Over the last couple of months I've become the unofficial AI-fitness guy in my group chat, which is a sentence I never expected to type. People send me their plans and ask why theirs is 'mid' and mine is good. We are using the exact same prompt and the exact same model. The difference, every single time, is the brief. The Weekly Training Block From Your Goal, Gear, and Real Schedule prompt is only as smart as the honesty you feed it.
A vague brief — 'build muscle, three days, dumbbells' — produces a perfectly competent beginner plan that ignores everything specific about you. A precise brief produces a program you can run for a season. Here's the difference, field by field, learned from watching people get it wrong.
The fields that decide everything
- Goal: 'lose 5kg and build visible arms' beats 'get fit' — specificity tells the model how to bias volume and which lifts to prioritise
- Equipment: list everything, including the kettlebell under your bed you forgot about — a missing item becomes a missing exercise
- Limits: name every cranky joint and every movement you avoid; the model can't program around a knee it doesn't know exists
- Time: put your real cap, not your fantasy cap — '45 minutes' gets you a 45-minute plan, '90 minutes' gets you a plan you'll quit
- Experience: 'intermediate but undertrained' is a legitimate, useful input — it changes starting loads and progression speed
The mistake almost everyone makes
People inflate their availability and lowball their limitations, because we all want to be the version of ourselves who trains five days a week and has no injuries. Then they get a five-day plan, do it twice, miss Wednesday, feel guilty, and abandon the whole thing. When my friend Priya re-ran it with her actual three days and her actual bad wrist, Claude Opus 4.8 swapped barbell curls for cable work and gave her a plan she's still doing six weeks later. The honest brief isn't the modest choice — it's the effective one.
The second-most-common mistake is being vague about equipment, and it's sneakier because it feels harmless. My mate Tom wrote 'dumbbells and a bench' and got a perfectly fine plan — but he also owned a pull-up bar and a kettlebell he'd forgotten to mention, so his whole posterior chain was under-served by a program that didn't know they existed. One re-run with the full inventory and suddenly he had loaded carries and pull-ups, and his back actually started growing. The model can only build with the bricks you hand it. List the kettlebell under your bed.
There's a second-order lesson here too: the prompt's structured output is doing quiet work. Because it forces a table of Exercise, Sets x Reps, Rest, and a coaching note, you can't get the hand-wavy 'do some pulling movements' answer that makes a plan useless. Every line is a thing you can actually go and do, with a number attached. I've come to think of that table as a filter — it physically prevents the model from giving you the kind of mushy non-advice that reads well and means nothing.
Why Claude specifically, and why the structure matters
People ask why I default to Claude Opus 4.8 for this one when there are faster, cheaper models around. The answer is that this prompt leans hard on following an explicit structure and respecting a stack of constraints at once — the time cap, the injury list, the equipment limit, the rest-day rule — and Opus is unusually obedient about holding all of those in its head simultaneously without quietly dropping one. I've watched lighter models cheerfully ignore a stated 45-minute cap and hand back a 70-minute session, because honoring six constraints together is genuinely harder than it looks. For a workout you're going to run for a month, I'd rather wait the extra few seconds and get a plan that actually fits the box I drew.
What to do with the plan once you have it
Don't tweak it for four weeks. I mean it. The temptation to optimise on day three — to swap an exercise because TikTok said something — is how people end up back at freestyle. Run the block, log it, and only re-brief the prompt when you have real data: a lift that stalled, a session that consistently runs long, a movement that started bothering a joint. Then update the relevant field and regenerate. The loop — brief, run, log, re-brief — is the entire game.
Once I have a few weeks logged, I feed that log into my Are You Actually Getting Stronger? Honest Read of Your Training Log prompt to find out what's working before I touch the plan. Data first, opinions second. Grab the Weekly Training Block From Your Goal, Gear, and Real Schedule prompt on Prompt Dock, write the brief like your future self is the one who has to do the workouts — because they are — and you'll skip the generic version entirely. The prompt is good. The brief is what makes it yours.