ChatGPT does a very good job of creating training plans. They aren’t perfect but very good. Getting one from a coach also wouldn’t be perfect.
It really depends on why you need a coach. Does AI do accountability or motivation? When you need feedback about how you are feeling during training can it do that?
ChatGPT training plans are garbage without a good prompt.
Most runners will be unable to engineer a good prompt without a deep understanding of training.
Yes the prompt is the most important part, but it doesn't have to be super complex when you already have a lot of data available.
For instance, for a runner who has 10 years of training behind him and his data in Garmin connect, he can export all his activities, and feed the engine with it.
In addition, you provide your race results during these 10 years and then it is easier to establish to what training you respond the best. The engine can establish your training profile and prepare an adequate training plan.
I asked one for a sub 16:00 5k training plan and it wanted me running 10 mile long runs at 5:20 pace, 8x800m in 2:10, a 20 minute tempo at 4:58 pace, ETC.
Trying to get training ideas from AI made me realize I can’t trust AI at all.
What I would say is rather that you cannot trust your prompt. GIGO.
I asked one for a sub 16:00 5k training plan and it wanted me running 10 mile long runs at 5:20 pace, 8x800m in 2:10, a 20 minute tempo at 4:58 pace, ETC.
Trying to get training ideas from AI made me realize I can’t trust AI at all.
What I would say is rather that you cannot trust your prompt. GIGO.
I experimented with prompts and higher and higher levels of details with the prompts.
If an AI model generates an utterly garbage training plan based off of as much information as…
- Height/weight/BF%
- Current PBs, Recent time trials/races, past improvement curves
- Avg Mileage and workouts when PBs were achieved, current mileage and workouts
- Up to a 1 year schedule of upcoming races
- HR/Cadence/Pace data from recent runs/workouts at different efforts
- Past and current Vo2Max data with past Vo2Max improvement curves
…Then it’s definitely not outperforming even half qualified coaches and it’s definitely not the prompters fault for getting ridiculous workout/schedule recommendations.
There are so many different training strategies, nuances, philosophies, and runfluencers/bloggers/crappy coaches who are muddying the waters that language learning models at their current level of development simply fail to write good training. They cannot detect what is and isn’t good training information online, and they can’t seem to figure out appropriate training paces for anything. The only way i’ve found to build a decent training plan is to tell the LLM your training plan and have it regurgitate it back to you, and even then you’ll get a hiccup here and there like “6x800m in 2:30-2:35 may be too strenuous for someone approaching 16:00 5k fitness, try pulling the reps back to 2:55-3:00 for a true Vo2Max stimulus”
Anyone saying that training plans generated by mass-market AI models are worth anything are stupid at best or intellectually dishonest at worst.
I'm sure that with proper model training and prompting, you could get a half decent plan. But that takes time and effort, and still wouldn't be as good as an actual qualified coach. It also requires high-quality user input. This presupposes that the user knows what they're doing and talking about, which kind of defeats the purpose of using AI to write a plan to begin with.
Give it a try, the plans AI give you are completely nonsensical.
For instance, for a runner who has 10 years of training behind him and his data in Garmin connect, he can export all his activities, and feed the engine with it.
With Runna, ChatGPT and NSM, why would an amateur runner need a running coach?
Is the jig up for unqualified coaches charging $100 a month for a cookie cutter training plan and an annual zoom call?
For 99.9999% of runners, just downloading any of the generic training plan off the web will work. If the coaching is in person that is a different story.
my .02 but for running related advice gemini absolutely slaps gpt out of the water. as a sub 2:30 marathoner i used gemini extensively in my last cycle to design workouts and get super personalized feedback based on my splits, ehart rate data, HRV, etc.