Why do ChatGPT and Copilot forget what my team knows?
ChatGPT and Copilot do remember things, but per person, per project or per tool. Within one shared ChatGPT project you share context, but what a coworker agreed yesterday in their own chat or in Copilot is unknown to your AI today. A shared memory underneath the AI your team already uses solves that: explained once, known to everyone.
Together we map what your AI forgets. Free, and the plan is yours, with or without Bombos. Join the plugin waitlist →
Why do ChatGPT and Copilot forget what your team knows?
ChatGPT and Copilot forget what your team knows because their memory belongs to one person or one place, not to the team. As assistants they are strong: writing, summarizing, looking things up. But what they remember, they remember for you. If your office team tells ChatGPT that a regular customer only takes deliveries on Thursdays, your salesperson's Copilot will not know that tomorrow.
A shared project in ChatGPT helps within that one project: whoever works in it sees the chats, files and instructions of the others. But it stays per project and per tool. A coworker's personal memory cannot be reached from it, and what was learned in ChatGPT does not exist in Copilot or Claude, and the other way around. So outside that one project, every employee explains the same things again.
What do ChatGPT and Copilot remember, and where is the limit?
They remember more than they used to, but every form of memory stops at a person, a project or a tool. This is the state of things according to OpenAI's and Microsoft's own documentation, November 2026.
ChatGPT has a memory that draws on your past chats, memories it saves, your custom instructions and connected apps such as Gmail. You manage it in your own settings, and what is available differs by plan, region and workspace.
Projects in ChatGPT keep the chats, files and instructions for one piece of work together. When you share a project with coworkers, it automatically uses project-only memory: it cannot access a member's memories outside the project, and chats in the project do not see conversations from another project.
Copilot searches your email, chats, calendar and documents through Microsoft Graph, and only surfaces organizational data you have at least view permission for. An agreement that was never written down is not part of it.
Copilot Memory remembers things like your writing style, your work goals and recurring tasks. According to Microsoft Learn, those memories are stored in a hidden folder in the user's own Exchange mailbox, the feature is still in preview, and memory actions do not appear in the Purview audit log.
| What | What it remembers | Where the limit is |
|---|---|---|
| Memory in ChatGPT | Past chats, saved memories, custom instructions and connected apps | Per person: it belongs to your own account |
| Shared project in ChatGPT | The chats, files and instructions in that one project | Per project: no personal memories of members, no other projects |
| Copilot with Microsoft Graph | What is already in email, Teams, calendar and documents | Per person: only what you have access to, and only what was written down |
| Copilot Memory | Writing style, work goals, recurring tasks and details from your chats | Per person: in your own mailbox, not in the audit log |
So the limit is always the same. ChatGPT's memory sits in your account at OpenAI, Copilot's in your own mailbox: what one learned, the other does not see. If your team also uses Claude, Cursor or Grok, each tool adds a memory of its own. How that works per tool is in how to share one AI memory across a team and does ChatGPT Business share memory across a team.
What does that look like in an ordinary small business?
It looks like a mistake nobody made. Take a wholesaler of installation supplies with fifteen people. On Monday, Sanne in the office team hears that a regular customer now gets a discount from fifty units instead of a hundred, and only wants deliveries on Thursdays. She corrects the draft ChatGPT wrote for her. If ChatGPT remembers it, it remembers it for Sanne.
On Tuesday, Mark in sales has Copilot draft a quote for the same customer. Copilot finds the previous quote in SharePoint, with the old discount and a Monday delivery. Mark sends it and the customer calls. On Wednesday a new employee starts, asks her AI how things stand with this customer, and gets last month's version.
With a shared memory underneath, it goes differently. Sanne's correction becomes a rule for the team. When Mark asks Copilot for the quote on Tuesday, Copilot brings in that rule with its source: Sanne, Monday, the customer's email. The new employee gets the same answer from her own AI on Wednesday. Nobody had to write anything down or hand anything over.
What does a shared memory underneath your AI change?
A shared memory changes where knowledge lives: no longer in one employee's account, but underneath the AI of the whole team. Bombos is such a memory layer: one plugin underneath the AI your team already uses.
- One memory for the whole team, whatever AI someone uses: Claude, Claude Code, ChatGPT, Copilot, Grok, Codex, Cursor or an agent of your own.
- A correction becomes a rule. Whoever improves a proposal improves it for everyone, including whoever starts next month.
- It reads what is already there: email, documents, business software and repos. Nobody has to write everything down again first.
- Traceable: with every answer you see where it came from, which email, which document, whose agreement.
- In the cloud, or standalone on your own server.
