Can ChatGPT actually write you a good workout? (I tried it)

TL;DR

• ChatGPT can write a workout plan that looks good and is mostly safe. The catch is how partial it is. When researchers scored an AI chatbot's exercise advice against gold-standard guidelines, it was 90.7% accurate but only 41.2% comprehensive (Zaleski, Berkowsky, Craig and Pescatello, JMIR Medical Education, 2024).
• The bigger problem is not accuracy. It is that a chatbot writes the plan once and then stops. In an expert review of ChatGPT's training programs, it "offered no potential real-time adjustments, or revisions to training protocols, based on feedback or individual progression" (Washif et al., Biology of Sport, 2024).
• A plan that never updates hands the deciding back to you every single week. You become the coach, which is the exact job you went to AI to avoid.
• Actus does the part ChatGPT can't: it re-plans each week off your own logged training, so the next session is already decided when you walk in.

So I tried it. I sat at the kitchen table one night, typed my lifts, my three days a week, and my goal into ChatGPT, and asked it for a program. Thirty seconds later I had a clean four-day split. Sets, reps, the works. It looked like something a coach would hand you.

Week one felt great. Then week two my schedule fell apart, I missed a day, and the plan just sat there, exactly the same, with no idea any of that had happened.

This post is about that gap. Not whether ChatGPT can write a plan. It can. About what happens after it does.

Can ChatGPT actually write you a good workout plan?

ChatGPT can write a workout plan that is broadly safe and follows standard training principles, but it is only partially complete. When researchers scored an AI chatbot's exercise recommendations against gold-standard guidelines, the advice was 90.7% accurate yet only 41.2% comprehensive (Zaleski, Berkowsky, Craig and Pescatello, JMIR Medical Education, 2024). What it tells you is mostly right. It just leaves a lot out.

That matches what I saw. The split it gave me was reasonable. The lifts made sense, the weekly layout was fine, nothing was dangerous. If you have never followed a structured program, it is genuinely better than wandering the gym floor picking machines at random.

The trouble is you do not know what is in the missing 59%. If you already know enough to spot the gaps, you did not really need the plan. If you do not, you cannot tell what it skipped. That is the quiet problem with asking AI to coach you: it sounds equally confident whether it is right or whether it is leaving out the thing that mattered for you.

So why does the plan it gives you feel a bit generic?

A ChatGPT plan feels generic because it is built from the average of everything written about training, not from anything about you. Expert reviewers found AI-generated programs "lack specificity, progression, and adaptability," often defaulting to safe and general rather than tailored to the person in front of it (Washif et al., Biology of Sport, 2024). It writes a sensible plan for someone roughly like you, not the plan for you.

You can feel it in the output. It does not know that your left shoulder hates overhead press, that you actually have 45 minutes not 60, or that you have stalled on the same squat number for two months. You can paste all of that in, and it will fold some of it into the answer, but it is still pattern-matching a typical case rather than reading your actual training.

A real coach earns specificity over time by watching what happens. The chatbot has no memory of last week, so every plan starts from the generic middle.

What happens to a ChatGPT plan after week one?

Nothing happens to it, and that is the real problem. A ChatGPT plan is a one-time document: it is written once and never changes, because the chatbot has no idea how your week actually went. In an expert evaluation, ChatGPT "offered no potential real-time adjustments, or revisions to training protocols, based on feedback or individual progression" (Washif et al., Biology of Sport, 2024). It hands you a plan and then goes quiet.

This is exactly where mine broke. After I missed that day in week two, the plan still wanted three clean equal sessions I no longer had. So I went back, re-explained what I had actually done, and asked it to adjust. It did. Then the same thing happened the next week. And the week after.

That is the part nobody mentions. To keep a ChatGPT plan useful, you have to re-prompt it every time real life touches your week. You become the coach feeding the assistant. The deciding you wanted to hand off is right back on your plate, just with an extra step.

Isn't the real problem just that I need better prompts?

