Some of the most worthwhile conversations I've had with ChatGPT (and sometimes Claude) aren't planned. The first question I ask isn't a straight line to the last answer.
It's like having a human conversation over coffee. One thought leads to another and before long you've covered a lot of ground.
What I was struggling with is what to do with the ideas that bubbled up during the chat. There were things I wanted to come back to at some point but rereading long conversations wasn't practical.
So, I created an instruction doc that does the work for me. It isn't a summary of the conversation. It goes deeper. It extracts opportunities, insights, topics, newsletter ideas, and more. Each of these has to connect with other work I'm doing and have a life of at least 6 months.
I save this as a working doc in my local Discovery Library. Once a week I add these to a Project and ask AI to give me a fresh review with specific instruction criteria.
This discovery process has given me a place for my thoughts to land until it's time to do something with them.