Bombos does not act on its own. Your own AI writes, looks things up and prepares the work; Bombos gives it what your team knows. Whether your AI can put something into Exact Online, AFAS or other software after your approval depends on the connections your AI itself has. More questions like these are in the guide to AI in your business.
Where do you start?
You start with the work your team explains again most often, and where you can see right away whether the answer is right.
The status question that comes up every day. The answer is spread across email, software and one coworker's head; with a shared memory, everyone's AI gives the same answer, with the source.
The status question that comes up every dayThe recurring customer questions. Whoever answers them best corrects their AI; that correction then applies to the whole team.
The same ten questions that come back every weekThe quote request on Friday afternoon. Whoever is still working does not always know the price agreement a coworker made with this customer; the shared memory does.
Quote requests arriving outside office hoursThe complaint that fades after the first reply. What was promised earlier is in the memory, not only in the chat of whoever promised it.
Complaints that fade after the first reply
If you want to see how it feels first, one person can try the plugin in their own AI before the whole team joins. Or book your half hour: together we map what your AI forgets today. It is free of charge, and the plan is yours, with or without Bombos.
What does it cost and what does it deliver?
The pilot costs €3,000 €2,000 for three months, then €1,000 per month, excluding VAT. You keep your ChatGPT, Copilot or Claude subscription; Bombos goes underneath.
What it delivers is a team that no longer explains the same things every day. A correction counts for everyone, a new hire starts from what the team already knows, and whoever leaves does not take the agreements with them. The full calculation is in what AI costs for a small business.
What does the person do?
The person decides. Your AI prepares the work with what the team knows, and you approve before a message goes to a customer or anything is posted; you set that up in the AI or agent itself. With every proposal you see the source: which email, which document, which earlier agreement.
People also keep the memory up to date, simply by correcting. If a rule no longer holds, you change the proposal and the new rule applies to everyone. Why approval for customer messages should be a fixed boundary is explained in should an AI assistant send email on its own.
When is ChatGPT or Copilot alone enough?
ChatGPT or Copilot alone is enough when there is little to share. If you work alone, or everyone mostly writes their own texts and notes, your AI's personal memory helps right away.
The same applies when everything your team agrees on is already written down in documents everyone can reach, and nobody uses a different AI from the rest. And with fewer than five people, what one knows and another does not is usually manageable without a shared memory.
Frequently asked questions
Do coworkers share memory in ChatGPT Business?
No. ChatGPT's memory belongs to your own account. A shared project does share chats, files and instructions, but according to OpenAI it automatically uses project-only memory and cannot access a member's personal memories.
Does Copilot know what my coworker told their Copilot?
No. According to Microsoft, Copilot Memory is stored in each user's own mailbox. Through Microsoft Graph, Copilot only surfaces organizational data you have at least view permission for, and what a coworker only corrected in their own chat, your Copilot does not see.
Do we need to stop using ChatGPT or Copilot if we use Bombos?
No. Bombos is a plugin underneath the AI your team already uses. ChatGPT and Copilot keep doing what they are good at; with Bombos underneath, they also know what your coworkers know and corrected.
Does it work if one person uses ChatGPT and another Claude or Copilot?
Yes. The memory sits underneath the tools, not inside them. Whether someone uses Claude, Claude Code, ChatGPT, Copilot, Grok, Codex, Cursor or an agent of your own, they read from the same memory.
Can Bombos run on our own server?
Yes. Bombos runs in the cloud or standalone on your own server. With every answer you can still see where it came from, wherever the memory is stored.
Does Bombos do anything in our software or email by itself?
No. Bombos does not act on its own. Your own AI does the work and you approve it. Whether your AI can put something into your software after your approval depends on the connections your AI itself has.
Further reading
- Email into the client file for insurance, bookkeeping and homeowners associationsThe status question that comes back every dayEvery status answer is ready for you to send.
- The shared inbox at bookkeeping firms and homeowners associationsThe same ten questions that come back every weekRecurring questions answered, leaving your team more time.
- Planning for installation and technical maintenanceQuote requests that arrive after hoursEvery request gets a reply before the customer calls elsewhere.
- The shared inbox at bookkeeping firms and homeowners associationsComplaints that stall after the first replyEvery complaint tracked until it is truly resolved.
- Eén taak van begin tot eind: Postbus leeg, mens beslist
- Next guide: What does it cost to automate administrative processes with AI?
- Would you rather talk now? Contact us
What does your AI forget?
- 01You tell us which AI you use and where it keeps starting over.
- 02We show what a shared memory changes there, on your own examples.
- 03You keep the plan, with or without Bombos.
