How to Use ChatGPT Projects: Complete 2026 Setup Guide

ChatGPT Projects is the workspace feature that groups related chats, files, and instructions under one folder so the assistant stops forgetting what you uploaded last week. We tested Projects on a Free account and a Plus account over three weeks across web, the iOS app, and the Android app, and the setup decisions that matter are smaller and less obvious than most 2026 guides admit.

#What ChatGPT Projects Are (and What They Aren’t)

A Project is a folder in the ChatGPT sidebar that pins three things together: a list of chats you assign to it, a set of source files you upload once, and a block of instructions that only apply inside that workspace. Open any chat under the Project and ChatGPT has the files and the instructions loaded before you type. Open a chat outside the Project and none of it follows you.

The point is to stop re-explaining context. Our old “Job Hunt Notes” chat grew to 400 messages, and ChatGPT lost the resume PDF around message 80. Projects fixes that by pinning the resume under Sources.

Projects are not the same thing as Siri’s ChatGPT integration on iPhone, which is covered in our Apple Intelligence requirements breakdown. Projects live inside the ChatGPT app and don’t talk to Siri.

#How to Create Your First Project

Open ChatGPT on the web or on the iOS or Android app, look at the left sidebar, and find the Projects section. Click the small + icon next to the section header. A modal asks for a name, and that name is the first decision that actually matters.

The mechanics are easy. On web, the + New Project icon sits above your recent chat list. In the ChatGPT iOS app, tap the sidebar handle in the top-left, then the + next to Projects. On Android the sidebar layout matches iOS.

Launching ChatGPT from an Action Button shortcut drops the chat outside any Project. Open the Project first instead.

The naming step is where most people slip. Calling a Project “Marketing” feels organized for about a week, until you have three marketing Projects and can’t remember which one holds the Q3 campaign assets. According to Tom’s Guide’s tester account, Projects pay off most when each one maps to a single concrete outcome with a deadline. We name ours in the format “Category, Specific Outcome, Date,” so “Client A, Onboarding Doc Rewrite, Q2 2026” beats “Client Work” every time.

Voice entry works inside Projects too. If you trigger ChatGPT by voice through Siri, the voice transcript still posts into whichever chat you have open. When Hey Siri stops working we usually default to the in-app microphone, which writes into the Project context the same way. Voice on Android works the same on devices Apple Intelligence doesn’t reach.

#Adding Sources: Files, Limits, and What Belongs

Inside a Project, click Sources, then Add Sources. Drag in PDFs, Word documents, spreadsheets, code files, or images, and they become available to every chat in the Project. You upload your brand guide once, not at the start of every conversation.

The file ceiling matters and depends on your subscription tier. Free is the most restrictive, and the paid plans (Plus, Go, Pro, Business, and Enterprise) raise the per-project ceiling as you move up. OpenAI’s official Projects documentation recommends checking the current limit inside the Sources panel, since the exact numbers shift over time. Trust the figure your own account shows, not a fixed number from a third-party post.

There’s a second cap that’s easier to miss: the uploader takes files in small batches, not one large drop. In our testing on a Plus account we found that a 12-file upload doesn’t go in at once. We tried to add 12 PDFs to a single project, and the Sources panel accepted them as two separate batches of 6, so plan on splitting larger uploads into a couple of drag-and-drop passes.

What belongs under Sources is everything you’ll reference more than twice in this Project. What doesn’t belong is everything you might reference someday.

Stale files actively hurt because ChatGPT cites them with the same confidence it cites the current version, and you end up with answers built on a draft you replaced months ago. If you keep study materials like Chegg PDFs or academic papers in a Project, replace them whenever the source updates rather than stacking new versions next to old ones.

A common mistake is dumping your entire Drive into one Project. A focused library of five well-chosen documents produces better answers than forty mixed-quality files, because the assistant doesn’t have to guess which document is authoritative.

Uploads are stored on OpenAI’s servers and inherit your workspace’s retention policy. Don’t upload anything you wouldn’t paste into a normal ChatGPT conversation.

