On this page
I asked ChatGPT to build my plan
In December 2025 I wrote a prompt that asks ChatGPT the questions I would ask a new client, and I used it to build a plan for myself, on camera, from start to finish.
The questions are what make it work. Mine ask how active the person is through the day, honestly: an office job is sedentary even with a few walks, a teacher or a sales assistant is moderately active, and a builder is very active. They ask for an average step count, which most phones already track. They ask for a specific goal for the next six months, with a number in it, so it can work out the calories. They ask how much time there is to train, what equipment, and whether it is a gym or home. And they ask how the person likes to eat, in as much detail as they can give.
What came back covered everything: a calorie target, a protein target, a training split with rep ranges and rest times, rules for when to add weight, and a daily step goal. It even chose the exercises I would have chosen. I was honestly impressed, and I would call it expert advice on the diet and on the training plan.
Some of its advice was exactly right. Never try to make up missed workouts, because stacking them up is how people overtrain, get injured or burn out. Never slash calories after a day of eating more, because that habit can turn into an unhealthy relationship with food.
So yes, ChatGPT can write a training plan, and a good one. It can calculate the calories and structure the training. That is exactly why I say plans are cheap.
Where it still needed a coach
Even a plan that good had details I would change.
- The calories. It put my maintenance at 2,700 calories a day. That is a little on the high side, which you see with a lot of these calculators, so I would trim 200 to 300 calories to start, see how it goes, and adjust from there.
- Training to failure. It said to leave one or two reps in reserve on most sets. On the big compound lifts I agree, because they are there to build strength. On the accessory exercises, in the 8 to 15 rep range, I would train to failure.
I said it in my own profile and I stand by it: AI can answer questions correctly, but if you don't ask the right questions about problems you don't even know you have, it will never really help you.
The plan was never the problem
A successful man once came to me asking for a training plan. When I offered him one, he said no, he wanted the coaching I was talking about, because he understood something that changes everything: the plan was never the problem. He could have found a plan in five minutes. What he could not find in any app or any program was someone to help him work out why he keeps stopping, and how to become the man who never stops. That is what I do, and no AI in the world does that.
Plans are cheap. Coaching is the human work AI cannot do. What actually changes a client's body and life is the coach who watches their real week, sees what is about to break, and adjusts before it does.
AI only gives the first step towards a plan that works. To get results, keep them and really change the body, the training plan has to keep adapting, and there is no perfect plan you can buy once. Real personalization is done over time, not at one moment.
What I hand to AI
- A first plan, from good questions. Given the questions a coach would ask, it builds a solid first step for a coach to adjust.
- The question in the kitchen. A client with 600 calories and 40 grams of protein left for the day can tell it what is in the fridge and ask what to make.
- Logging by talking. I prefer talking to typing, and describing a meal by voice takes seconds.
That is how Evo SI works in Evobody. It answers a client's questions between sessions from their own plan and numbers. A fridge, a shelf or a restaurant menu: it suggests what fits, checked against that client's allergies and diet rules and what is left of their day. It reads photos of meals and logs food only once the client approves it. It works for the coach too, trained on the coach's philosophy and methods, set up in the app. SI stands for Super Intelligence, previously known as Artificial Intelligence.
What I still check by hand
Calories from a photo. AI is not there yet with a photo of a plate. The same potatoes can be very different cooked at home or in a restaurant, because some cooks use more fat and some use cream, and a photo cannot see it. In Evobody nothing from a photo is logged until the client approves it, and where a checked food matches, the checked numbers are used instead of an estimate.
Advice that only sounds personal. A lot of apps open with a huge questionnaire and then give generic advice anyway. That is automation, not personalization. I know the difference, because personalizing is my job.
An app that promises everything. If I told you an AI app would solve all your problems and get you fit for $10, I could probably sell it to a lot of people. That does not mean it works, or that it fits anyone's life.
What AI must never answer
Some questions are not for software, however good it gets: pain, an injury, anything medical, signs of disordered eating, a request to change the plan, billing, and a client in distress. Those need a person.
In Evobody, Evo SI never diagnoses, never gives medical advice and never changes the plan. When a client raises one of those things, it tells them it has passed it on, and their coach sees it on the dashboard with a note. Anything about pain, an injury or health beyond coaching should get a reply from the coach the same day, and where it is beyond the coach's scope, a clear suggestion to see the right professional.
The reason is simple. AI has not lived, and it cannot teach what it has not survived. And AI does not have a standard. A coach does.
The coach stays in charge
Coaching is a partnership, not a transaction. The work AI cannot do is the coach's: watching a client's week, adjusting before it breaks, motivating them when they need it, and reminding them what matters so they stay focused on the one goal.
Evo SI proposes rather than decides, and never overrules the coach. When I tell people about my own coaching app, I say it plainly: the AI coach is there to use or not. Every client gets ten free questions, and after that it is $3 a month for each client the coach switches it on for. Switched off for a client, it is hidden from their app and costs nothing.
Questions coaches ask about AI
Can AI write a training plan for my clients?
It can write a good first plan. When I tested ChatGPT with my own questions, it built a complete plan with calories, protein, a split, rep ranges and rules for adding weight, and chose the exercises I would choose. It still needed a coach's eye on the details, and it only ever gives the first step.
Will AI replace personal trainers?
Not the coaching. Plans are cheap: ChatGPT can write a good one in five minutes. The work AI cannot do is watching a client's real week, adjusting before it breaks, and helping them work out why they keep stopping.
What should AI never answer for a coaching client?
Pain, an injury, anything medical, signs of disordered eating, a request to change the plan, billing and distress. Those need a person. In Evobody, Evo SI passes them to the coach with a note, never diagnoses and never changes the plan.
Can AI count calories from a photo?
Not reliably yet. The same food can be very different depending on who cooked it and with how much fat or cream, and a photo cannot see that. In Evobody a client approves every photo entry before it is logged, and checked foods are used where they match.
How do I use AI with clients without losing the personal touch?
Let it handle the questions with an answer in the plan, like what to make from the fridge or how to log a meal, and keep the coaching for yourself: reading the week, adjusting the plan, and the conversations that need a person.