Projects, Custom GPTs, Gems or Copilot: which one holds team knowledge?

Short answer: Projects, Custom GPTs and Gems work well for one person or a defined piece of work. Microsoft 365 Copilot is good at finding information in your Microsoft environment. But none of them is built to keep a whole team's shared knowledge current, approved and identical across every tool. That takes one shared source that all the tools read from.

Most teams have already tried. Someone built a Custom GPT with the price list, someone else has a Claude project with the sales process, and leadership bought Copilot licences. The question is which of these tools can actually carry company knowledge once more than one person relies on it.

How we compare

We look at four things that decide whether a tool works for a team, not just for one person:

• Where knowledge lives. Is it a copy someone uploaded, or does the tool read from the source?
• What happens when something changes. Do answers update on their own, or does someone have to remember to replace files?
• Who sees what. Do permissions follow the person asking, or does everyone see the same thing?
• Whether it works in other tools. Can the knowledge be used in more AI tools than the one it was built in?

The comparison in one table

These tools change quickly. The table describes how they work at their core, not every setting in every plan.

ChatGPT Projects and Custom GPTs

Projects bring chats, files and instructions together for one piece of work. Custom GPTs are your own versions of ChatGPT with instructions and knowledge files someone has uploaded. Both are quick to set up and work very well for one person or a defined project.

The problem starts when they become the team's knowledge base. The files are copies. When the price list changes, someone has to remember to replace it in every GPT and every project. Often only the person who built the GPT knows it exists, and ChatGPT memory is personal and cannot be reviewed by the team.

Read more about how ChatGPT can connect to one shared knowledge source.

Claude Projects

Claude Projects work in a similar way: files and instructions per project, which can be shared with colleagues on Team and Enterprise. For a defined client engagement, that is a good setup.

But company knowledge is needed in almost every project. The price list, tone of voice and sales process get uploaded again and again, and each upload becomes its own copy. When the original changes, the copies drift apart, and Claude answers correctly based on a file that no longer applies. How to give Claude your whole company's knowledge.

Gemini Gems

Gems are Google's equivalent: your own versions of Gemini with instructions and attached files, which can be shared in Google Workspace. The same strength and the same limitation apply. A Gem knows what someone gave it, but it does not know when that became outdated.

Microsoft 365 Copilot

Copilot is different from the others. It does not upload copies. It searches SharePoint, OneDrive, Teams and email directly, using the user's permissions. That solves the problem of stale copies but creates another one: Copilot finds everything, including drafts, old versions and folders that were shared too widely years ago.

Copilot searches, but it does not decide what applies. When three versions of the same price list exist, the answer can be based on the wrong one, or on all three. How Copilot can access what actually applies.

The shared problem: copies instead of one source

All five tools are built to help the person using them. None of them is built to keep a team's shared knowledge current. The result is the same whatever the tool: colleagues get different answers to the same question, and nobody knows which one is right. We cover that problem in Same question, different answers.

The tools are not the problem. The problem is that every tool has its own copy of the truth.

When are the built-in tools enough?

Often they are. You do not need anything more if:

• One person uses the knowledge, or a small team that talks every day.
• The material belongs to a defined piece of work with a clear end.
• The information rarely changes, or changes in a way everyone notices.

When do you need one shared knowledge source?

It is time to rethink when any of these are true:

• More than a handful of people use AI in customer-facing work.
• You have several GPTs or projects with the same documents in them.
• Pricing, terms or procedures change several times a year.
• You use more than one AI tool, for example Copilot and ChatGPT.
• You want AI agents to carry out work, not just answer questions.

Then you need one shared source where every piece of knowledge has an owner, a date and an approved version, and that every tool reads from. That is what a Company Brain does. It connects to the tools through MCP, the standard we explain in What is MCP?

Frequently asked questions

What is the difference between ChatGPT Projects and Custom GPTs?

Projects are a workspace for one piece of work, with chats, files and instructions in one place. A Custom GPT is your own version of ChatGPT that several people can use. Both rely on files someone uploaded that must be updated by hand.

Can Copilot replace a knowledge base?

Copilot is good at finding information in Microsoft 365, but it does not decide which version applies and cannot capture knowledge that is not written down. It needs a verified foundation to answer correctly.

Can we use several AI tools with the same knowledge?

Yes, if the knowledge sits in one shared source that the tools read from through MCP. It is then maintained in one place and used in Copilot, ChatGPT, Claude and AI agents.

Do we need to stop using our GPTs and projects?

No. Use them for what they do well, such as a specific piece of work. Let the shared source provide what applies to the whole company.