Use case P60 · Human
ChatGPT subscriber uses Claude 4 Opus to help analyze his exported conversation history and check how much he uses his plan
- Money and shopping
- Advanced
- Claude 4 Opus, ChatGPT, Jupyter notebook
- Source checked · blog post
The short version
What they did
Max Song exported his own ChatGPT conversations and built a Jupyter notebook, starting from Chip Huyen’s heatmap script. Claude 4 Opus helped write the token-counting algorithm and code so he could account for conversation history and regenerated branches. He used the charts to inspect how often he used Deep Research and which models he chose; his comparison with hypothetical API prices is not verified savings or a reason to consume more tokens.
Result (as claimed)
“You can probably imagine my disappointment lol, as you can see in this graph - my usage is nowhere near the maxmium quota.”
“These are quite interesting charts, not only does the token count better reflect the fact that I started giving ChatGPT more and more difficult tasks in late 2024 - 2025, it also revealed the dramatic shifts in my model choice and when that started to occur.”
Prompt
I’m reviewing how I use my ChatGPT subscription. Use Claude Code to help build a local Jupyter notebook for my own conversations export at [file location]. The date range I want is [period].
1. Inspect the export structure and explain how conversations, models and dates are recorded. Flag missing fields.
2. Chart conversation counts by date, model choice over time and Deep Research frequency where the export identifies it.
3. Before estimating tokens, explain how you will count repeated conversation context and regenerated branches. Show the assumptions separately.Give me a readable notebook with labeled charts and a short plain-English summary. Keep the analysis local and do not upload the export. Check a small sample against the underlying records. Label token totals as estimates, and do not describe hypothetical API spending as actual money saved.
Do it yourself
These steps are my suggested way in, not Max’s setup.
Level: Advanced for token counts. The first version below is Intermediate.
What you need
- Your ChatGPT data export, requested in Settings under Data controls. It arrives by email as a zip file.
- Python with Jupyter (free), or a coding agent such as Claude Code.
- Your plan’s price and limits, from the provider’s pricing page.
Start simple
- Request the export and unzip it on your computer.
- Ask a coding agent to write a local script that counts your conversations by month and by model.
- Run it and look at the chart.
- Compare your busiest month with what your plan allows.
Take it further
- Estimate tokens per conversation, counting the history each reply re-reads and any regenerated answers.
- Chart how often you use features like Deep Research.
- Run the same review for every AI subscription you pay for, and drop the ones you barely use.
Check your work
- Pick one month and count its conversations in the app. Compare with the script.
- Ask the agent to list every assumption behind the token estimate.
- Your export holds everything you’ve typed, so keep it on your computer. Hypothetical API prices are not money saved.
Source checked. The link, author, date and quote were confirmed against the source. The result itself has not been reproduced. Prompt adapted for this library, inspired by the case. Business size: Individual. Results are quoted as the source claimed them and never improved. Spot an error or a broken link? Email charlie@cbarmorecpa.com.