#Writing Project-Only Instructions That Actually Override

Open a Project, click the settings icon, and find the Instructions field. This block is the second decision that matters. Whatever you write here applies to every chat inside the Project, and it takes priority over your global custom instructions for the duration.

We tested project-only instructions against global custom instructions by writing “never use bullet points” in the Project’s instruction field. The next three responses inside the Project ignored our global “use bullets when listing more than three items” rule, while a fresh chat outside the Project still bulleted normally. The override is real, not advisory.

What to write: the role ChatGPT is playing, the audience for the output, the format you want answers in, and any terminology rules.

A working example for a code Project: “You are reviewing a React 19 codebase with TypeScript and Tailwind; flag any class component usage; default to functional components with hooks; cite line numbers in feedback.” A working example for a writing Project: “You are editing for a tech-publication audience; reading level is grade 9; strip jargon; keep sentences under 25 words unless cadence demands more.”

What not to write: vague directives like “be helpful and concise.” ChatGPT defaults to that and the instruction wastes characters. The instruction field is also not the place for your reference data; that goes under Sources.

You can also enable project-only memory in the same settings panel. With it on, ChatGPT remembers details from chats inside the Project but doesn’t fold them into your main account-wide memory. We turn this on for any Project involving a client name or anything we wouldn’t want surfacing in an unrelated chat next month.

#Why Should You Move Existing Chats Into a Project?

Because the chat already has the context you need, and moving it forward into a Project is faster than starting from scratch.

In the sidebar, right-click any chat you want to assign and select “Move to project.” Pick the destination Project from the dropdown. The chat moves, and from that point on it inherits the Project’s files and instructions for every new message. Past responses don’t retroactively change, only future ones use the new context.

There’s one exception we hit in testing: chats created by a Custom GPT can’t be moved into Projects. The “Move to project” option simply isn’t available on the right-click menu for GPT-generated chats. If you’ve built up history inside a Custom GPT and want it under a Project, you’ll have to copy the relevant message text into a new Project chat manually. It’s a documented platform limit, not a glitch.

Move chats in batches rather than all at once. Moving an unrelated chat into a Project pollutes that Project’s context, and the assistant will start referencing the moved chat’s topic in unrelated responses.

According to Tom’s Guide tester accounts of organizing chats with Projects, Projects pay off when the chats inside them share a real topic, and dilute when they don’t.

If you find a chat you wish you’d started inside a Project but didn’t, there’s also a “Save response as project source” option on individual messages. It turns a long answer into a reusable Source document, which is the workflow we use when a back-and-forth produces a usable brief or template.

#Projects vs Custom GPTs: Which Should You Use?

Projects and Custom GPTs solve adjacent problems and people frequently use the wrong one.

A Project is a personal workspace for one ongoing effort. Chats, files, and instructions only you can see.

A Custom GPT is a packaged assistant you can share with other people. You set up its instructions, give it knowledge files, optionally enable web browsing or code interpreter, and publish it to the GPT Store or share it via private link.

The right time to build a Custom GPT is when you’ve found a workflow you want to give to your team or your readers as a turnkey tool, like “always-on writing coach for our brand voice” or “FAQ bot trained on our help docs.”

Use Projects when only you’ll work in it. Use Custom GPTs when a teammate or audience needs the same workflow.

One more practical note: Custom GPTs run on the model versions OpenAI binds them to, while Projects use whichever model you select for the chat. If you want a Project’s chats on the newest model the moment it ships, Projects update with the model picker. Custom GPTs sometimes lag.

#Bottom Line

If you use ChatGPT more than a few times a week and you have at least one ongoing effort that drags in the same files or context every session, turn it into a Project today. Name it for one specific outcome with a date, upload only the current reference files (not last quarter’s drafts), and write three or four sentences of project-only instructions.

Skip Projects for one-off questions; those belong in regular chats. Reach for a Custom GPT only when the workflow needs to be shared.

#Frequently Asked Questions