Every run you do lands in the same place: a grey row in an app. Distance, pace, a heart-rate trace, a map. Strava is a near-perfect logbook and an almost useless coach — it records everything and interprets nothing. A good coach does the opposite. They glance at your week and tell you the one thing that matters: you jumped your volume too fast, you spent the whole week in the grey zone, that “easy” run was secretly a threshold effort. Most of us never get that read. So I handed the job to Claude.
How do you set Claude up as a running coach?
Setting Claude up as a running coach takes three moves: connect Strava, connect email, and schedule a written brief. The mechanics are almost boringly simple. In Claude you connect the tools it needs, give it a standing instruction, and let it run on a schedule — that’s what the connectors and recurring tasks are for. Mine fires every Sunday evening and lands in my inbox before I’ve finished stretching.
- Connect Strava (read-only) so it can pull the week’s activities — distance, moving time, pace, elevation, heart rate, and which shoes you wore.
- Connect email so it can send the finished report as a clean HTML summary rather than a wall of chat.
- Write the brief and schedule it — a recurring task, in my case Sunday at 19:00, covering the week just gone.
What should the coaching brief ask for?
The coaching brief should ask for judgements on your actual numbers, not a recap of them. “Summarise my week” gets you a summary. A coaching brief gets you a coach. The entire difference is in what you ask for, and it pays to be demanding. Mine tells Claude to read the week the way an experienced endurance coach would, and to stay specific to the actual numbers:
- Lead with the headline totals — distance, elevation, moving time, number of runs — and a week-over-week comparison, flagging any jump beyond roughly 10%.
- Judge the intensity distribution — the easy / moderate / hard split — and say whether it’s sensibly polarised or stuck in the moderate grey zone.
- Check pacing within key sessions — did the effort hold, drift, or fall apart, and did that match the point of the session?
- Look for fatigue and progression — heart-rate drift at a steady pace, pace decaying at constant effort, or gains over comparable past efforts.
- Place the week in a block — base, build, peak, taper or recovery — and flag overreaching, or any run whose execution betrays its intent, like intervals run too timidly.
It asks for tight prose over bullet spam, metric units, British English, and no motivational filler. What comes back reads like a note from someone who actually watched your week.
A brief along these lines is enough to start with: “Read my last seven days of Strava runs as an experienced endurance coach would. Give the week’s totals and the change from the week before. Say how much of the running was easy, moderate and hard. For each key session, say whether the pacing held. Name what you cannot judge from this data. No encouragement, no bullet lists.” Tune it after two or three weeks, once you see which parts of the report you actually read.
The part that makes it trustworthy
Here’s what surprised me most, and it’s why I keep it: the brief tells Claude to admit what it can’t see. Training load is only half the picture. Readiness and recovery live in sleep, HRV and resting heart rate — none of which are in Strava — so a good report says “I can’t assess recovery from this data” rather than inventing a verdict from mileage alone. When a run has no heart-rate data it estimates zones from pace, and says it’s estimating. That honesty is the whole difference between a useful assistant and a confident fabricator.
A coach who tells you what they can’t see is worth more than one who cheerfully guesses.
Can Claude replace a human coach?
No: a weekly report from Claude doesn’t replace a human coach who knows your body, your life and your goal race. But as a weekly sanity check it’s fast, direct, and free of flattery. Did I overcook it? Am I drifting into the grey zone? Is this week actually building anything?
The limits are worth writing down before you trust it. The report sees only what Strava recorded: no sleep, no resting heart rate, no niggle you didn’t log, and no idea that Tuesday’s slow run was slow because of a heatwave or a hangover. Heart-rate zones are only as good as the maximum they were set from, so check yours first with the heart rate zones calculator. And it can’t write the plan for you. It reads the week you ran against the week you meant to run, so keep the plan somewhere it can see, even if that is a line in the brief.
Two ways to make the numbers mean something
A Sunday-night email is one way to make your training data matter: the cold, analytical read that keeps you healthy and pointed in the right direction. The other half is warmer, and it arrives at the end. When the block is done and the race is run, the same numbers behind this week’s report can become something you hold, not an email you archive. Let Claude read your training week by week; keep the season somewhere it won’t get archived.