ChatGPT Projects: What They Are, How to Use Them, and Projects vs Custom GPTs (2026)

Every long-running task in ChatGPT eventually hits the same wall: you spend the first two messages of every new chat re-explaining the background. Projects fix that. A Project holds the chats, files, and instructions for one effort in a single workspace with its own memory, so ChatGPT already knows what you're working on. It's free, and it's genuinely useful, once you know what it is and, just as importantly, what it isn't.

The confusion worth clearing up first: a Project is not the same as a Custom GPT or custom instructions, even though all three let you set context. A Project is a workspace, a Custom GPT is a shareable tool, and custom instructions are global rules. We'll sort out which is which below, along with how to set a Project up, the file limits, and the memory quirks nobody warns you about.

What are ChatGPT Projects?

A ChatGPT Project is a workspace that groups the chats, files, and instructions for one body of work in one place, with its own memory. OpenAI introduced Projects in December 2024, and the idea is simple: instead of scattering a big effort across a dozen unrelated chats, you keep it together so ChatGPT holds the context.

Each project carries three things. It stores reference files you upload, PDFs, spreadsheets, docs, or pasted text, that every chat in the project can draw on. It holds project instructions that set how ChatGPT should behave inside that workspace, like the tone to use or the libraries your code project relies on. And it has project-only memory, so chats in the project can reference each other and pick up where you left off, without that context bleeding into the rest of your account. If you've read our guide to build a second brain, a Project is the cleanest way to actually do it now.

How to create and use a project

To make one, open Projects in the left sidebar, create a new project, and give it a name. From there you add files, set the instructions, and move any existing chats into it that belong together.

The instructions field is where most of the value hides. Tell the project who ChatGPT should be and how to respond, once, and every chat inside it follows those rules without you retyping them. A book project's instructions might look like this:

You are helping me write a nonfiction book for beginners. Keep the tone plain and warm, use short paragraphs, and avoid jargon unless you define it. When I paste a chapter, critique the structure first, then the line-level writing. Never rewrite a whole chapter unless I ask; suggest edits I can accept or reject.

New chats you start from inside the project inherit its files, instructions, and memory automatically, which is the whole point.

Projects vs Custom GPTs vs custom instructions

The single most common question about Projects is how they differ from Custom GPTs, and the short answer is that a Project is a workspace while a Custom GPT is a tool. A Project organizes an ongoing effort and keeps it private. A Custom GPT packages a repeatable task into an assistant you can reuse and share, by link or through the GPT Store. Custom instructions are the third thing again: global rules that apply to every chat in your account, not to one project or one GPT.

The practical rule: reach for a Project when you're working through something over time and want the context to accumulate, a Custom GPT when you want the same kind of output every time or need to hand the tool to someone else, and custom instructions for the defaults that should apply everywhere. They also stack. Power users set a base voice in a Custom GPT, layer project-specific context in a Project, and let global memory hold the background facts.

What people use Projects for

Projects suit any effort that runs across many sessions and builds on itself. The pattern is repeated, evolving work rather than one-off questions.

Common ones include a piece of long-form writing, where the project holds your outline, style, and research; a research topic, where uploaded papers and notes stay together; a job search, with your CV, target roles, and tailored applications in one place; a coding project scoped to a language and its libraries; and ongoing planning like a launch or a house move. The test is whether you'll come back to it. If a task is one message and done, a Project is overkill; if you'll return to it next week, it earns its place.

Getting the most out of Projects

A few habits separate a project that helps from one that just holds clutter. Scope each project to one real effort, not one topic or one day, so the memory and files stay relevant to what you're actually doing. Write the instructions the way you'd brief a new hire: what the work is, how you want responses, and what to avoid, since a clear "don't" often does more than a vague "do." Keep the file list short and current, because three to five strong references beat twenty weak ones that just dilute the context. And name projects so you'll still recognise them in a month; a sidebar full of "Untitled" defeats the point of organising anything.

The honest limits

Projects are useful, but the memory isn't as seamless as the pitch suggests, and it's worth going in knowing where it slips. In hands-on use, the project memory sometimes fails to pull in relevant details from other chats in the same project, so you still end up copying context across occasionally. New chats you create don't always show up in the project until you refresh the page. And files ChatGPT generates for you don't automatically save back into the project as reference files; you have to add them yourself.

Two more traps catch people. You can't share files across projects, so the same reference has to be uploaded into each one that needs it, or moved into a Custom GPT that several projects lean on. And more files isn't better: a handful of focused, current documents beats dozens of mediocre ones, which just dilute the context and slow things down. Prune the stale stuff every so often.

What this post does not cover

This is a plain-English guide to the ChatGPT Projects feature, not a full manual or a promise of exactly how it will behave on your account. It doesn't cover the developer API or team administration in depth, and it isn't legal or professional advice. Features, file limits, and which plans include what change often, so confirm the current details in ChatGPT or OpenAI's help pages. For a different way to work on a single document or script, ChatGPT Canvas is the companion feature worth knowing.

Sources

1. OpenAI Help Center: Projects in ChatGPT
2. OpenAI Academy: Using projects in ChatGPT
3. ChatGPT Learn: Projects, chats, and tasks