chat in that project:
- Files: PDFs, spreadsheets, docs, images or pasted text. Free allows 5 per project, Go and Plus 25, and Pro, Business, Enterprise and Edu 40.
- Links to Google Drive and Slack, in a private project, and answers you save from a chat.
- Instructions that apply only inside the project and override your custom instructions.
Its chats can also refer to earlier chats in the same project. On every plan you can create as many projects as you like and share them: members see the chats, files and instructions, and ChatGPT leaves each member’s personal memories out. Sharing a ChatGPT project has its own steps and limits.
What is a custom GPT?
A GPT is a version of ChatGPT “configured for a specific purpose”. Its builder sets:
- Instructions, applied to every conversation.
- Knowledge: up to 20 files, each up to 512 MB.
- Capabilities: web search, image generation, Canvas, and Code Interpreter & Data Analysis.
- Apps or actions: apps use services people have connected, and actions call APIs the builder defines. A GPT uses one of the two at a time.
OpenAI says GPTs “do not use saved memory, custom instructions, or previous conversations”, and builders can’t read people’s chats with them. In a Business, Enterprise or Edu workspace, a GPT can be shared with up to 100 people or groups, the whole workspace, a link, or the GPT Store. On a personal plan, invite-only means only you.
Can you use a GPT inside a project?
Partly. A GPT can answer in a project chat that already exists. A chat that starts with a GPT sits outside the project, and chats made with a GPT can’t be moved into one.
What is happening to custom GPTs?
On 11 September 2026 OpenAI announced that it plans to retire custom GPTs on every plan, and it recommends moving them to plugins.
- New GPTs. OpenAI’s help centre says Free, Go, Plus and Pro can no longer create or publish them, though its pricing page still lists creating GPTs on Go, Plus and Pro. On 30 September, Business, Enterprise and Edu workspaces still could where settings allowed; OpenAI planned to end creation in Enterprise workspaces on 26 October 2026 and expected other plans to follow.
- Retirement. Scheduled for 11 December 2026, or 11 February 2027 for Enterprise workspaces with an approved deferral. On that date GPTs and their pages become inaccessible; until then they keep working.
- A GPT you built. When migration reaches your account, My GPTs shows Migrate to plugin. The instructions become a skill, the knowledge files become reference files, and connected apps come across. Custom actions, the chosen model, past conversations and sharing settings stay behind, and the plugin starts private.
- A GPT someone else built. Only its builder can migrate it (or, in an Enterprise workspace, an admin). Check its notice for a replacement.
What replaces a custom GPT?
A plugin: a package of skills (reusable instructions, examples and code), connected apps and reference files for a task. ChatGPT uses an installed plugin when it fits your request, or when you name it with @. To build one, install Plugin Creator from the plugin directory and describe what you want. The directory is on every plan; on Business and Enterprise you can share a plugin with colleagues or publish it to your workspace’s directory, which stays separate from the public one.
Skills also work on their own. OpenAI’s skills article lists them for Business, Enterprise and Edu, and its pricing page shows a skills beta on the personal plans.
Which plans can create and share each?
Projects are the only one of the three that every plan can both create and share.
| Create | Share | |
|---|---|---|
| Project | Every plan, as many as you like | Every plan: up to 5 people on Free, 10 on Go and Plus, 100 on Pro, Business, Enterprise and Edu |
| Custom GPT | New ones only in Business, Enterprise and Edu workspaces, until creation ends there | Workspaces: people, groups, the workspace, a link or the GPT Store. Personal plans: existing GPTs only |
| Plugin | With Plugin Creator; a personal plugin starts private | Business and Enterprise: colleagues or the workspace directory. The public directory takes a separate submission |
When should you use a project, and when a GPT or plugin?
