ChatGPT Projects: What They Are and When to Use One

If you have ever pasted the same background into five different chats, Projects exist for you.

Projects keep related chats, files, and instructions together. ChatGPT can use that context to support ongoing work, such as writing, research, or planning.

Source: OpenAI, Projects in ChatGPT, checked November 1, 2026. The page had been updated two days before that check.

That is the whole idea. A project is a container that holds three things — conversations, uploaded files, and standing instructions — and applies all three to anything you do inside it.

The problem it actually solves

Ordinary ChatGPT is stateless in a way that bites you slowly. Each new chat starts from nothing unless memory carries something over, and memory is a summary rather than a document. So you re-explain the same context, and the model’s answer drifts depending on how much you remembered to repeat.

A project changes that in one specific way: anything you put in the project applies to every chat inside it. Upload the brief once, set the tone once, and every conversation in that project starts already knowing.

Three patterns where this matters:

None of these are one-off questions. That is the dividing line: if you will come back to it, a project helps. If it is a single question, a normal chat is faster.

How to create one

1. In the sidebar, select New project.
2. Give it a name, then pick an icon and colour.

That is it. There is no setup wizard and no configuration step you have to complete before using it.

You can create as many projects as you want. OpenAI states that users can create an unlimited number of projects, so the limit is not how many you make — it is whether you can find anything in a sidebar full of them.

Naming matters more than it sounds. “Project 3” tells you nothing in three weeks. Name it after the thing it produces: “Q4 pricing page”, “Spanish B1”, “Kitchen renovation”.

Three things you can put inside

Chats

Start a new conversation inside the project, or move an existing one in: drag it onto the project, or open the chat menu and select Move to project.

Once moved, the chat inherits the project’s instructions and file context. That inheritance is the point — and it is also a trap if you move a chat that was written for a different purpose. A conversation drafted under one set of instructions will now be read under another.

One limitation to know: chats created with a GPT cannot be moved into a project. If you built a conversation inside a custom GPT and want it in a project, you will need to start over.

You can move a chat back out with Remove from project. It returns to ordinary history.

Files

You can upload PDFs, spreadsheets, documents, and images, or paste text directly. Files can be previewed, downloaded, and deleted from the project’s source list.

Two practical details:

• Name collisions are handled explicitly. If you upload a file with the same name as an existing one, you choose Upload anyway or Skip.
• You can save a chat reply as a project source. Open the message menu and select Add to project sources. This is underused — it turns a good answer into reusable context without copying text into a document.

Google Drive folders and Slack channels can be linked, but with a caveat: the Google Drive app does not sync inside a project. It searches and accesses on demand rather than pre-loading, so “connected” does not mean “available offline”.

Project instructions

This is the feature that does the most work and gets configured the least. Find it under the ••• menu → Project settings.

OpenAI’s own example:

Act like my marketing mentor. Be concise. Use bullet points. Ask clarifying questions.

And the rule that surprises people:

Project instructions apply only within that project and override your global custom instructions.

Source: same page, checked November 1, 2026.

Override, not merge. If your global instructions say “explain in detail” and the project says “be concise”, the project wins inside the project. That is usually what you want. It is confusing when you forget you set it.

Limits by plan

Rate limits follow your subscription level rather than the project itself.

The file cap is the one that catches people. Five files on Free sounds like enough until you upload a syllabus, three lecture PDFs, and a reading list. When you hit the cap, OpenAI suggests deleting old files, merging them into one document, or splitting the work across projects.

Splitting is usually the right answer. A project holding 40 files about four different topics is harder to work with than four projects holding ten each.

What works inside a project, and what does not

Most tools behave the same inside a project. A few do not, and the exceptions matter before you build a workflow around them.

The Study mode exclusion is the surprising one. If you built a project for a course and planned to use the quiz-style mode inside it, that combination is not supported. You would run Study mode outside the project and paste results in.

A worked example: the weekly report

A complete setup, small enough to copy.

The project: named “Friday ops report”, not “Operations”.

Files: last quarter’s reports — two or three is enough — plus the spreadsheet template you fill in each week.

The instruction:

You are helping me write a weekly operations report. Structure it as: what changed, what is at risk, what needs a decision. Keep it under 400 words. Do not invent numbers — if a figure is missing from the files, say so and ask me.

The first conversation: “Here are this week’s numbers. Draft the report.”

Why each part matters. The name tells you what the project produces, so you can find it in three weeks. The old reports give the model your actual format instead of an invented one. The instruction fixes structure and length, which is what stops the output becoming three pages of preamble. And the last line — do not invent numbers — is the one that saves you from a confident wrong figure.

