Use your athlete context in conversation
Practical guides to giving Claude or ChatGPT reliable athlete context and activity data without re-explaining it in every conversation.
Why your AI plan paces you off your goal, not your fitness
Because the goal is usually the only number you gave it. Asked for a sub-3:30 plan with no current benchmark, an assistant derives every training pace from 3:30, so you train at the fitness you want rather than the fitness you have. The fix is one extra number: a recent result or a tested threshold, labelled as the benchmark.
Can ChatGPT predict your race time?
It can but it's accuracy can vary wildly. The maths is simple: most of the time it is applying Riegel's formula to a result or data point you gave it. That means the prediction inherits what you've told ChatGPT entirely, even though the context might not be right. A split inside a longer race, a downhill course or a four-year-old best all let ChatGPT produce a confident number about an athlete who is not you today.
Does ChatGPT remember your training numbers?
How long is. apiece of string? Sometimes it remembers, and far better than a year ago: OpenAI rebuilt ChatGPT’s memory in June 2026 and reports that context carries forward 82.8% of the time. Remembering is not the same as knowing which value is current, though. On OpenAI’s own scores the weakest objective is keeping up when a fact changes, at 75.1%, and the memory summary you can read is explicitly not everything it is working from. For training facts that change, that gap is the one worth planning around.
Cycling training with ChatGPT: the numbers it needs from you
Give ChatGPT your FTP, your real weekly hours, which days can take intensity, your event and its date, and anything that limits you. The catch is that cycling prescriptions are generally percentages of FTP, so a stale or estimated FTP quietly rescales the whole plan: sweet spot written from an old 308 W lands at threshold if you are actually 285 W.
Did my training block actually work?
Only if it can see the before as well as the after. A list of sessions describes what you did; whether it worked is a comparison between two dated anchors, the intent of the block, and what actually happened to you during it. Overwrite the old threshold and the comparison disappears.
What the Strava connector doesn't tell Claude
A lot, and it is genuinely useful: real activity history, heart-rate and power zones, an FTP figure, gear, and best efforts, all read live from your account. What it cannot give Claude is anything you have not recorded as an activity: your race, your goal time, an injury, or which of two FTP figures you actually trust. Strava's connector reports what it computed from your training. It has no field for what you are training towards.
Connecting training data to ChatGPT or Claude: what works today
What each assistant connects to differs more than how well either writes a plan. Claude reaches Strava on any plan, including the free one, which covers most devices. ChatGPT reaches Apple Health through ChatGPT Health, bringing in Garmin, WHOOP, Oura and Strava, but only on a United States account with an iPhone. COROS connects directly to both. Neither route holds what you are training for.
Why your AI training plan ignores your schedule
It did not ignore your schedule. It never had it. ChatGPT and Claude both respect constraints when given them, and neither can discover yours, so a plan written without your hours, your hard days and your equipment is written for a notional athlete with open evenings. Restating it works once; every new conversation starts empty again.
Triathlon training with ChatGPT: three sports, three thresholds
A triathlete has three thresholds, not one: FTP in watts, run threshold pace per kilometre and critical swim speed per hundred metres. They come from different tests and go stale at different rates, usually because training time is split unevenly. Give ChatGPT one of them and it might even build a week around all three, inventing the two it does not have.
How to use ChatGPT or Claude as a running coach
Give the AI enough context about who you are and what you are training for and then put them to work, creating clean coaching plans that work. A 2024 peer-reviewed study found ChatGPT-generated training plans rated dramatically better by coaching experts when given detailed athlete input instead of a bare request in a new chat. Read on to find out: what to include, a worked prompt, the honest limits, and why keeping that input current is a separate problem from writing it well once.
Using ChatGPT when you already have a coach
Both can work in tandem and ChatGPT can help you understand a coach's prescribed session, solving logistics around it, and preparing better before it starts. Rewriting the week isn't a good idea though, because the AI does not know what your coach was building towards, duplicates work and lets doubt start to creep in.
What to tell ChatGPT or Claude before a goal race
Update four things and leave the rest alone: your current fitness (not the fitness you had when you started training), whether the taper is going to plan, any niggle that has appeared or resolved, and the exact goal, course and conditions for the day. Do not introduce anything new this week, and say so explicitly so the assistant does not suggest it. Below: why this check-in is a different job from asking for a plan, what to update, what to say once and let go, and a worked prompt for race week.
What to update after a race
A race changes three things independently: the result, possibly a personal best, possibly your threshold, and a goal that is now finished. None of them reach your assistant on their own, and the old threshold keeps producing paces until you replace it.
Why your assistant agrees with everything you say
Because it is trained on human approval, so it learns to pick the answer most likely to please rather than the one most likely to be true. With training questions that is costly: told you can go sub-three, it builds the plan. Facts are what give it grounds to push back, and an opinion gives it none.