OpenAI draws the line itself: a GPT is “single player (just you and chat)” and “static content”, while a shared project is “multi-player” and a “live context hub for recurring shared work around a topic”. So material that changes, and that several people add to, goes in a project. A procedure or an assistant that should behave the same way in any chat goes in a skill or a plugin.
| What you have | Where it goes | Why there |
|---|---|---|
| The brief, research and files for one client or piece of work | Project files | Every chat in the project uses them |
| Rules for that work, such as the client’s spelling and tone | Project instructions | They apply only in that project |
| Your role, tone and format for every chat | Custom instructions | They apply to all your chats |
| Work several people add to week after week | A shared project | Everyone sees the chats, files and instructions |
| A procedure you repeat anywhere, like the weekly report format | A skill, alone or in a plugin | ChatGPT uses it when the task comes up |
| An assistant the whole company uses, with fixed instructions, files and tools | A plugin shared to the workspace | It does a GPT’s job, on Business and Enterprise |
| A GPT you built and rely on | Migrate it to a plugin before it retires | Instructions, files and apps carry over; custom actions need rebuilding |
| A GPT someone else built | Ask its builder about a replacement | You can’t migrate a GPT you only use |
Does Claude have custom GPTs?
In Claude, three pieces do that job. A project holds files and instructions for one piece of work; every plan has projects (five on Free), and Team and Enterprise can share them with colleagues. A skill is a procedure Claude loads when a task needs it, on every plan with code execution on. A connector gives Claude tools and data, as apps and actions do for a GPT.
What if your team uses ChatGPT and Claude?
Say your strategist builds in ChatGPT and your designer works in Claude. A shared project or a GPT opens only for someone signed in to ChatGPT, so the designer either joins with a ChatGPT account and works there, or keeps a Claude project with its own copy of the brief, which someone updates by hand.
Two things travel better. Skills do: OpenAI says its skills “build on the open agent skills standard” at agentskills.io, and Anthropic’s help centre points to the same standard and says skills you create “aren’t locked to Claude”. Going by both vendors’ documentation, a skill written to that standard can move between the two. So does the project itself, if it lives outside both apps and each person opens it through MCP, the open standard that lets an AI app use tools and data held somewhere else. Claude adds an MCP server as a custom connector on every plan (Free allows one). For ChatGPT, OpenAI’s help centre gives full read and write on Business, Enterprise and Edu, where an admin adds it, and read and fetch only on Pro; its developer docs list more plans, so check yours.
Where Alcoven fits
Your AI does the reading and writing (Alcoven runs no model for this). Alcoven holds a project’s brief, documents, tasks and history in one place, shared by the people and agents on it. Add Alcoven to Claude, or to ChatGPT on a plan where an MCP tool can write, at https://app.alcoven.co/mcp, and choose the project when you sign in; clone it with git to work from Claude Code or Codex; or ask the project in the Alcoven app. Whatever your AI changes comes back under your name, for everyone on the project, and every document keeps every past version. Your ChatGPT projects and plugins keep their jobs, and Alcoven sits beside them. Free covers one project and up to five people writing to it, with no card. Start a project free
Sources
Checked on 30 September 2026.
- Projects in ChatGPT, OpenAI Help Center
- GPTs in ChatGPT, OpenAI Help Center
- Creating and editing GPTs, OpenAI Help Center
- Sharing and publishing GPTs, OpenAI Help Center
- Configuring actions in GPTs, OpenAI Help Center
- Custom GPT retirement and migration FAQ, OpenAI Help Center
- ChatGPT release notes, OpenAI Help Center (planned custom GPT retirement, 11 September 2026)
- Plugins in ChatGPT, OpenAI Help Center
- Skills in ChatGPT, OpenAI Help Center
- ChatGPT custom instructions, OpenAI Help Center
- Developer mode and MCP apps in ChatGPT, OpenAI Help Center
- ChatGPT developer mode, OpenAI developer docs
- Build skills, OpenAI developer docs
- ChatGPT pricing, OpenAI
- What are projects?, Claude Help Center
- What are skills?, Claude Help Center
- Get started with custom connectors using remote MCP, Claude Help Center