Coaching prompts that work

Most training apps ship one coaching model and you get what you get. Parameter ships none. It holds your plan, your equipment, your history and your logs, and it lets whichever AI you already pay for do the thinking.

That changes which prompts are interesting. “Give me a workout” wastes it. The prompts below all have the same shape: your AI goes and does something Parameter has no ability to do on its own, checks it against your real numbers, and pushes back finished workouts.

What your AI can reach beyond Parameter depends entirely on which assistant you use and what you have connected to it. Web search, document reading and other MCP servers are commonly available, but may not work everywhere. Parameter’s half never changes: it answers with your actual data, and it accepts a plan. Where a prompt below needs something extra, it says so.

Have it research a method properly, then program it

Your AI can go read about a training methodology in depth, then apply it to your twelve weeks of real numbers. No fitness app is going to add support for a method three thousand people use.

Research how Norwegian singles actually work for a runner at my weekly volume. Use primary sources and coaches who publish their reasoning, not blog summaries. Then read my last twelve weeks in Parameter and tell me honestly whether I have the aerobic base to start it. If I do, build the first four weeks and push them. If I don’t, tell me what to do for two months instead.

Swap in whatever you are curious about: 5/3/1 with a running block on top, polarized versus threshold distribution, block periodization, a masters lifter’s frequency, heat acclimation before a summer race. The useful half of the prompt is the second half, where it has to check the method against what you have actually been doing rather than describing it in the abstract.

Give it your coach’s PDF, a book chapter, or a physio protocol

If your assistant can read documents, a training source stops being something you interpret by hand every week.

Attached is the rehab protocol my physio gave me. My status in Parameter is set to injured. Turn phase two into real sessions using only the equipment I own, keep every load inside what the protocol allows, and tell me if anything already in my plan breaks it.

The same works for a program out of a book, a race-specific plan a coach wrote for someone else, or the four pages of notes you took at a clinic. Your AI reads the source, resolves it against your equipment list and your history, and the output is scheduled sessions rather than a PDF you keep meaning to follow.

Setting your status first matters here. A status other than ok is a constraint your AI plans around, so it does real work rather than sitting there as a note to yourself.

Let it look out the window before planning

Training doesn’t happen in a vacuum. If your assistant can search the web, it can make sure your plan acknowledges what is going on in the real world.

Check the forecast here for the rest of the week. Put my long run on whichever morning is driest and coolest, move the strength day off that morning, and leave Sunday as a rest day.

Useful variations: schedule around air quality during fire season, put the track session on a day the local track is not booked for a meet, or move a session because sunset now beats you home.

Plan against a city you have not arrived in yet

Travel is where a generic plan falls apart, and it is where an AI that can look things up is worth the most.

I’m in Chicago Tuesday through Friday, staying near the Loop. Find out what the hotel gym actually has in it and what is runnable nearby, rebuild those four days around what is really there. Keep Thursday’s intervals if there is a track I can get on.

It comes back with sessions built for dumbbells that go up to 50 pounds and a lakefront path, because it went and checked, instead of assuming a full rack. Add your flight times and it will stop programming a hard session for the morning after a red-eye.

Combine Parameter with your other data

Parameter deliberately holds training data only. It has no idea what you eat or how you slept. Your AI can hold both, if you have those sources connected to it.

Pull my last three weeks from Cronometer alongside my training from Parameter. My hard sessions are Tuesday and Saturday. Tell me whether I am actually fueling those days, and if I am not, either fix the intake or move the sessions to days I do.

The same pattern covers sleep and recovery data, a calendar that knows when your work travel lands, and bodyweight trending against a cut. This needs the relevant server connected to your assistant. Parameter’s side is the training history, and that is all it contributes.

Adapt from the gym floor, mid-session

You do not have to accept the session in front of you. Many people don’t think to ask for adjustments while they are standing in the gym.

Squat rack is taken and I have 35 minutes. Same stimulus, different equipment, and do not drop the main lift.

Or, when the machine is the constraint rather than the clock:

This treadmill tops out at 10 miles an hour and today’s intervals ask for 11. Rebuild the session inside what the machine can hold and keep the total work the same.

You can also adjust these types of things from the workout execution screen, so you can take the answer and apply it without leaving the workout. Substitute swaps an exercise from the set row, and Adjust changes a target mid-segment, with a scope control that decides whether the change applies to just one segment or the rest of the workout.

Ask it to argue with your plan

Assistants default to encouragement. You have to explicitly ask for pushback, and the planned-versus-actual log is what makes the answer honest rather than generic.

Read my last eight weeks and make the strongest case that my plan is wrong. Where am I genuinely progressing, where am I just accumulating fatigue, and what would you cut if I told you I only had four hours a week?

Here are some followups that can add more value: “which of these sessions did I skip more than once, and what does that tell you” and “check my notes from Tuesday before you plan legs.” Notes you dictate mid-workout are timestamped to the set or segment you were on, so a mumbled observation about a cranky knee is sitting there attached to the exact set that caused it.

Give it a standing job

Make your planning process even easier. If your assistant supports scheduled or recurring work, hand it the loop.

Every Sunday evening, read what I actually did last week against what was planned, then write next week and push it. Keep my Tuesday and Thursday strength days. If I missed more than two sessions, make next week easier instead of trying to catch up.

Where the answers end up

Everything above lands in Parameter as real workouts on real days, not as chat you have to transcribe. Open the Plan tab and it is there, and you can still edit any of it by hand.

Parameter’s job is to hold accurate data and execute the session well. Deciding what the session should be is the part your AI is already good at, and it gets better every time the model behind it does, without anyone shipping an app update.