# ChatGPT Projects: What They Are, How They Work, and Where They Stop

A ChatGPT project is the answer to a small, constant annoyance: starting a new chat and pasting in the same background again. Put the chats, files and instructions for one piece of work in a project, and every new conversation inside it starts with that context already there.

<div class="quick-answer">
<p><strong>Quick answer</strong></p>
<p>Projects in ChatGPT are workspaces that group related chats with shared files, sources and instructions. Create one with New project, add instructions and files in its settings, and start chats inside it. Project memory can be default (can also use your saved memories and other chats) or project-only. A project lives inside ChatGPT, so if you also use Claude or another assistant, keep the context they share in a knowledge base all of them can read.</p>
</div>

Details below are as of November 2026. OpenAI changes this feature often.

## What is a ChatGPT project?

A project is a workspace for ongoing work, with three things in it:

- **Chats.** Every conversation you start in the project, plus any you move into it.
- **Sources.** Files you upload once, and connected sources, that every chat in the project can draw on. Your plan sets how many files.
- **Instructions.** Standing guidance for the project, like "answer as a copy editor for my novel" or "this is a Rails app, keep answers in Ruby." They apply inside the project and override your account-wide custom instructions.

That's the whole idea. Instead of one long chat that eventually runs out of room, you get many short chats that share the same starting point.

## How to use projects in ChatGPT

1. **Create the project.** Click **New project**, in the sidebar or under **Choose project** below the message box, and give it a name you'll recognize in a month.
2. **Write the instructions.** Open the project's settings and say what the work is, who it's for, and how you want answers. Short and specific beats long and vague.
3. **Add the sources.** Upload the documents the work depends on: a brief, a style guide, a spec, the chapters so far.
4. **Pick the memory setting.** Default or project-only (more on that below).
5. **Work inside it.** Start new chats from the project page. Older chats that belong to the work can be dragged into the project from the sidebar.

One habit makes projects much more useful: when a chat reaches a decision, ask for a short summary, save it as a file and add it to the project's sources. The decision becomes something you can find and review later, not just something the model may remember.

## How project memory works

Projects have two memory settings.

**Default memory.** Chats in the project can use your saved memories, and on most plans they can draw on conversations outside the project too. Good when the project is part of your normal life and your general context helps.

**Project-only memory.** ChatGPT only draws on chats inside this project. Nothing from outside comes in, and nothing from the project shapes your other chats. Good for client work, a novel with its own world, or anything where outside context would confuse things.

As of November 2026 you can switch an unshared project between the two in its settings. A shared project always uses project-only memory, so everyone in it works from the same context.

If you've hit the limits of saved memory itself, that's a different problem: see [ChatGPT's memory is full](/blog/chatgpt-memory-full).

## ChatGPT projects vs custom GPTs

They look similar, since both bundle instructions and files, but they're built for different jobs.

| | Project | Custom GPT |
|---|---|---|
| What it is | A workspace for ongoing work | A reusable assistant for one repeatable task |
| Chats | Kept together inside the project | Each chat stands alone |
| Changes over time | Grows as you add chats and sources | Fixed until you edit it |
| Shared with others | Invite people to the project | Share the GPT with others |

OpenAI is moving reusable GPT workflows to plugins. If your GPT only held the background for ongoing work, move its instructions and files into a project instead.

## Claude Projects vs ChatGPT Projects

**Claude** has Claude Projects, built on the same idea: a set of chats that share instructions and files. The differences, as each company documents them in November 2026:

- **Memory.** ChatGPT lets you choose default or project-only memory per project. In Claude, a project's knowledge and instructions apply only to chats inside that project.
- **Sources.** Both take uploaded files. ChatGPT Projects can also use connected sources.
- **Sharing.** ChatGPT Projects can be shared on every plan. Claude documents project sharing on its Team and Enterprise plans.

The bigger point is that the two don't connect. A ChatGPT project and a Claude project on the same work are two copies you keep in sync by hand. [Hjarni vs Claude Projects](/compare/hjarni-vs-claude-projects) goes into the details.

## Does Gemini have projects like ChatGPT?

The closest match is **notebooks** in the Gemini app, which keep sources, instructions and ongoing chats together. Gems, Gemini's reusable instructions, are closer to custom GPTs. Like a ChatGPT project, all of it stays inside one assistant.

## Where projects stop

Projects work well as long as the work stays in ChatGPT. Here's what happens when it doesn't.

Say you're writing a novel. Your ChatGPT project holds the outline, the character sheets and fifty chats of drafting. It knows Ada's sister died in chapter three and that the town is called Marrow. Then you open Claude for a line edit, because you like how it handles prose. Claude knows none of it. You paste in the character sheet, it's out of date by Friday, and now you keep two versions of your own story.

That's the main edge: **a project lives inside one assistant.** Two smaller ones come with it:

- **Uploaded files are snapshots.** Change the outline and you upload it again, or the project quietly works from the old one.
- **Memory isn't something you can read.** Whatever ChatGPT carries over from your chats isn't a list you can open, check and correct line by line.

Writers have the full version of this problem in [keeping your story consistent across Claude and ChatGPT](/blog/keep-your-story-consistent-with-claude-and-chatgpt). Teams have it in [Claude Projects and ChatGPT Projects don't scale to a team](/blog/claude-and-chatgpt-projects-dont-scale-to-a-team).

## Keeping context outside the project

The fix is to keep the context that matters outside any one assistant, as notes you can read and edit, that every assistant can reach.

That's what a knowledge base with a built-in MCP server does. MCP is the open standard that lets ChatGPT, Claude and other assistants connect to outside tools. [How to use MCP with ChatGPT](/blog/how-to-use-mcp-with-chatgpt) explains the mechanism. With Hjarni connected, ask ChatGPT or Claude to search the relevant notes, and to save new decisions as they happen. Both work from the same notes, and so do Claude Code, Codex, Cursor, and anything else that speaks MCP.

Back to the novel. Your ChatGPT project still holds the drafting chats. The facts that outlive any one chat, like characters, places, timeline and style guide, live as Markdown notes. When something changes in the story, you say:

> Hjarni, update Ada's character note: her sister died in chapter three, not chapter five.

The next chat, in this project, another project, or Claude, can look it up. Notes don't fill up like saved memory, and every change is in the note's history if an assistant gets something wrong.

[Hjarni vs ChatGPT Projects](/compare/hjarni-vs-chatgpt-projects) compares the two side by side, and the [ChatGPT setup guide](/docs/chatgpt) gets you connected in a couple of minutes.

## When a project is enough

If your work happens in ChatGPT, alone or with people who also use ChatGPT, a project is all you need. Set it up, keep the instructions honest, and add decisions to its sources as you go.

Reach for a shared knowledge base when the same context has to work in more than one assistant: when people using Claude and ChatGPT need to retrieve and update the same notes, or when you want your context in a form you can read and correct yourself.