The short answer: I asked Claude to analyze more than 100 of my own work chat sessions against current leadership best practices. It called me a system builder whose step-by-step method costs more tokens than one big brain dump up front, and it noted that I often skip its follow-up questions. The same exercise may help anyone who uses an AI tool regularly.
Key takeaways
- Your chat history is a record of how you work. An AI tool can review it for patterns you do not see yourself.
- The first goal was lower token cost. The more useful result was a working-style review.
- Treat the output as input, not a verdict. The tool only saw one account and one slice of my work.
- I have only tried this in Claude, so I cannot say whether it works the same in Copilot or ChatGPT.
What I did
I have hundreds of chats in Claude Cowork, and I decided to analyze my chats with Claude. The results were quite interesting.
I started out by trying to analyze my token use and see whether my prompt pattern could be improved to reduce my token cost. I had over 100 chat sessions to analyze and felt that was a good sample to work from. What Claude ended up doing was always giving me a working-style analysis, at a very high level.
I found that interesting and decided to go further and see what it would tell me about my overall working style. I told Claude to analyze my chat sessions and tell me my strengths and weaknesses based on current best practices for leadership.
What it told me
It went through my sessions and analyzed my working style and how I can best work with teams. Some of it was interesting.
- I skip follow-up questions. Claude called out that I often do not answer the questions it asks after producing output, and said I do not complete tasks as a result. I just do not like paying for questions I did not ask it to bring up.
- I am a system builder. It said I think through a process, and that this method, although great, was costing me more tokens than a big brain dump up front. Fair enough, it was a good call-out. It also said that working through problems step by step might be good for some team members, but as a way to keep token costs down, it is not the best.
Why I think this is useful
I think this kind of analysis would help many people who use Claude on a regular basis and want to improve how they use the tool. I am not sure whether the method works in Copilot or ChatGPT; I only tried it in Claude.
One caution. I have the benefit of two accounts, one for work and one for home, and I analyzed only my work account. If I had blended the two, the outcome would likely be different. Keep that in mind if you want to do this kind of analysis.
A suggested way to run it yourself
This section is a suggestion, not a description of what I did.
- Check your policy first. Make sure your employer allows you to analyze your chat history, and strip out names or confidential details if needed.
- Pick one account and one time period. Mixing work and personal use will blur the picture.
- Ask for a specific frame. For example: "Review these sessions and tell me my strengths and weaknesses based on current leadership best practices. Give examples from the sessions."
- Choose one change. Pick a single habit to adjust, such as answering follow-up questions or batching your instructions up front, and see what happens to your results and your token use.
- Check the work. The tool is reading patterns, not knowing you. Keep the parts that ring true and discard the rest. This is the same discipline described in The Three C's of AI.
Frequently asked questions
Can I analyze my own AI chats to improve how I work?
Yes, that is what I did. I asked Claude to review more than 100 of my work chat sessions for strengths and weaknesses against current leadership best practices. I have only tried it in Claude, so I do not know if it works in Copilot or ChatGPT.
What did the analysis find?
Claude described me as a system builder who thinks through a process step by step. It said that approach, while effective, cost more tokens than a single big brain dump up front. It also noted that I often do not answer its follow-up questions after an output, which left some tasks unfinished.
Is it safe to analyze work chats this way?
Treat chat history like any other work record. Check your employer's data and AI policy first, and consider removing names or confidential details before you ask a tool to review them. I analyzed only my own work account.