LearnBuild the program
Can you use ChatGPT to write workout programs for clients?
The short answer
Yes, for a first draft. Not for the finished program. In published tests, coaches rated AI-written plans as safe and sensibly structured but generic: weak on progression and vague about how hard to train. More detailed prompts produced better plans. Check every draft against the person it is for, keep their health details out of the prompt, and make the final decisions yourself.
2 to 4
out of 5: how coach ratings rose as the prompt gained detail1
90.7%
accurate, but only 41.2% comprehensive, on exercise advice for clinical groups2
0 of 3
chatbots gave enough detail on how hard to train3
Many trainers already do this, and many clients arrive with a plan a chatbot wrote for them. What matters is where these tools are reliable and where they are not. That has been tested, by coaches, more than once.
What did the studies find?
| Study | What was tested | What the experts found |
|---|---|---|
| Dergaa and colleagues, 20244 | GPT-4 wrote 30-day programs for five hypothetical people | Safe, but lacking precision for the person's condition and goals, and "prioritizing excessive safety over the effectiveness of training". Progression was mostly in duration, not intensity or load |
| Washif and colleagues, 20245 | GPT-3.5 and GPT-4 wrote 12-week resistance programs | The plans reflected training principles such as progressive overload, but were suited to use as a draft that needs human expertise |
| Düking and colleagues, 20241 | Ten coaches rated three running plans from prompts of increasing detail | Median ratings of 2, 3 and 4 out of 5 as the prompt got more detailed. Even the best plan scored below neutral for how training volume progressed |
| Havers and colleagues, 20253 | Ten coaches rated muscle and strength plans from three chatbots on 27 criteria | "Moderate" quality, with most plans under 4 out of 5. Plans lacked detail on intensity and leaned conservative |
| Zaleski and colleagues, 20242 | Exercise advice for 26 clinical populations, checked against established guidelines | 90.7% accurate but only 41.2% comprehensive. Most of the inaccuracies concerned when to get medical clearance. The text read at college level |
Three findings repeat across them.
The structure is fine. Sensible splits, sensible exercise choices and a recognisable shape. For a new trainer that is the slow part of a first draft. For an experienced one it is typing saved.
The hard decisions are missing. How much load, how close to failure and how it gets harder from week to week. In the 2025 study, the missing detail on intensity was a limitation across every model tested.3 Those decisions are the coaching.
The prompt matters. The running study saw the median rating go from 2 to 4 out of 5 as the prompt gained detail.1 That is ten coaches rating three plans, so read it as a direction, not a measurement. A later study found that the same detailed prompt gave plans of the same quality each time.6
- Least detailed prompt2
- More detail3
- Most detailed prompt4
Median rating for each of three plans. The only change was how much the prompt said.1
Where does it help, and where does it fall short?
| Good use | Why | Poor use | Why |
|---|---|---|---|
| A first draft of a week | Structure is what it does well | A finished program, sent unread | Progression and intensity are its weak points |
| Swaps for an exercise | Fast, and easy for you to check | A plan for someone with a medical condition | Most inaccuracies in testing were about medical clearance2 |
| Rewriting your shorthand in plain language | The decisions are already yours | Deciding loads for a real person | It has no record of what they lifted or how it felt |
| Translating a program you have written | Same reason | Advice outside your scope, such as diet plans or injuries | Your professional limits still apply |
What should a prompt include?
Everything you would ask a new client, minus anything that identifies them. A brief that states the person, the constraints and the format gets a draft you can work with.
Prompt 1: the brief
Act as a strength coach writing a draft that I, a qualified trainer, will review and edit. Write week 1 of a training program for this person: - Goal: [one outcome, such as a first pull-up or a stronger squat] - Training experience: [months or years, and what they have done] - Days per week: [number], about [minutes] per session - Equipment: [full gym / dumbbells and a bench / bodyweight only] - Movements to avoid: [any, described generally, with no medical details] Rules: - 4 to 6 exercises per session, hardest lift first. - For every exercise give sets, a rep range, reps in reserve and rest time. - Letter the order (A, B1, B2) and say how many rounds a superset runs. - Use one standard name for each exercise. - Give one easier and one harder swap for each main lift. - Do not add nutrition, supplement or medical advice.
Prompt 2: make it progress
Here is week 1 of the program: [paste it]. Write weeks 2 to 4. - Keep the same exercises, so progress is visible. - Use double progression: add reps within the range, then add load and return to the bottom of the range. - State the rule for adding load in one line the client can follow alone. - Make week 4 lighter: the same exercises with about one third fewer sets. - Show each week as a table and list what changed from the week before.
