AI prompts for cycling training, sorted by what riders actually ask

Is this goal reachable? A coach checks a goal against your training before anything else. A model skips that step and writes the plan.

Here is my goal and my last twelve weeks of load. Before you write a single week of a plan, tell me whether this goal is reachable from this starting point, in this time, on these hours. Give me the fitness I would need at the event and the weekly ramp to get there from where I am. If the ramp is more than is safe, say the goal or the target time has to move, and tell me what is reachable. Do not write the plan until you have answered this.

Give it: the event, the date, your hours, twelve weeks of load.

Watch for: a plan appearing anyway.

Build the plan.

Build me a plan from today to my gran fondo on 18 October: 140 km, 2,400 m of climbing, goal under five hours. I have 6 to 8 hours a week on the days in my brief. Show the phases, the purpose of each week, a rest week every third or fourth week, and the taper. Then give me the first two weeks in detail, each session in Intervals.icu step syntax, one step per line. Sum the hours and show me the sum. Do not exceed my hours.

Give it: the brief, the event, your last six weeks.

Watch for: no rest weeks, intensity in week one, or a total above your hours. Each one is common.

Should I ride today?

HRV this morning 61, my 30-day average 68; both are rMSSD in milliseconds from the same watch, and my usual day-to-day swing is about 6. Resting heart rate 52 against a usual 48. Sleep 5 h 40 min against 7 h 10 min. Yesterday 90 minutes easy; the day before 4 x 8 min at threshold. Today's plan is 3 x 12 min sweet spot. Before you decide, state the rule you will apply. Then apply it and decide: ride as planned, ride easy, or rest. Say which numbers triggered the decision and which did not. Ignore how I say I feel. Then tell me what tomorrow becomes under each answer.

Give it: this morning's HRV with its unit and source, resting heart rate, sleep, the last two days, today's plan.

Watch for: a verdict with no rule in front of it. A rule stated afterwards was written to fit the verdict.

Both cases.

From the numbers above, build the case for training as planned and the case for resting. For each, list the numbers that support it and the numbers that argue against it. Do not use anything I said about how I feel. Then say which case has more support, by how much, and name the one measurement that would most change the answer if I took it now.

Watch for: agreement with whichever way you leaned in the question.

What did that ride say about me?

Attached is Saturday's ride as a per-second CSV with time, power, heart rate, cadence and altitude. FTP 265, threshold heart rate 168. Load it in code and first tell me the row count, the elapsed and moving time, and any gaps over ten seconds. Then find the longest steady segment after the first 20 minutes with no stops, split it in half, and give me power divided by heart rate for each half and the drift as a percentage, with both halves' averages shown so I can check you. Then normalized power, intensity and load for the whole ride. One thing to change next time. No praise.

Give it: the ride file from Intervals.icu or a TCX from Garmin Connect. Garmin's "Export to CSV" is the lap table, not the ride.

Watch for: a drift number with no segment named. Drift over a ride with a cafe stop in it is noise.

Why did that feel so hard?

Yesterday's 4 x 8 min at 100% FTP felt like 110%. Here is the ride file and the last ten days. Rank the likely reasons: fatigue, sleep, fuelling, heat, an FTP set too high. Say which one the data supports and what would confirm it.

Give it: the ride, ten days of load, sleep if you have it, the temperature.

Watch for: "it could be many things". Make it rank.

Was that a personal best?

Here are my best 1, 5, 20 and 60-minute powers from the last 12 months with dates, and today's ride. Tell me whether today set a new best at any duration, and what that says about where my fitness is compared with three months ago.

Give it: your bests by duration (Intervals.icu's power curve or Strava's Power Skills show them; copy the numbers) and today's ride.

Plan my week around the hours I have.

This week: Tuesday 60 min indoors, Thursday 60 min indoors, Saturday 2 h outdoors, Sunday 3 h with the club and it will be hard whatever you write. FTP 265, last week's load attached. Hard rules, check each at the end and print pass or fail: total time under 7 hours; no two sessions above threshold on consecutive days; Sunday counted as a hard group ride, not a planned interval session. For each session give the purpose, duration, target watts and percent of FTP, and the workout in Intervals.icu step syntax, one step per line, so I can paste it and send it to my head unit.

Give it: your hours per day, last week's load, FTP.

Watch for: two hard days in a row, and a total that ignores your cap. The pass-or-fail list is there so it has to check.

Review last week.

