Sometimes I have long and relatively wide-ranging conversation threads with Claude. It’s almost always related to a specific topic, but includes tangents and diversions and even some dead ends. Managing these kinds of long Claude AI conversations effectively is crucial for anyone working on complex projects or exploring multifaceted topics with the AI assistant.
When those conversations get too long, Claude gives you a warning message: “Tip: Long chats cause you to reach your usage limits faster and suggests starting a new chat.”
The problem is that starting a new chat means that Claude is oblivious to all the good stuff we discussed in this chat.
A prompt for managing long AI conversations
So I have a prompt I use to ensure that the best of this conversation is captured and added to the project knowledge documentation (Projects are only available in the paid version of Claude):
I want you to analyse all the chats in this thread and separate them into different categories. I want you to use those categories to create a set of artifacts (one for each category) that I can add to the project knowledge base. Each artifact should capture the gist of our discussion in that category, as well as the conclusions and decisions about how I’m going to move forward.
You then have the option to add these artifacts to the project documentation, which means that all future chats will include this context. The point isn’t so much about ensuring that the artifacts perfectly capture everything discussed. This of them more like a handover to another team member where some of the nuance will be different. The goal is to ensure that future conversations take this context into account.
Once I’ve added the artifacts to the project knowledge base, I can close off that conversation thread and start another one. Sometimes the new chat includes specific reference to the artifact, in cases where I didn’t get to a conclusion:
I want to continue our conversation on the topic of XXX (refer to the document entitled: “XXX”)…and so on.
And sometimes that artifact just becomes part of the background knowledge in the project, and it gets reviewed as part of every chat, only coming into play when Claude decides that it’s relevant.
While managing long AI conversations can seem challenging at first, the process of creating and organising knowledge artifacts becomes second nature once you understand how to categorise and document key insights.