ChatGPT Projects and Folders: Organise Your AI Work Like a Professional

What ChatGPT Projects Actually Are

ChatGPT Projects are containers that give every chat inside them the same starting context: the same files, the same instructions, the same history. A plain chat starts from nothing every time. A project chat starts from everything you have already put in.

The word “folder” undersells it, and that is why so many people set one up, drag two chats in, and conclude it changed nothing. A folder sorts. A project informs. Nothing improves until you actually load it with the context you keep retyping.

Same output quality, roughly a tenth of the typing. Multiply that across a week of client comms and the saving stops being trivial — most professionals who set this up properly report the first real benefit lands in about three days, once the project has enough history to draw on.

Three things carry across every conversation you open inside a project, and it is worth being precise about them because each one solves a different daily irritation:

• Your files, so you stop attaching the same brief to five separate chats.
• Your instructions, so you stop retyping “keep it formal, British English, no bullet points”.
• Your earlier chats, so ChatGPT can refer back to the decision you made on Tuesday without you summarising it again.

Projects Versus Loose Chats

Use a plain chat for anything you will never think about again — a quick rewrite, a definition, a one-off calculation. Reach for one of your ChatGPT Projects the moment a piece of work has a second session in its future. That is the whole test, and it is deliberately low effort, because the cost of creating a project you end up not needing is close to zero.

The people who get the least out of ChatGPT Projects are the ones who build an elaborate structure first and find work for it afterwards. Let the repeating work tell you what projects to make.

Every Plan Gets ChatGPT Projects Now

ChatGPT Projects are available on every subscription type, free and paid alike. What changes between plans is not access — it is how many files you can park in each project, and how many people you can invite into it.

According to OpenAI’s Help Center documentation on Projects, the number of ChatGPT Projects you can create is unlimited on every tier, and no more than ten files can be uploaded at once regardless of plan:

Sharing used to be the genuine upgrade reason. It isn’t any more — since October 2025, Free, Go, Plus and Pro users can all invite people into a project too. OpenAI’s own Free tier documentation confirms the free plan carries file uploads, web search and data analysis alongside it.

So if you have been putting off organising your work because you assumed it needed a subscription, start today on whatever plan you already have. If you find yourself hitting the five-file ceiling every week, that is a real upgrade signal — and the honest breakdown of Free versus Plus covers whether the rest of the package earns its money for your kind of work.

Set Up Your First Project

Setting up ChatGPT Projects takes about five minutes, and the order matters more than the speed. We call this the Seal-First Method at PromptPeakAI: decide how sealed the project is before you put anything in it, because that is the one choice ChatGPT never lets you revisit.

Every other setting here is editable forever. Rename the project, swap the files, rewrite the instructions, change the colour — all reversible on a Tuesday afternoon. The memory mode is not, which is exactly why it belongs at step two rather than as an afterthought.

Click New project in the left sidebar, then give it a name you would recognise in a hurry six months from now. “Acme Corp” beats “Client Stuff”. You can also pick an icon and colour, which sounds cosmetic until you have nine projects and are scanning for one mid-call.

The New Project dialog asks whether this project uses default memory or project-only memory. Answer it deliberately rather than clicking past it.

Client work, HR casework, board material, anything covered by a confidentiality clause: choose project-only. General personal work where you actively want ChatGPT to remember your writing style from elsewhere in the account: default is the better pick.

Upload the two or three documents you keep pasting from — the brand guide, the SOP, the signed scope of work. PDFs, spreadsheets, documents and images all work, and you can paste raw text in as a source too.

Resist the urge to dump everything you own. A project with four sharp files gives better answers than one with twenty-five overlapping drafts, because ChatGPT has no reliable way to know which of two similar documents is the current one. On Free you only get five slots anyway, which turns out to be a useful constraint.

Open the three-dot menu at the top right of the project and choose Project settings. This is where the standing rules live. They apply to every chat in the project and they override the global custom instructions you have set at account level.

That override is worth pausing on. If your account-wide instructions tell ChatGPT to be brief and punchy, and this project needs formal client prose, the project instructions win — which is exactly what you want, but it does mean thin project instructions can quietly disable rules you rely on everywhere else.

You are supporting me on [project or client name].

My role: [job title] at [company type].
Audience for most outputs: [who reads this work].

Always:
- Use the tone and terminology in the project files.
- Ask me for missing facts instead of inventing them.
- Default to [British/American] English and [format, e.g. short paragraphs with bolded actions].

Never:
- Add statistics, dates or names that are not in the files or my message.
- Open an email with "I hope this finds you well".

The Memory Choice You Cannot Undo

Memory inside ChatGPT Projects decides what a chat is allowed to draw on. Choose it at creation, because ChatGPT does not let you switch an existing project from default to project-only afterwards — you would have to build a new project and move the chats across.

This is the single most consequential thing about ChatGPT Projects, and it is also the thing competing guides get wrong most often. Plenty of them describe a settings menu where you flip memory modes on an existing project. That menu does not exist.