What you stop doing: pasting the format every week, re-explaining what the report is for, and trimming the length after the fact.

Sharing changes the memory rules

This is the part most guides skip, and it matters if you work with anyone else.

A project can use one of two memory modes:

And then the rule:

Shared projects are automatically set to project-only memory and cannot be switched to default memory.

Source: same page, checked November 1, 2026.

Read the consequences carefully:

• Anyone you share with cannot see your context from outside the project. Their chats, their instructions, their memories — none of it crosses in.
• You cannot switch a shared project back to default memory. The isolation is not optional.
• Switching a project to project-only memory removes its information from your memory outside the project. The isolation works in both directions.
• There is no global setting. Each project’s memory mode is set individually, and there is no list where you can review them all.

For a solo project, this is mostly invisible. For a shared one, it means the project is genuinely a separate room — which is the point, and also why a shared project feels less “smart” than your personal chats. It does not know what you know.

What collaborators can see

Members see the project’s chats, files, and member list. Each chat shows the creator’s avatar, so authorship is visible.

Everyone in the project can view and download the project files. Edit access allows members to update instructions, upload or remove files, and invite others — but not remove existing members.

Sharing a single chat link does not grant access to the rest of the project. If someone receives a chat shared from a project, they need to join the project before they can view it.

What happens to your data

Deleting a project removes its chats, files, and instructions from OpenAI systems within 30 days, except where legal or security retention applies.

For training:

• Business, Enterprise, Edu, and Healthcare: not used for training by default.
• Personal plans: a shared project’s data may be used for training only if the owner and every contributor have “Improve the model for everyone” enabled.

Projects inherit your workspace’s retention period and data residency. You do not configure those per project.

Projects versus custom GPTs

They look similar and are not.

GPTs are single player… shared projects are multi-player.

Source: same page, checked November 1, 2026.

A GPT is a packaged assistant you publish. A project is a workspace you and other people work inside. If you want to hand someone a tool, build a GPT. If you want to work alongside someone on the same material, make a project.

When a project is the wrong choice

Three cases where a normal chat is better:

One question, one answer. Creating a project to ask “what does this word mean” is overhead with no payoff.

You need a clean answer, not a contextual one. Project context biases every reply. If you want an unbiased take on something, ask outside the project.

You are testing an idea. Exploration benefits from starting fresh. A project anchors you to the framing you set on day one.

A short practice sequence

1. Pick one recurring task you already do — a weekly report, a course, a side project.
2. Create a project named after the output, not the topic.
3. Upload the two or three files you keep re-pasting.
4. Write one project instruction: what role, what format, what to avoid.
5. Run three conversations inside it and notice what you no longer have to explain.

If step five produces nothing, the task probably did not need a project. That is a useful result too.

Questions beginners often ask

Are projects free?

Yes. OpenAI states that access to projects is included with every qualifying ChatGPT plan. The limits differ by plan, but the feature itself is not a paid add-on.

How many projects can I make?

Unlimited. The constraint is your ability to find things, not a cap.

Can I use a project on my phone?

Yes, projects work on web and mobile, and a conversation started on one continues on the other.

Why can’t I move my chat into a project?

If the chat was created with a GPT, it cannot be moved. Start a new conversation inside the project instead.

Does deleting a project delete my files everywhere?

It removes them from the project and from OpenAI systems within 30 days. Copies you downloaded are yours and stay wherever you put them.

Why does a shared project feel dumber than my normal chats?

Because shared projects use project-only memory. It cannot see your saved memories or your conversations outside that project. That is a privacy feature working as designed.

Can I move a project between accounts?

No. A project belongs to the account that created it. If you need the material elsewhere, download the files and rebuild it.

Do projects count toward my message limit?

Project conversations use the same models and tools as ordinary chats, so they count the same way. Creating a project does not buy extra usage.

What happens to a shared project if the owner deletes their account?

The project goes with the account. If other people depend on it, export the files before that happens.

When to ask a person

Ask before uploading material you do not own — a client’s files, a colleague’s draft, a licensed document. Ask your workspace admin before sharing a project externally, since workspace settings can restrict sharing. Ask before putting anything sensitive in a project you share, because every member can download the files.

If you want the wider picture on what these tools do with your data, the privacy guide covers the settings that apply across projects and chats. And if you are still deciding whether to pay for a plan at all, the Plus decision guide compares what each tier actually unlocks.