ChatGPT and Claude can both remember things between conversations, which makes each person's assistant more useful over time. But that memory belongs to one user or one project. It is not shared, curated or owned by the business. Company memory has to live somewhere else: in shared files every person's AI reads.
Key takeaways
- ChatGPT's memory keeps saved details and draws on a user's chat history; Claude keeps a separate memory for each project.
- Both are personal or project-level memory, built from what one person told their assistant.
- According to Microsoft and LinkedIn, 78% of AI users bring their own tools, so much of this memory sits in accounts the business does not control.
- According to Panopto's 2018 survey, 42% of institutional knowledge is unique to the individual, and AI memory can deepen that concentration.
- The fix is shared, owned knowledge underneath, with personal memory used for personal preferences.
What do ChatGPT and Claude memory actually do?
Both let an assistant carry information from one conversation to the next. ChatGPT keeps "saved memories" a user asks it to remember and can draw on that user's past chats. Claude creates a separate memory for each project. In both, users can view or switch memory off, and temporary or incognito chats are not saved.
The details matter, and they change, so check each vendor's current documentation. As of late 2026, the published position looked like this:
| ChatGPT | Claude | |
|---|---|---|
| What it remembers | Saved memories, plus insights from the user's chat history | A memory summary per project |
| Whose memory | The individual user's | The user's, within each project |
| User controls | Turn saved memories or chat history off; Temporary Chat | View and edit memory; incognito chats |
| Admin controls | Varies by plan | Enterprise admins can turn memory off |
OpenAI introduced memory in February 2024 and expanded it in April 2025 to reference past conversations. Anthropic launched memory for Team and Enterprise plans in September 2025 and extended it to Pro and Max in October 2025.
Why isn't personal memory company memory?
Because it is built from one person's conversations, held in their account or project, and nobody curates it. Two people's assistants can remember contradictory things about the same product. Neither memory has an owner checking it, a source for each fact, or a rule about what is approved to say publicly.
Picture two salespeople. One told their assistant last month that the starter plan includes onboarding. The other told theirs, correctly, that onboarding is extra. Both assistants now remember different facts and will apply them confidently in every follow-up.
Scale makes it worse. According to Microsoft and LinkedIn's 2024 Work Trend Index, 78% of people who use AI at work bring their own tools. A good share of the business's working knowledge is therefore accumulating in personal memories the company cannot see, check or keep.
What happens to AI memory when someone leaves?
It usually leaves with them. Memory in a personal account was never the company's. Memory in a company account belongs to a user who no longer works there, and it is rarely organised for anyone else to use. Either way, whatever that person's assistant learned about how the business wins is effectively lost.
This deepens an old problem. According to Panopto's 2018 Workplace Knowledge and Productivity Report, 42% of institutional knowledge is unique to the individual who holds it. AI memory adds a new place for that individual knowledge to pile up, out of sight. The guide to offboarding in the AI era covers how to recover what you can.
How should personal memory and shared knowledge work together?
Use personal memory for personal things: how someone likes their drafts formatted, what they are working on, their preferences. Keep business facts in shared, owned files that every assistant reads, such as product facts, pricing rules, objections and the brand voice. Then the assistant is personal in style and consistent in substance.
A useful rule of thumb is to ask whether a fact would be true for everyone. "I prefer bullet points" belongs in personal memory. "Onboarding is not included in the starter plan" belongs in shared knowledge, with an owner and a source.
That shared layer is what a company brain holds. When someone teaches their assistant something genuinely useful about the business, it should be written back to the shared files, so the whole team benefits and it outlasts the person.
What memory settings should a business choose?
Put staff on business plans, decide whether memory is on, and write a simple rule: personal preferences may live in memory; business facts live in shared files. Where available, use project-level memory for client or initiative work, and review what assistants remember periodically. Settings vary by plan, so confirm current options.
A short checklist:
- Business plans, not personal accounts, so memory sits under company control.
- A clear rule about what belongs in memory and what belongs in shared files.
- Projects for bounded work, so a client's details stay with that client's project.
- Shared knowledge for facts, maintained by named owners.
- A habit of writing back useful learning to the shared files.
The guide to Claude Projects for teams covers project structure in more detail.
Skein builds the shared layer as the shared knowledge layer of a Company OS, so useful learning moves from individual memory into company files. That suits a team where several people rely on AI daily. A solo user can reasonably rely on personal memory alone.
Frequently asked questions
Does ChatGPT memory work for teams?
ChatGPT memory works per user. It keeps details a user asks it to remember and can draw on that user's chat history, but it does not share memory across a team. For facts everyone should use consistently, keep shared, owned files that every person's assistant reads.
How does Claude memory work?
Anthropic's Claude creates a separate memory for each project, so work on different projects stays apart. Users can view and edit what Claude remembers, and incognito chats are not saved. Enterprise administrators can turn memory off for the organisation. Check Anthropic's current documentation, as features change.
Is AI memory the same as a company knowledge base?
No. AI memory is built from one person's conversations and held in their account or project, without owners or sources. A company knowledge base holds curated, sourced facts that every person's AI reads. Memory personalises style; a knowledge base keeps substance consistent.
What happens to ChatGPT memory when an employee leaves?
On a personal account, it leaves with the employee. On a company account, it belongs to a user who has gone and is rarely usable by anyone else. Useful knowledge should be written into shared company files as people work, so it is not lost at offboarding.
Should businesses turn AI memory off?
Not necessarily. Memory helps with personal preferences and ongoing work. The key is a clear rule: preferences can live in memory, business facts belong in shared files with owners. Use business plans, project-level memory for client work, and review settings as vendors change them.
Related reading
- Anthropic. "Bringing memory to Claude." Anthropic, 11 September 2025, updated 23 October 2025. https://claude.com/blog/memory
- VentureBeat. "ChatGPT's memory can now reference all past conversations, not just what you tell it to." VentureBeat, April 2025. https://venturebeat.com/ai/chatgpts-memory-can-now-reference-all-past-conversations-not-just-what-you-tell-it-to
- Microsoft and LinkedIn. "2024 Work Trend Index." Microsoft, 8 May 2024. https://blogs.microsoft.com/blog/2024/05/08/microsoft-and-linkedin-release-the-2024-work-trend-index-on-the-state-of-ai-at-work/
- Panopto. "Workplace Knowledge and Productivity Report." 2018, reported by HR Dive. https://www.hrdive.com/news/inefficient-knowledge-sharing-costs-large-us-businesses-47m-a-year/527892/