Here are last week's seven days against the plan you gave me. Score how I executed each session, say whether the week did what it was meant to, and tell me the one thing to keep and the one thing to change.

Give it: the seven rides, and the plan pasted back in. It does not remember last week.

Race weekend.

I have a road race Saturday, about 90 minutes, and a criterium Sunday, 45 minutes. Both matter. Here is the week so far and my load. Give me Wednesday to Friday: what stays, what goes, and the openers on Friday, with the reason for each.

Give it: the two events, the week so far, your load.

Watch for: a taper written for a gran fondo. Two short races need you fresh and a little sharp, not a week of nothing.

Repair, do not rebuild.

I missed Tuesday and Wednesday, work. I still have 40 min Thursday, 60 min Friday and 3 h Sunday. This week's purpose was threshold volume, week three of four before a rest week. Repair the week: keep the purpose ahead of any single session, move what you must, drop the rest. If the purpose can no longer be served, say so and tell me to bring the rest week forward instead of patching. List what moved and why. Do not add sessions to make up the lost ones.

Give it: the current week's plan and its purpose, your remaining hours.

Watch for: catch-up training. Missed sessions are gone; a model will try to squeeze them back in.

I was ill.

I had a cold from Monday to Friday, no riding, no fever now. My race is in five weeks. Tell me how to come back: how many easy days before intensity, what the first hard session should be, and what changes in the race-week plan. Tell me the symptoms that mean I see a doctor instead of coaching around it.

Give it: the illness dates, the race date, the plan.

Watch for: intensity in the first two days back.

Two weeks off.

I have not ridden for 16 days: work travel. Before that I was at 8 hours a week. Give me a two-week return that rebuilds without a big first week, and tell me what I have most likely lost and what I have kept.

Watch for: a first week back at your old volume.

Estimate my FTP.

Attached are my best powers by duration from the last 90 days, one row per duration from one second to one hour, each with the date it was set. My FTP is set at 265. Estimate threshold three ways, in code, showing the arithmetic: 95 percent of my best 20 minutes, and say whether that effort had a hard five-minute opener before it or your estimate is unanchored; a critical-power fit over the 3 to 12 minute range, with the residuals; and whatever your modelled curve gives. Then list the dates each estimate leans on, say whether they describe the same fitness, say which you trust least and why, and give one number to set.

Give it: the power curve with dates (Intervals.icu exports it).

Watch for: the 20-minute number times 0.95 presented as truth, and a 60-minute "best" from a group ride dragging the estimate down.

Zones from a test.

My threshold heart rate is 168: the average of the final 20 minutes of a 30-minute solo test on 3 September. Do not use max heart rate and do not ask my age. Give me zones in the Coggan five-zone scheme and the Friel seven-zone scheme side by side, each as a percentage range and a beats-per-minute range, and tell me in one line where they disagree so I know which app is showing me what.

Give it: a real test result. Never your age alone.

Watch for: 220 minus age anyway. If it appears, the model ignored your test.

Explain the number my app gave me.

My app says fitness 58, fatigue 71, form minus 13, using the usual model: fitness a 42-day weighted average of daily load, fatigue a 7-day one, form the difference. Attached is my daily load for the last 120 days including zeros. In code, reproduce the three numbers and tell me how far you land from the app's. If you cannot match them within a couple of points, say what is missing rather than explaining the mismatch away. Then say in plain words what this state means and whether minus 13 ten days before a race is where I want to be.

Give it: the three numbers, their definitions from the app, 120 days of daily load. Four weeks is not enough to reproduce a 42-day average.

Watch for: confident definitions of your app's private metric. Paste the app's own.

Pace this course.

Course: 140 km, 2,400 m of climbing. Climbs at km 30 (4.2 km at 6.1%), km 65 (8.0 km at 4.8%), km 95 (3.1 km at 8.2%), km 120 (5.5 km at 5.5%). Rider 78 kg plus 9 kg of bike and kit, FTP 265, goal under five hours, in a group on the flat and alone on the climbs. In code: for each climb, a target in watts, percent of FTP and watts per kilo, and the climbing time that gives. A group-riding power for the flat with its time. Sum it. If the total is over five hours, tell me the goal is not reachable at safe targets instead of bending the targets until it fits. Then the two kilometre markers where I am most likely to blow up and the warning sign at each.

Give it: the climbs with lengths and gradients (a GPX, or the numbers), FTP, weight.

