i tested 4 free AI running coaches so you don’t have to

if you read my last post about building a strava app with AI, you know i’ve been thinking a lot about where this stuff is actually headed for regular runners. not elites with coaches and altitude camps, people like me, running 3x a week, trying to get a half marathon done in the fall without breaking something.

so i decided to run an experiment. i took my current stats, roughly 5:00/km pace, averaging around 30km a week, with a 21.9km long run on my birthday that i did not plan but apparently my legs did, and fed the same prompt into four free AI tools to see what kind of training plan they’d spit out.

the prompt:

“i’m a returning runner in my mid-30s based in toronto. currently running 3x per week at around 5:00/km. my longest recent run was 21.9km. i’m targeting a half marathon in fall 2026 with no specific race date yet. build me a 12-week training plan.”

same input, four tools. here’s what happened.

tool 1: chatgpt (free tier)

chatgpt didn’t ask a single follow-up question. i hit enter and within about ten seconds i had a fully formatted 12-week plan with monday through sunday laid out, rest days included, long run progressions, the whole thing. it looked extremely official.

it was also completely generic. the plan could have been written for literally anyone. there was no acknowledgment that i’m a returning runner, no adjustment for the fact that i’ve already done 21.9km, no toronto winter consideration, nothing. it gave me week 1 starting at 5km easy runs like i’m brand new to this.

to its credit, when i pushed back (“i’ve already run 21km, can you adjust?”) it immediately rebuilt the plan around that. so it’s not dumb, it just doesn’t ask questions it should probably ask.

verdict: 3/5. good starting point, bad at reading the room.

tool 2: claude (free tier)

yes, i used the AI i literally wrote about last week to coach me. the irony is not lost on me.

claude actually asked a few things before generating: did i have any injuries? was i targeting a time goal or just completion? how had my recent training felt? it felt less like a vending machine and more like a conversation.

the plan it produced was more nuanced. it acknowledged my base fitness, built from where i actually am rather than square one, and flagged that without a race date it was working with a rough window. it also told me what it didn’t know and where i should adjust based on feel, which is either genuinely useful or a very polished way of covering its bases.

downside: it’s still a chatbot. it has no idea what i actually ran last tuesday. every conversation starts from scratch.

verdict: 3.5/5. smarter out of the box, but still only as good as what you type.

tool 3: athletica.ai

athletica is running-specific, which already puts it in a different category. signing up was pretty smooth. it connected to strava during onboarding, which means it’s actually pulling my real data rather than trusting me to self-report accurately. that matters.

the plan it generated felt noticeably more calibrated. it wasn’t starting me at week one because it could see i’ve been running. it also builds in something called “adaptive scheduling.” if you skip a workout or tank a session, it adjusts the following week rather than just assuming you nailed everything.

the interface is clean. it’s not flashy but it’s functional. the free tier gives you enough to actually train on, though some of the deeper analytics are paywalled.

this is the one that felt most like an actual coaching product rather than a chatbot wearing a bib.

verdict: 4/5. best option for apple watch + strava runners who want something that actually adapts.

tool 4: trainasone

trainasone has been around longer than most AI coaching apps. the interface shows it. it’s not ugly, but it has that “built by engineers for engineers” energy where you’re clicking through more tabs than you expected.

it also connects to strava, pulls your history, and builds a plan around what it actually sees, not what you tell it. the AI here leans harder into machine learning than the others, meaning it’s trying to model your fitness curve over time rather than just generating a template.

the plan it gave me was solid but conservative. more recovery weeks than i’d have chosen, more “easy” labels on runs. whether that’s the AI being smart about injury prevention or just being cautious, i genuinely can’t tell. probably both.

the free tier is functional but limited. you’ll hit the ceiling faster than with athletica if you want more than a basic plan.

verdict: 3.5/5. smart under the hood, but the experience needs some work.

so which one is actually worth using?

for an apple watch + strava runner with no garmin in the picture, athletica is the clear winner. it connects to your existing setup, adapts when life happens, and doesn’t make you start from scratch every time you open it.

chatgpt and claude are useful if you want to think through your training or get a second opinion on a plan, but they’re conversation tools, not coaching tools. there’s a difference.

trainasone is worth a look if you’re data-obsessed and patient with interfaces, but athletica gets you there faster.

am i going to follow one of these for my fall race? probably athletica, loosely. the honest answer is i’ll use it until i miss a week and then feel too guilty to open the app and quietly revert to winging it, which if you’ve been reading this blog for more than five minutes, you already knew.

if you’ve actually trained off an AI plan and survived, drop it in the comments. i want to know if anyone is doing this for real.