The red row in that left column is the one that surprises people. On Free, Go, Plus, Pro and Business, a default-memory project chat can still reach conversations outside the project — and be reached by them. That is the “context bleed” complaint you see in forums, and it is a setting rather than a bug.

Two further wrinkles are worth knowing before you commit. Sharing a project with anyone switches its memory to project-only automatically and permanently — sensible, but not reversible. And on Enterprise and Edu workspaces, project chats stay contained under both modes, so the choice matters far less there than it does on an individual plan.

One habit that pays off regardless of mode: when ChatGPT produces something you will reuse — a decision note, an approved summary, a final draft — save that response into the project sources from the message menu. It becomes reference material for every later chat instead of something you have to go hunting for.

Three Workspace Blueprints You Can Copy

Most people over-think the architecture. Three or four ChatGPT Projects covering your actual recurring work will outperform a beautifully designed system of fifteen that you abandon in a fortnight.

Before uploading anything, run the same test you would apply to any tool that stores company documents — the judgement covered in our guide to using ChatGPT safely at work:

The Marketing Command Centre

Files: brand voice guide, audience personas, last quarter’s best-performing posts. Instructions: your tone rules and banned words. This is the project that stops you pasting three hundred words of brand background into every copywriting chat.

Using the brand voice guide and persona documents in this project, draft a 200-word LinkedIn post announcing [feature or news].

Audience: [persona name from the files].
Angle: [the one point you want remembered].
Format: short paragraphs, one concrete example, no hashtags.

Do not use any claim or number that is not in my message or the project files.

The HR Policy Hub

Files: the employee handbook, remote work policy, the current benefits summary. Set this one to project-only memory without thinking twice — employee matters have no business drifting into your other work, and HR is the function where a stray cross-reference does the most damage.

Using the policy documents in this project, write a reply to an employee who has asked about [topic].

Tone: warm, plain English, no policy jargon.
Length: under 150 words.
Structure: direct answer first, then the relevant policy point, then who to contact for exceptions.

If the policy documents do not cover this, say so instead of guessing.

The Client Account Room

One project per significant client, each holding the scope of work, the kickoff notes and the latest status. Consultants and agency staff feel this one immediately — it is the difference between five clients living in one sidebar and five clients living in five sealed rooms.

Review the scope of work in this project and draft a short status email to the client lead.

Cover: what shipped since [date], what is in progress, anything I need from them.
Tone: confident, no apologising for normal timelines.
Length: under 180 words, no bullet lists.

Flag anything in the scope that looks at risk, but do not invent progress I have not told you about.

If you would rather not write these from scratch, the free AI Work Templates library has instruction sets you can drop straight into a project. The same architecture also transfers between tools — Claude Projects work on the same principle if your workplace runs something other than ChatGPT.

Projects or Custom GPTs?

This is the question that stalls people for weeks, and the answer is simpler than the debate suggests. ChatGPT Projects are for work that keeps going. Custom GPTs are for a task you repeat identically.

The clue is in who the output is for. A project serves you and the people inside it, evolving as the work does. A Custom GPT serves anyone who opens it and should behave the same way on the fiftieth use as the first.

When to Build Each

A campaign, a client, a hiring round, a report cycle — all projects. A tool that converts messy notes into your company’s meeting-minutes format, used by eleven people who should all get identical output — that is a Custom GPT. Plenty of professionals run both, and there is no penalty for doing so.

Two practical wrinkles catch people here:

• A chat you started with a Custom GPT cannot be moved into a project afterwards. If you want it in the project, start it there.
• You can call a Custom GPT from inside an existing project chat — just not as the opening message, which would pull the conversation out of the project entirely.

Where ChatGPT Projects Still Fall Short

Three limits are worth knowing before you build a system around them, because each one tends to surface after you have already committed:

• There is no subfolder layer. ChatGPT Projects sit in a flat list. If one gets crowded, use a naming convention for the chats inside it or split the project in two.
• Temporary chats cannot join a project, and neither can chats begun with a Custom GPT.
• Connected Google Drive files are searched, not pre-synced inside a project, so treat linked sources as a lookup rather than dependable background knowledge.

And the honest one, which no amount of setup fixes: ChatGPT Projects make the tool better informed, not better judged. It will still misread which of two conflicting documents is current. It will still write with total confidence about a policy it half-understood.

Most experienced users land on the same working rule — a project raises the floor on quality and speed, but the ceiling still depends on you reading the output before your name goes on it. That is not a flaw to design around. It is just the job.

Common Problems and Fixes

Underneath all of this sits ChatGPT’s account-level memory, which keeps working across your chats whether or not you use ChatGPT Projects. The two systems interact more than most people expect, and understanding the account-level layer makes the project-level choice much easier to reason about.

Pick the one piece of work that has cost you the most re-explaining this month. Build a project for it this afternoon, choose the memory mode deliberately, and give it a week before you add a second. The second one usually takes four minutes, because by then you know exactly what belongs in it — much the same way working with a spreadsheet in ChatGPT gets faster once the structure stops being a decision.