ChatGPT Projects vs Custom GPTs: which one holds team knowledge?

Picture a marketing lead partway through the quarter. She runs a Project called Q4 Campaign and a Custom GPT called Campaign Writer, and the same product one pager is uploaded into both. Ask her why both exist and she has to pause, because she built the second one after she could no longer remember what the first one was for.

The short version: a ChatGPT Project is a workspace that groups your own chats, files and instructions around ongoing work. A Custom GPT is a packaged assistant, with instructions and knowledge files, that you share so others can run a repeated task. Neither one reviews its knowledge or shares it beyond ChatGPT, which is the job Company Brain does.

Her confusion is common, and the naming carries some blame. The two features live side by side in ChatGPT, both accept uploaded files, both take custom instructions, and both make ChatGPT answer with more context than a blank chat gives you. The differences only appear when you ask who the thing is for.

What a ChatGPT Project is

A Project is a folder with powers. You create it in your own ChatGPT, drop in files, and write instructions that apply to every chat inside it. From then on, each new chat you open in the Project starts with those files and instructions already in scope, so you stop re-uploading the deck and re-explaining the goal each morning.

The center of gravity is your own work. A Project collects one person's continuing effort on one topic: the campaign you are running, the hiring round you are managing, the report you are writing over three weeks. Chats accumulate inside it, each starting from the same files and instructions. It behaves like a desk you have set up for one piece of work, with the reference material already laid out.

What a Custom GPT is

A Custom GPT is a packaged assistant. You write instructions, attach knowledge files, give it a name, and share a link. The team opens the link and chats with the assistant you configured. The builder does the thinking once and hands the finished behavior to other people.

That makes a Custom GPT a small product. It shines on a repeated, well-defined task: a proposal writer that always follows your structure, a support drafter that always uses your tone, a job ad generator that knows your format. I have written about where Custom GPTs work for team knowledge and where they break, and the honest summary is that for task knowledge they are genuinely good.

Which one should you use?

One question settles nearly every case: is this my ongoing work, or a repeated task I want to hand to others?

If it is your ongoing work, use a Project. The campaign, the fundraise, the product launch: anything where you will open many chats over weeks and want them to share files and instructions. You are the main user, and the value is continuity.

If it is a repeated task other people will run, build a Custom GPT. The value is consistency: everyone who opens it gets the same rules and the same structure, whether or not they would have thought to set that up themselves.

The marketing lead's setup sorts itself out with that question in about a minute. Q4 Campaign is her ongoing work, so the Project stays hers. Campaign Writer is a task two colleagues will run, so it stays a GPT. Both survive the sorting.

The blind spot both of them share

Underneath the different shapes, both features hold knowledge the same way: as files uploaded into a container, one container at a time.

That design has three consequences worth seeing clearly.

Every container keeps its own copy. The one pager in the Project and the one pager in the GPT are two separate files. Update one and the other keeps answering from the old version. Our marketing lead's Custom GPT is by now quietly two revisions behind her own Project.

Nothing gets reviewed. Whatever the builder uploaded or typed is what the container treats as true. There is no step where anyone confirms the pricing sheet is current or the positioning line is the approved one. A draft and a decision look identical inside a knowledge file.

Both go stale on the same schedule. Company facts change: pricing, product names, policies, positioning. Every Project and every GPT holding a copy has to be found and updated by hand, and there is no list of which containers hold what. Staleness arrives silently and gets discovered by a reader, often an external one.

There is a fourth limit that follows from where these features live: all of it stays inside ChatGPT. The colleague who prefers Claude and the developer working in Cursor get none of it. ChatGPT's own memory does not close the gap either, because memory is personal to each account and picks up whatever it happens to notice. I covered that in what ChatGPT memory actually remembers.

When Projects and Custom GPTs are enough

Plenty of teams need nothing more than this, and it is worth saying plainly.

If your team is small, works mostly in ChatGPT, and the knowledge involved changes rarely, a few Projects and one or two Custom GPTs will serve you well. A five person team whose pricing changes once a year can update two files once a year and move on.

Watch for three signals that you have outgrown the setup. The same company fact lives in three or more containers. A fact changed, somebody forgot one of the copies, and a customer or candidate saw the old version. Or teammates now work in tools beyond ChatGPT and keep asking for the files.

What a team knowledge system does differently

Company Brain holds company facts in one reviewed place and delivers them through MCP to ChatGPT, Claude, Copilot and Cursor. Facts go in through a proposal and review step, so the shared set stays deliberate rather than accumulated. The whole team connects through one shared MCP address for the company, so nobody maintains a private copy. There is a free plan, and it stays free.

Your Projects and Custom GPTs keep their jobs in that picture. Task instructions stay in the GPT, working files stay in the Project, and the company's facts arrive from one source underneath both. When pricing changes, the change is one reviewed edit, and every container and every tool sees it. For a side by side look at where each approach wins, see our comparison of Custom GPTs and a shared company context.

The marketing lead keeps her Project and her GPT in that picture. The one pager ends up living in neither.