The difference in one line each
OpenAI’s own help center puts it like this: a GPT is single player and static, curated by an expert and meant to scale knowledge across the organization, like an “HR Benefits GPT”. A shared project is multiplayer and evolves as people add chats, files and instructions, like a “Q4 planning hub”.
Side by side
| Custom GPT | Project | |
|---|---|---|
| Who works in it | Each user alone with the GPT | Everyone invited, who can see each other’s chats |
| Changes over time | Only when the builder edits it | With every chat and file the team adds |
| Uses earlier conversations | No, every chat starts fresh | Yes, chats in the same project |
| Uses your memory and custom instructions | No | Project instructions override your custom instructions; memory depends on the project setting |
| Tools | Capabilities plus apps or custom actions, not both | Same tools as normal chats, plus connected apps |
| Can new ones be created | Not on personal plans; ended for affected Enterprise workspaces on October 26, 2026 | Yes, on every plan, unlimited |
| Future | Scheduled to retire December 11, 2026 | Continuing |
Sources: OpenAI’s GPTs, Projects and GPT retirement articles, November 2026.
When a project is the better fit
- The work is ongoing and the context grows: a client, a campaign, a quarterly report.
- Several people contribute and should see each other’s work.
- You want ChatGPT to remember earlier conversations about the same topic.
OpenAI’s retirement FAQ makes the same recommendation: if you used a GPT to keep instructions and reference files together for ongoing work like writing, research or planning, start a project.
When you need what a GPT did, after December 11
A GPT’s strength was a reusable, packaged workflow that many people can call: the same instructions, the same reference files, the same tool, every time. That role now goes to plugins. A plugin bundles skills (instructions), reference files and connected apps, can be shared within a Business or Enterprise workspace and is called with an @ mention, also inside other chats. Creators of an existing GPT can use My GPTs > Migrate to plugin; new ones are built with Plugin Creator. For an assistant that runs on a schedule or in Slack, Business and Enterprise also have workspace agents.
Moving a GPT to a project, step by step
- Select New project in the sidebar.
- Copy the GPT’s instructions into Project settings.
- Upload the GPT’s knowledge files (up to 40 on Business, Pro, Enterprise and Edu).
- Start a new chat in the project and test it with prompts you used with the GPT.
- If the GPT used a custom action, plan a replacement: actions do not transfer to projects or plugins and may need an app or a custom MCP server.
The common mistake: a GPT learns from its users
Many teams assume their company GPT gets smarter as colleagues use it. It does not. GPTs do not use memory or previous conversations, and builders cannot even see users’ chats. Every improvement is a manual edit to the instructions or files. Projects do accumulate context, but only inside that one project. More on GPT building and its retirement is in can I build a custom GPT on my company documents, and on projects in what are ChatGPT Projects.
What it means for a team
The retirement forces a choice that was always there: is this knowledge a fixed tool, or living team context? Fixed workflows become plugins. Living context goes into projects, one per topic, with someone responsible for keeping instructions and files current. Shared projects keep their context within the project. Private projects using default memory can reference conversations outside the project on supported plans, subject to memory settings, but nothing carries the team’s lessons from one shared project to the next.
Where Bombos fits
GPTs froze knowledge, projects keep it per topic, and both keep it inside ChatGPT. Bombos is a plugin that gives every AI tool your team uses one shared memory: team knowledge, learned rules about how you work and what already lives in mail, documents and repos, with traceable answers, in the cloud or on your own server. A correction made in one tool is then available to the others. See also what is AI agent memory.
Related questions
Can I turn my custom GPT into a project?
Not automatically. OpenAI suggests creating a project, pasting the GPT's instructions into Project settings and uploading its reference files. Chats with the GPT cannot be moved into the project, and custom actions and sharing settings do not carry over.
Can I use a custom GPT inside a project?
Yes, for messages in an existing project chat. If the first message of a chat uses a GPT, that conversation starts outside the project.
Which one remembers previous conversations?
Projects can draw on other chats in the same project. GPTs do not use saved memory, custom instructions or previous conversations; each GPT chat starts fresh.
What replaces custom GPTs?
Plugins. GPT creators can use Migrate to plugin under My GPTs: instructions become a skill, knowledge files become reference files and connected apps carry over. Custom actions do not.