Better prompts help, but they cannot fix the core gap, because the limit is not how you ask. It is that ChatGPT has no ongoing record of your training. A great prompt produces a better first plan. It does not give the chatbot a memory of your last eight sessions or the judgement to weigh them. Progression and week-to-week adaptation are exactly what expert reviewers found AI plans missing (Washif et al., Biology of Sport, 2024).

So you can get very good at prompting and still be stuck doing the work yourself. Every week you are the one logging what happened, summarising it, pasting it back, and deciding what to change. The prompt is just the interface for a job that is now yours.

Prompting harder optimises the wrong thing. It makes the document better. It does not take the recurring decision off your plate, which was the actual point.

What's the difference between a plan and an app that decides?

A plan is a static document you have to interpret and update yourself. An app that decides is a system that reads your real training history and re-plans the next session for you. Actus is the second kind: it builds and weights today's session from what you actually logged, so the deciding is already done when you arrive. A ChatGPT plan is step one frozen in time. Actus runs the loop every week.

The difference shows up on a bad week. When your schedule breaks, a frozen plan still expects the clean version of you that never existed. Actus works off what you really did, not what you were supposed to do on Tuesday, so the next session reflects reality instead of ignoring it.

I built Actus because I was tired of being my own re-prompting machine. I did not want a better document. I wanted to walk in and already know, every week, without sitting at the kitchen table explaining my own training to a chatbot again.

So is ChatGPT useless for training?

No, ChatGPT is genuinely useful for training, as long as you use it for what it is good at: explaining a concept, suggesting an exercise swap, or sketching a first draft. It is a strong thinking tool and a weak coach. The line is simple. Ask it to teach you something once and it shines. Ask it to manage your training week after week and the lack of memory and adaptation catches up with you (Washif et al., Biology of Sport, 2024).

Use it to understand why progressive overload matters, or to find a substitute when the rack is taken. That is a great use of a chatbot, and you should.

Just be honest about what it is not doing. It is not watching your training, not remembering last week, and not deciding tomorrow for you. The moment you want that part handled, you have outgrown what a one-time plan can do.

FAQ

Can ChatGPT replace a personal trainer? Not really. ChatGPT can explain training and draft a reasonable plan, but an expert review concluded it "cannot replace the judgement and empathy of a human practitioner" and works best as a supplementary tool, not a replacement (Washif et al., Biology of Sport, 2024). It has no memory of your sessions and makes no real-time adjustments, which are central to what a coach actually does.

Is a ChatGPT workout plan safe? Mostly, but with gaps. One study found AI chatbot exercise advice was 90.7% accurate, with most of the inaccuracy relating to when someone should get medical clearance before starting (Zaleski et al., JMIR Medical Education, 2024). Treat its plan as a draft, not a prescription, and check anything that affects an injury or health condition with a qualified professional.

Why does my ChatGPT plan stop working after a few weeks? Because it never updates itself. The plan is written once and has no idea how your weeks actually go, so it keeps expecting the schedule and progress you had on day one. Keeping it useful means re-explaining your training to it regularly, which puts the deciding back on you.

What should I use ChatGPT for in the gym then? Use it for one-off questions: explaining a concept, suggesting an exercise swap, or sketching a starting structure. It is a good teacher for a single question. It is a poor manager of an ongoing training week, because it cannot track or adapt to what you actually do.

How is Actus different from asking ChatGPT? Actus reads your logged training history and decides your next session for you, re-planning each week from what you actually did. You do not paste your week into a chat or re-prompt anything. You open the app and the session is already laid out.

About the author

Pedro Rodrigues is the founder of Actus and has been training since 2014. By day he is Operations Director and the AI Innovation lead at Monks, one of the largest marketing agencies in the world, where his job is putting AI to practical use at scale. He spends his working life building with these tools, which is exactly why he is clear-eyed about where a chatbot helps with training and where it quietly hands the work back to you. Connect with him on LinkedIn.

This article is for educational purposes and does not constitute medical or clinical advice. Strength training carries injury risk. Stop if you feel pain and consult a physician before starting any new exercise programme.

If you want an app that decides your session for you each week, based on your own training history, instead of a plan you have to keep re-explaining, get Actus on the App Store.