Using AI as a 10K Training Coach

I am a terrible runner. I suck at running. But I always wanted to try.

I love getting out and going for walks, but they take time and don't burn the calories my Tangfastics addiction require me to. I needed to find an exercise that helped get me moving, lose some weight, and that I didn't totally hate.

I got on a treadmill and stumbled my way through a few workouts. I did my first Parkrun in a few years. After a few more of those eventually made it round without stopping. After that I did my first official 5K race and did it in record time.

The next logical step is 10K right? The Bournemouth Bay Run is coming up in March. That's a terrible idea, right? I've managed to bungle my way over 5K so double that would just end badly, surely...

So I did what anyone else would do and booked my spot before I could decide against it.

Setting a training plan

I can now run 5K in 28 minutes. Even my basic maths skills gets that pace out at a 56 minute 10K. I set myself the goal of completing the 10K in under an hour, with anything faster than that a bonus.

I figured it would take more than a playlist of mid-00s bangers and a fancy breathing pattern to get me there. It's time to get a training plan in place.

Searching online for the right plan seemed impossible. There's so many different ones out there it's hard to know where to start. They vary from "try a light jog every other Sunday" to "you can never sit down". I am not a complicated man - give me something simple and repeatable and I will do it.

AI is a great tool for doing things you know are possible but can't find the means, so I thought it would be a great place to source a training plan. It's got the whole of human history to find the perfect plan for me - plus I don't need to deal with those other Matt types.

Comparing models

I have no idea how AI would do at a task like this. My expectations were low. If it could get something vaguely recognisable as a training plan for a human I'd be happy.

To give it a try, I gave all the models the same prompt. It's nothing sophisticated but gives it enough information to make a tailored suggestion, while also being open-ended enough to see their differences.

I have a 10K run coming up on the 23rd of March. I can run a 5K in about 28 minutes currently. Act as if you are my personal trainer and devise a training plan for me to get me a 10K time under an hour by 23rd March.

I can go to the gym on Mondays, Wednesdays and Fridays, and I run a 5K on Saturdays. Make a plan to accommodate those days.

ChatGPT

I'm not a heavy AI user, but ChatGPT would be my go-to. I'm comfortable with how it works and know how I can keep prompting to get what I need.

It wasn't particularly chatty about it, but gave me an 11 week plan to follow. It broke the phases before it in to 4 chunks - building a base, adding speed & distance, race prep, and then a taper week just before the race.

Every phase had a defined goal, describing why it was broken up as it was. Each phase was made up of a few weeks, with each week comprising of an easy run, a speed workout, a long run and that 5K run on Saturday. Each workout was accompanied with some strength training workouts. That made it really easy to follow.

The only thing I needed to correct was asking for it to put the paces for each runs in km/h instead of min/km that it defaulted to, which it did without issue.

Claude

I hadn't had much exposure to Claude to be honest. I understood it was better at more in-depth tasks so I was interested to see how it would turn out a training plan.

It gave a similar training plan to ChatGPT, with clear phases designed to work at different rates depending on how long it was before the race. It made it easy to skim through as I needed to.

What I did like was the separate artefact for the plan itself, while the main response was dedicated to explaining what it was about. While it didn't stick to the 5K Saturday runs it used them as parts of longer workouts. In its defence I didn't specify I didn't want that, but it's not how I intended the prompt to be interpreted.

In short, the plan it generated was more evenly spread throughout the week. However in the first pass it suggested I would run 10K at any point was 2 weeks before the race, which seems like it's cutting it fine! Further prompts pushed it earlier in the plan and adjusted surrounding runs to make them less intense.

Perplexity

I like how Perplexity cites its sources. It suggested that whatever it would generate would be well intentioned, if not a little impersonal.

It managed to give some kind of structured response, but I found it was lacking in a lot of detail. While ChatGPT and Claude both suggested specific weight training exercises for the days at the gym, Perplexity simply suggested "strength training" with absolutely no aerobic exercise at all.

The only running it suggested was the Saturday run, which at no point suggested I should try and run 10K. I feel like that's quite an important step when training for a 10K.

After some gentle prompts in the right direction, it did give me some more detail on what "strength training" was. I also liked how it also suggested follow-up questions about diet and progress tracking. But ultimately it didn't seem like it could give me anything more than surface level responses even after providing more information.

Gemini

This was my first time using Gemini. I figured because it was Google it would be as good, if not better, than what I'd get from searching the Internet myself.

I was wrong.

It took a couple of goes to get any kind of week-by-week plan out of Gemini. It gave some good tips about training for a running event and for following training plans generally, but took a few follow-up questions to get anything out.

Once I did it gave me a fairly detailed strength training routine to follow, which was great but I needed more from it.

While it was always polite enough when I asked it for corrections, it never seemed to try too hard to generate what I was asking for. Perhaps the prompts were too loose, but I would expect a 10K training plan to include some running before the event.

Let's get training

I decided to follow ChatGPT's suggestions in the end. It gave a simple enough structure with scope to adjust as needed. It had generated a table for each week's training plan, which made it easy to reference at the gym each time. While Claude's was equally as detailed, I found the plan it generated to be a little too safe and wanted to push myself a bit more.

As I go, I'll be prompting and changing the plan to suit the progress I'm making. I'll also use it to ask any follow-up questions for things as they come up in those sessions.

I'm tracking how that training plan is going on Strava, but I'll also post updates here soon to show how I'm getting on.

If I can get sub-1 hour on race day I'll count this as a success. Anything under that will be a fantastic bonus. Depending on how the training goes, I might even make the goal pace faster and see how it copes!

I think it'll be an interesting way to mix tech and fitness in a way I'm not sure I'll get to try again! Let's see how it goes 🏃‍♂️