Watch for: climb targets above FTP for a five-hour event, and a total that was made to fit.

Fuel it.

Five hours, moderate to hard, 22 degrees. I weigh 78 kg. I have used gels and a 60 g carbohydrate drink mix before without trouble and never trained above that. Give me carbohydrate per hour, and where you go above 60 g say what mix of sugars that needs and why. Sodium and fluid per hour with your sweat-rate assumption stated so I can correct it. A timing plan from the start, the evening before, and breakfast with the timing. If your number is above what I have tolerated, give me the gut-training ramp to get there and how many weeks it needs, and do not put an untested intake in the race plan.

Give it: duration, intensity, temperature, weight, what your stomach has handled.

Watch for: 90 g an hour recommended to someone who has never trained their gut.

Taper.

The race is in 14 days. My last big week is attached. Give me the taper day by day: volume down by how much, which intensity stays, when the last hard session is, and the two days before the race.

Watch for: a taper that cuts intensity along with volume. Ask why the openers are where they are.

Train around the knee.

My right knee aches on the outside after three hours in the saddle, no swelling, fine walking, fine on the bike under an hour. Before you touch the plan, ask me the fit questions that matter: saddle height and setback, anything changed recently, cleat position and float, whether it is one leg only. Do not diagnose. List the signs that mean I see a physio or a bike fitter instead of changing training. Only then tell me what in the attached five weeks to change and what to keep.

Give it: the symptom in plain words, the plan.

Watch for: a model coaching around something a fitter fixes in twenty minutes.

Strength that does not wreck the riding.

Two 40-minute gym sessions a week, a barbell and a rack, six years of riding and no lifting history. Put them into the week attached so they do not land the day before a hard ride, and give me the sessions.

Watch for: gym the day before intervals, and a program written for someone who lifts.

Lose weight without losing the season.

I want to go from 78 to 74 kg by the race in 12 weeks while keeping my threshold. Here are my hours and my rides. Tell me the deficit that is safe for this training, which rides to fuel fully no matter what, the weekly rate of loss above which I should stop, the signs I am under-fuelling that I would notice first, and the point at which you would send me to a sports dietitian instead of coaching me.

Give it: weight, target, hours, the plan.

Watch for: a deficit that ignores your long rides, and no mention of when to stop.

I am 58.

I am 58, riding 8 hours a week, thirty years on the bike. A hard Sunday now costs me until Wednesday. Take the week attached and change it for that: where the hard day goes, how many easy days follow it, what I drop, and what I keep doing that a younger rider would not need. Edit my week; do not describe a principle.

Give it: your age, hours, how long a hard day costs you now, the week.

Watch for: generic advice about masters athletes. If it explains physiology and moves no sessions, ask again.

Tell me the truth.

Here are my last three months and my best powers by duration. What kind of rider am I: where am I strong, where am I weak, and what does the power curve show that I have not wanted to admit? Be direct. Then name the one limiter that would move my gran fondo time the most.

Give it: three months of rides and the power curve.

Watch for: flattery. Ask it what you are worst at first.

Compare this year with last.

Attached are two files, each one row per week: week start, hours, load, time in each of five zones, and the best 20-minute power set that week. June to September, this year and last. In code, make a side-by-side table: total hours, total load, share of time above 90 percent of FTP, best 20-minute power with its date. Print the table before any conclusion. Then say whether I am fitter, and name the comparison that most supports your answer and the one that most contradicts it.

Give it: two weekly summaries, same months. Not the raw rides; a season of per-second data will not fit.

Watch for: a conclusion with no table in front of it.

Threshold pace from your best efforts.

Best recent efforts: 5 km 21:40 on 2 September, 10 km 45:30 on 20 August, both flat, solo and all out. Estimate my threshold pace two ways and show both: a Riegel extrapolation with the exponent stated, and a VDOT table lookup with the value used. Threshold is roughly one-hour race pace, so say which of my two results is the better anchor and why. Then give easy, threshold and interval paces per kilometre, and how much the two methods disagree.

The missed long run.

I missed Sunday's 28 km, the longest in the block. The marathon is in five weeks. Tell me whether to move it, shorten it or drop it, and what the next two weekends become.

Three sports, one week.

Swim Tuesday and Thursday mornings, bike Wednesday evening and Saturday, run Monday and Sunday, gym Friday. Here is last week's load across all three. Put the hard sessions where they will not collide, and tell me which day is the one I must keep easy.