Can AI replace a personal trainer?

You can now open an app, answer six questions and receive a twelve week training programme in about four seconds. It will look professional. It will have sets, reps, progression and rest days. It will cost a fraction of what a coach charges. So the question is fair: is there any reason left to pay a human?

The honest answer is that AI has taken over roughly half of what a personal trainer does, and it happens to be the half that was never the hard part.

1. What a personal trainer actually does

Ask people what they pay a trainer for and most say the programme. Ask trainers where the value sits and you get a different list.

• Assessment. How you move, what hurts, what you can load safely today.
• Programming. The plan itself, which is the most automatable piece.
• Coaching in the moment. Changing a cue, dropping a set, spotting that today is not the day for a heavy single.
• Adjusting over time. Reading what happened over six weeks and deciding what changes.
• Accountability and relationship. Someone expects you, notices when you disappear, and asks about it.

Software has largely absorbed the second item. It is beginning to touch the third. It has barely started on the rest.

2. What AI genuinely does well

Structure. A generated plan will typically respect sensible basics: a mix of push, pull, legs and core, two to three strength sessions a week, progressive overload, some conditioning. That already puts it ahead of what most people were doing on their own.

Progression maths. Working out loads, percentages, deload weeks and volume across a block is arithmetic, and machines are better at arithmetic than tired humans in a gym.

Constraint solving. Three days, forty minutes, one kettlebell and a bad wrist. AI is genuinely good at rearranging a plan around limitations like that.

Availability. A question at eleven at night gets an answer at eleven at night. Small, but it removes friction that used to stop people entirely.

Explanation. Asking why a movement exists, what a tempo means, what tightness in the hips during a squat might indicate. Used as a fitness encyclopaedia it is excellent.

Tracking and logistics. Logging, reminders, trend charts, comparing this month to last. Nobody misses doing this by hand.

3. Where AI training falls short

It cannot see you. Everything it knows about your body comes from what you typed. If you underestimate your fitness, overstate your experience, or forget to mention the shoulder that has ached for two years, the plan is built on that.

It has no memory of your bad weeks. A coach who knows you slept badly all month programmes differently. Most apps will happily hand you your heaviest session of the block on the worst day of your year.

It assumes compliance. Plans are written for a person who completes every session. Real training histories are full of gaps. What matters is what happens after the gap, and that is exactly where generated plans go quiet.

It over prescribes. Ask for a plan and you often get five or six sessions a week, because more looks more thorough. Research on training frequency suggests that two or three well structured strength sessions per week already produce most of the adaptation available to a non athlete. The extra sessions mainly buy you a reason to quit.

It cannot triage pain. Distinguishing normal training discomfort from a problem is a clinical judgement. Software will suggest stretches for things that need assessment.

It does not know why you are here. Someone training after a bereavement, during recovery from disordered eating, or to feel less anxious needs a completely different approach from someone chasing a number. That context rarely fits in a form.

4. The form feedback question

Camera based coaching is the feature most often advertised and most often overstated. Markerless motion capture has improved substantially, and in controlled conditions it can estimate joint angles reasonably well. In a real gym it is working with one phone, poor lighting, a partial view and clothing that hides the joints it is trying to track.

Realistically, it can tell you whether you reached depth, whether your back changed shape noticeably, and whether your knees collapsed inwards on a clean side view. Useful things.

It cannot tell you whether you are bracing, whether you are breathing, whether you are shifting weight onto one leg to protect something, whether the movement feels wrong to you, or whether the load is simply too heavy today. That is most of coaching.

Treat automated form feedback as a mirror that occasionally comments. It is not supervision.

5. How to brief an AI for a better plan

The gap between a mediocre generated plan and a good one is almost entirely in the brief. Give it the following.

• Your training age. Not your goal, your history. Years of consistent training, and how long since the last consistent stretch.
• Days you will genuinely train. Then remove one. Plans that survive are slightly smaller than your ambition.
• Equipment, precisely. A home setup with two dumbbells produces a very different plan from a full gym.
• Injury history and current niggles. Including the ones you have stopped mentioning to people.
• What you enjoy. Adherence research is consistent on this point. Enjoyment is not a luxury, it is the mechanism.
• What you have quit before and why. The most useful single sentence you can give it.

Then ask for specific outputs: a four week block rather than twelve, explicit progression rules, a deload, a minimum version of each session for bad days, and one clear rule for what to do after a missed week.

6. The hybrid approach most people need

The interesting model is not human versus machine. It is a small amount of human input placed where it changes outcomes.

A session or two at the start to check how you move and set direction. Software to run the daily work. A human check in when something changes: a new pain, a plateau that lasts, a life event that reshapes your week, or a goal that shifts.

That is a handful of expert sessions a year rather than weekly coaching, and it keeps the parts that software genuinely cannot do.

7. Training, mood and the part nobody programs

Most people who start training say they want to change their body. A large share stay because of how it makes them feel. The evidence linking regular physical activity with lower rates of depressive symptoms is strong, and the effect does not depend on the plan being optimal.

This matters for the AI question, because a plan optimised purely for physical adaptation can quietly work against the reason you turned up. Sessions that leave you flat, a schedule that generates guilt, a tracker that makes movement feel like an audit. None of that shows up as a programming error and all of it shows up as quitting.

If exercise is part of how you manage stress or low mood, say so in the brief, and be willing to change a plan that is technically correct and emotionally exhausting.

8. When to book a human

Choose a person, not software, when any of these apply.

• You have pain that persists beyond normal soreness, or any pain that changes how you move.
• You are pregnant, postnatal, or returning after surgery or illness.
• You have a health condition that affects exercise, including cardiac, metabolic or joint conditions.
• You are entirely new to resistance training and want to learn technique properly.
• You have a history of disordered eating or compulsive exercise.
• You have followed sensible plans for months without progress.
• You keep starting and stopping, which is a coaching problem rather than a programming one.

So, can AI replace a personal trainer? It can replace the spreadsheet. It cannot yet replace the person who notices.

If you are not sure which side of that line you are on, you can describe your situation anonymously through Ask for help in the BYOU app and get a straight answer from a qualified coach. Your name and contact details are never shared with the experts who reply, and nothing is booked unless you decide to.