Prompt 3: rewrite it for the client
Rewrite this program so a beginner can follow it alone in a gym, reading on a phone: [paste it]. - Short lines. No abbreviations unless explained once. - Keep every number exactly as written. - Add one coaching cue per exercise, 10 words or fewer. - Add one line each for: the set felt too easy, the set felt too hard, the equipment is taken.
What should you check before a client sees it?
The eight-point check
What about privacy and responsibility?
Two things are worth knowing before you paste anything in.
The responsibility stays with you. OpenAI's terms say you should not rely on output "as a sole source of truth or factual information, or as a substitute for professional advice", and that you must evaluate it "for accuracy and appropriateness for your use case, including using human review as appropriate".7 An article in a National Strength and Conditioning Association journal makes the same point from the trainer's side: blind use of a generated program could lead to a liability claim, and the tool is better suited to generating ideas than to prescribing.8
What you type in is your responsibility too. The same terms say you must have the rights to anything you input, and that content may be used to improve the service unless you opt out.7 The simple rule: describe the client in general terms. No names, no dates of birth, no medical history, no screening forms.
What if a client brings you a plan a chatbot wrote?
It happens, and it is a good opening, not a threat. Run the same eight-point check on their plan, in front of them. The gaps it shows are the ones the studies found: no real progression, no stated effort, nothing specific to them. That conversation explains what a coach is for better than any sales page.
Should you tell clients you use it?
We found no professional body that requires it. Clients pay for your judgement, and that is still what they are getting if you draft with a tool and decide yourself. If a client asks, a plain answer costs nothing: "I draft with it. Every number in your program is my call." What damages trust is a plan that reads as if nobody checked it.
How DUAL handles an AI draft
Paste a draft from ChatGPT into DUAL and it becomes a session people can follow, with exercises matched to a demo where the library has one, and the supersets grouped. You edit anything before you share it, and nothing is published until you say so.
Paste a workout and see itQuestions people ask
It writes reasonable first drafts. In published tests, coaches rated AI-written plans as moderate in quality: safe and well organised, but generic, cautious and thin on progression and intensity. Plans improved when the prompt was more detailed.
No. A draft is a starting point, like a template. It becomes a problem when the draft is sent unchecked, when a client's private details are typed in, or when the tool is used for advice outside a trainer's scope.
The goal, training experience, days per week, session length and equipment, plus rules for the output: how many exercises, a rep range, an effort target and rest time for each, and standard exercise names. Leave out anything that identifies the client.
The studies do not name a clear winner. One compared two chatbots and rated one higher on several criteria.6 Every tool tested shared the same weak points: progression and intensity. A detailed prompt is the part you control.
Sources
Every source was read and checked on 10 November 2026. Spotted something out of date? Tell us.
- Düking P, Sperlich B, Voigt L, Van Hooren B, Zanini M, Zinner C. ChatGPT generated training plans for runners are not rated optimal by coaching experts, but increase in quality with additional input information. Journal of Sports Science and Medicine, 2024.
- Zaleski AL, Berkowsky R, Craig KJT, Pescatello LS. Comprehensiveness, accuracy, and readability of exercise recommendations provided by an AI-based chatbot: mixed methods study. JMIR Medical Education, 2024.
- Havers T, Jelonnek C, Masur L, Isenmann E, Sperlich B, Geisler S, Düking P. A professional assessment of training plans for muscle hypertrophy and maximal strength developed by generative artificial intelligence. Biology of Sport, 2025.
- Dergaa I, Ben Saad H, El Omri A, et al. Using artificial intelligence for exercise prescription in personalised health promotion: a critical evaluation of OpenAI's GPT-4 model. Biology of Sport, 2024.
- Washif JA, Pagaduan J, James C, Dergaa I, Beaven CM. Artificial intelligence in sport: exploring the potential of using ChatGPT in resistance training prescription. Biology of Sport, 2024.
- Havers T, Masur L, Isenmann E, Geisler S, Zinner C, Sperlich B, Düking P. Reproducibility and quality of hypertrophy-related training plans generated by GPT-4 and Google Gemini as evaluated by coaching experts. Biology of Sport, 2025.
- OpenAI. Terms of use. Effective 1 January 2026.
- Chelette A, Whitehead R. The use of artificial intelligence applications for exercise prescription by personal trainers: a call to action. NSCA Personal Training Quarterly, 2023.