Short answer: maybe. Longer answer: I’m a runner. On September 21, I’ll be shoulder to shoulder with 50,000 other masochists at the Berlin Marathon, trying to shave three minutes off my time from the previous year.
(Idea initially shared in my newsletter, then made into a posting.)
I have a solid training plan. Problem is: Six weeks out, I wasn’t hitting my paces, which is the moment you either double down on wisdom or panic-Google. I chose a third option: ask ChatGPT to fix my life.
First, I had to get my data out of Strava, where I track my runs. There’s no easy export for specific runs, so I asked ChatGPT to write me a script.
Let's make an app that pulls activities from Strava. First, let's make a plan. I want to be able to select the type of activity (runs) and the starting date after which we pull every activity to a CSV file. I need everything but GPS data. I want to later analyze times, paces, intervals.What I got back was 536 lines of Python and step-by-step instructions for setting up Strava API access. I put the code into Google Colab. It threw an error. I pasted the error back into ChatGPT and got a fix. I ran the script again, and it gave me the requested data. The whole thing took maybe five minutes.
It still feels kind of magical.
Armed with my data packed neatly in a CSV file, I asked several LLMs to analyze my training and build a plan to hit my goal.
You are an experienced marathon coach. I want to run the Berlin marathon on September 21st. This is a file of my latest training sessions. Note the interval sessions and look at the splits and laps. During my last long run, I picked up the pace in the last 3k - you should be able to see this in the data. How is my training going?Claude Opus lacked details and was a bit cautius, o3 seemed more optimistic, but Gemini was outright delusional. The analysis from GPT-5 Thinking was extensive, with charts, and aligns with Strava’s prediction.
GPT-5 researched plans on the web, did the math, and returned a schedule that cut my rest days from three to one. One. I mentioned travel and two tune-up races, and it reshaped the plan instantly, like a coach who’s also a spreadsheet with feelings.
After each new run, I provide the latest data export to GPT-5. I’ve found it helpful to verify that GPT-5 can actually see and understand the data before proceeding with analysis. For example, I ask GPT-5 to confirm whether it can identify specific elements I know are present in the data, such as intervals. This validation step ensures the model truly comprehends the dataset before running any analysis.
Now, is this actually a good idea?
Probably not. I’ve ran Berlin five times before, and I like to think I know what I’m doing. But we’ll find out on September 21. Either ChatGPT helps me PR, or it breaks me in some spectacular, very public way in front of thousands of people.
Either way, I’ll let you know.
