"Sorry, this conversation has reached its limit. Let's start a new chat."

I hope this helps,

Microsoft does not publish a specific conversation-length limit for Copilot for Microsoft 365. When the message “Sorry, this conversation has reached its limit. Let’s start a new chat.” appears, it is normal behavior and enforced by the Copilot service. This limit cannot be configured or changed by administrators.

Currently, there is no official Microsoft Learn documentation that defines the maximum number of messages, the token or context-size limit, or any differences between applications such as Teams, Outlook, Word, or Microsoft 365 Chat. Only Copilot Studio (custom agents) has publicly documented quotas. The built‑in Copilot for Microsoft 365 chat experience does not.

Although this limit is not formally documented, many customers observe that conversations tend to reach the limit after roughly 80 to 120 total messages or when the conversation reaches an approximate context size of 25,000 to 40,000 tokens. These numbers are based on common usage patterns and can vary depending on message length, the workload being used, and general service load. They are not official or guaranteed values.

The limit exists to maintain performance, reliability, and answer quality. As a conversation grows longer, Copilot must reprocess the entire thread each time it generates a response. Larger context windows increase compute requirements and can reduce clarity or accuracy in extended threads. Session limits also help minimize the risk of sensitive information being carried forward unnecessarily.

Example:
Consider a long conversation in which Copilot has already processed dozens of questions, references to documents, and earlier parts of the discussion. Each new message requires Copilot to reinterpret the entire history so it can respond correctly. As the conversation becomes larger, this processing becomes more resource-intensive and may slow performance or affect answer quality. The conversation limit ensures that responses remain accurate, efficient, and focused on the most relevant context.

For longer workflows, one effective approach is to ask Copilot to summarize the discussion and start a new chat using that summary. Another option is to move the ongoing context into a document such as Word, OneNote, or Loop and rely on that file instead of maintaining an increasingly long conversation thread.

Thanks! ✌️

I have been using Copilot for a while now, and it's not a tool that can be used for anything but simple short tasks.

Trying to use it for anything complex will result in the error you see. Which typically leaves the chat thread in a state where you can never reliably reopen it. So, all you work you have done is lost.

Medium complexity tasks where you have not hit that limit are equally compromised, as it's often impossible to re-open a thread and continue.  As Copilot will hang when you re-open the thread.

The only workaround is to break everything into a simple task and keep track when you have had more than a few interactions and know to stop. As you are close to losing everything you have done.

Ask for a document summarizing where you are and then start a new chat from that document.

Copilot is simply not ready for anything more than simple short interactions.

Why the limit is 30 has to do with performance and user experience. Why it is 30? I can only assume it was the result, after testing phase, they ended up with before performance started to drop. 

The more detailed your prompt is, the better the result will be which results in fewer attempts to get to the result you desire. Once you figure out which prompt gets you the response you need, simply save the prompt to your prompt gallery so you can reuse it.

Same issue here.  Last week, I was looking for some information within 6 months of emails.  We were just about to be to the point where the response info needed to be and I was cut off.  Now, I have to start over.  What good are these products if they can't complete a project with me?  Chatgpt does this too and so does Claude but Claude seems to be the best one out of Gemini, Copilot and Chatgpt.