Alcoven

Why is ChatGPT mixing up my chats? How to keep clients apart

Blog · Published 11 November 2026

ChatGPT mixes up chats because its memory reaches across them: saved memories, chat history and projects on default memory can all carry a detail from one conversation into another. To keep clients apart, give each client its own project set to project-only memory, keep client details out of your global custom instructions, and use a temporary chat for one-off questions. Claude keeps each project’s memory separate by default; the steps for it are further down.

Why does ChatGPT mix up my chats?

Say you ask for a launch email for one client and the draft opens with another client’s tagline. When ChatGPT mixes up conversations like this, it has pulled the detail from somewhere it is allowed to look. There are five usual places.

  • Saved memories. These are details you asked ChatGPT to remember, or that it saved as useful context, and they carry into your other chats. Deleting the chat a memory came from can leave the memory behind: OpenAI’s help centre says deleting a chat “does not necessarily delete a separate saved memory created from that chat”.
  • Chat history. With Reference chat history on, ChatGPT can use relevant information from past conversations. In OpenAI’s words, “ChatGPT decides which available information is relevant to a response”. Past chats about your other clients are part of what it can draw on.
  • Projects on default memory. A project keeps a client’s chats and files together. On every plan except Enterprise and Edu, though, chats in a default-memory project “can reference both project and non-project conversations”, and chats elsewhere can reference them back. A client folder on default memory still lets the other clients in.
  • Global custom instructions. OpenAI applies them to all your chats, though a project’s own instructions override them. A client’s name or house style written there can follow you into chats about other clients.
  • Long chats. Earlier messages in a chat can be part of what the model reads for the next answer, so a thread that began on one client and drifted to another can keep both in play. Keep one client per chat.

Depending on plan and region, memory can draw on files in your Library and on connected apps such as Gmail as well. A saved memory can also be out of date: OpenAI has written that saved memories “tend to go stale over time and eventually become incorrect or irrelevant”, which its June 2026 memory update set out to tackle.

How do I use ChatGPT for multiple clients without mixing them up?

  1. Make one project per client. Select New project in the sidebar, add the client’s files, and put the client’s context in the project instructions. Project instructions “apply only within that project and override your global custom instructions”.
  2. Set each client project to project-only memory. Open the project, select the ••• menu, then Project settings. Under Memory, choose Project-only memory and save. Saved memories are then left out of the project’s chats, its chats can’t reference conversations outside it, and chats outside can’t reference it. There is no global switch, so set each project one by one, and allow a few hours for the change. Project-only memory needs memory turned on in your settings, and on Business and Enterprise in the workspace too.
  3. Keep client details out of your global custom instructions. Use them for how you like answers written. Everything about a client belongs in that client’s project.
  4. Start each chat inside the right project. If a chat ends up in the wrong place, open its menu and choose Move to project, or Remove from project to take it out. If the option isn’t there, the chat may not be able to move, so start a new chat in the right project.
  5. Use a temporary chat for one-off questions. Choose Unpersonalized before the first message and it uses no memories or custom instructions. A Personalized temporary chat can still use them. Neither kind creates or updates memories.
  6. Trace a leak when you see one. When ChatGPT uses personal context, Sources may appear under the answer. From there, depending on the source, you can delete a referenced chat, correct or delete a saved memory, or mark a source as not relevant. To remove a detail ChatGPT saved as a memory, delete the saved memory and the chat where you first shared it.

A shared project is always on project-only memory and can’t be switched to default memory, which suits client work.

Should I just turn ChatGPT’s memory off?

You can, under Settings > Personalization > Memory. Where your settings show Reference saved memories, turning it off also turns off Reference chat history. Turning off only Reference chat history leaves your chats where they are; what ChatGPT remembered from them is scheduled for deletion within 30 days.

The cost: in each new chat, you explain again what earlier chats already settled. Project-only memory keeps each client’s history working inside its own project and keeps the other clients out, so try it first.

How do I keep clients apart in Claude?

Claude keeps each project’s memory apart by default. Anthropic’s help centre says each project “has its own separate memory space and dedicated project summary”, apart from other projects and from chats outside projects, and a search of past chats from inside a project stays within that project.

  1. Make one project per client. Projects are on every plan, up to five on Free. Add the client’s files to the project knowledge and the client’s context to the project instructions, which apply only to chats in that project. On Free, check that you can set project instructions: one of Anthropic’s help pages lists them for paid plans only.
  2. Keep client work out of standalone chats. Outside projects, memory and chat search draw on all your non-project chats together.
  3. Keep client details out of Instructions for Claude. They are account-wide and apply to every conversation, incognito chats included.
  4. Use an incognito chat for one-offs. Click the ghost icon when you start a chat outside a project. An incognito chat doesn’t use Claude’s existing memory and isn’t saved to memory or to your chat history. It isn’t available inside projects.
  5. Check what Claude remembers in Settings > Memory, where you can edit or delete each topic. Deleting a chat doesn’t remove memory already generated from it. Pause memory stops Claude using or adding memory; Reset memory deletes all of it, project memories included.

Memory is on by default on Free, Pro and Max, and off on Team and Enterprise until an owner turns it on. Long chats matter here too: when a chat nears its context limit and code execution is on, Claude summarises earlier messages to keep going, so one client per chat keeps that summary about one client.

What if my team uses ChatGPT and Claude?

Everything above keeps clients apart inside one person’s AI. In ChatGPT and Claude alike, personal memory belongs to one person inside one vendor. OpenAI says memories are “tied to each individual account” and can’t be shared with other members, even in the same Business workspace.

A shared ChatGPT project gives its members the project’s chats, files and instructions, and it is for people signed in to ChatGPT. A shared Claude project, on Team and Enterprise, gives members of your organisation its knowledge and instructions, while each person’s chats stay private unless they share them. Say your strategist uses ChatGPT and your designer uses Claude: each builds a separate picture of the client, and the two pictures drift apart until the team shares one project between ChatGPT and Claude.

Where Alcoven fits

To keep clients apart across the whole team, give each client its own Alcoven project: its brief, documents, tasks and history, shared by the people and agents you add. A tool you allow reaches only the projects you’re on, and opens one at a time, by name.

Your strategist adds Alcoven to ChatGPT, on plans where an MCP tool can write, and your designer adds it to Claude; both ask about the client’s project by name. Your developer clones it with git to work from Claude Code or Codex. Each person’s own AI does the reading and writing (Alcoven runs no model for this), and whatever it changes comes back under that person’s name, for everyone on the project. Every document keeps every past version; open one in the Alcoven app and you see what others changed since you last read it, and who changed it. Clients join as readers, free on any plan. Free keeps one project active; Solo and Team have any number.

Start a project free

Sources

All checked on 30 September 2026.