← Building with AI

Fitness tracking with Notion + Claude

How I run my own training. Claude plans each week from what I actually did, picks sessions from a library I built with it, and checks every claim against the evidence. If my long-term health is at risk, it pushes back.

How a week runs

I run it two ways. A weekly check-in closes the week that ended and plans the next one. An ad hoc check-in plans what is left of the current week, after an injury, a missed session or a change of plans.

How a training week gets plannedA weekly or ad hoc check-in starts. Claude checks the local date, asks two batches of questions, and plans from actuals using my rules and gate register, my weekly log and my workout library. It grades the evidence. If longevity is at risk it pushes back and adjusts. Then it writes the week to Notion and puts sessions and movement snacks on my calendar.Rules+ gate registerWeekly logactualsWorkout library30/45/60 min, home or gymWeekly or ad hocCheck local dateAsk two batcheslast week + week aheadPlan from actuals+ agreed gatesGrade the evidenceLongevity at risk?Push back and adjustWrite the weekNotionday-by-day sessionsCalendarsessions + movement snacksyesno
  1. Claude checks my local date and time first. A one-day error can quietly delete a real session.
  2. It asks me one batch of questions about the week that ended: what I did, the numbers, sleep, and why anything was missed.
  3. It asks one batch about the week ahead: travel, fragile days, equipment and weather.
  4. It plans from what actually happened, not from the old plan. Gates I agreed in advance decide what changes.
  5. It writes the week into Notion: each session day by day, with paces, heart-rate ranges, sets and reps, and a tutorial video for each exercise.
  6. It creates one calendar event per session, with the full workout in the description.
  7. It pushes back if a plan puts long-term health at risk.

The offseason setup

After my October half marathon I rebuilt the system for the offseason. The goals changed, so the tools changed too.

Two coaches, one rulebook

Claude loads two skills every session. One is a training coach. The other is a longevity coach. When they disagree, my own rules page wins. Every goal, target and rate gets an evidence grade: strong, moderate, emerging, controversial, or expert opinion only.

Five pages in Notion

  • Athlete state and gate register. My numbers, my schedule limits, and the gates. A gate is a rule agreed in advance: if a sign shows up, a set change happens. No negotiating in the moment.
  • Coaching rules. How plans get written. Example: paces come from my last measured run, never from a chart.
  • Weekly log. What I planned and what I actually did.
  • Strength library. Exercises by movement pattern, each with a home version and a gym version. I pick on the day. Both count.
  • Workout library. Ready-made sessions in 30, 45 and 60 minute versions. Each one is tagged by phase, setting, intensity and the gates that can block it. It stores structure and heart rate, never paces.

Small movement through the day

The library also holds short movement snacks for sitting, standing and walking. Claude put four on my calendar: an 8 minute morning start, and three short breaks on workdays. The morning start doubles as a daily body check.

An example of the evidence check

I asked whether athletes who lean on the back of the body (glutes, hamstrings, calves) are more effective. Claude answered with graded evidence, then checked my own strength library against it. It found five gaps, including too little calf work and no plyometrics. I chose which fixes to add. It wrote them into Notion with the evidence grade on each one.

How this differs from the marathon app

The marathon training agent is a product for other runners. This is my own working system. It needs no code. It is a database, a set of rules, and a thinking partner.

Related: Notion-to-Website Publishing Pipeline