AI for personal trainers: the hours it saves, where it gets clients wrong, and how to use it without losing the personal touch

Use AI for the first draft of almost everything and the final word on nothing. It is fast at the typing in coaching (programs, meal plans, check in forms, replies), and the studies so far find it weakest at what clients pay you for: fitting the plan to the actual person, and being the actual person they answer to.

Below: what to hand over and what to keep, what experts found when they graded AI programs and diet advice, whether clients mind, how to protect their data, and a routine that saves hours without anyone feeling coached by a robot. Full disclosure: we build an AI assistant for coaches, so we have opinions, and we have put a source under every one.

Will AI replace personal trainers?

Not on the evidence so far, and the official forecast is not betting against you. The US Bureau of Labor Statistics projects employment of fitness trainers and instructors to grow 7 percent from 2025 to 2035, "much faster than the average for all occupations", with about 68,000 openings a year.

What AI is replacing is the typing. A generic 12 week program or a 2,000 calorie meal plan used to cost an evening; now it costs a sentence and a minute. If your whole offer is a PDF, that is a real problem, because your competition just got very cheap.

What it cannot replace is the reason clients stick. A 2011 paper by Mohr and colleagues proposes a model it calls supportive accountability: "human support increases adherence through accountability to a coach who is seen as trustworthy, benevolent, and having expertise". A chatbot can be polite. It cannot be someone your client would feel bad letting down, and it has no idea that your client's "all good" normally comes with three exclamation marks, so this week's zero is news.

Clients have instincts about this too. In a Pew Research Center survey of 11,004 US adults, run in December 2022, 60 percent said they would be uncomfortable if their own health care provider relied on AI to diagnose disease and recommend treatments, and 57 percent said it would make the relationship between patient and provider worse. That is medicine, not coaching, but people bring the same instinct to whoever is working on their body. AI replaces the typing, not the trainer.

How can personal trainers use AI?

Think of AI as a fast, tireless junior assistant who has read every fitness textbook and met none of your clients. Brilliant at structure and words. No idea that Sarah's knee hates lunges.

• Programs. AI: the first draft, meaning the split, exercise picks, sets, reps and progression between blocks. You: injuries, equipment, the client's real week, and whether they will actually do it.
• Meal plans. AI: hitting the targets, food swaps, the grocery list. You: allergies, medical conditions, culture and taste, and when to refer to a dietitian.
• Check in replies. AI: a summary of the answers and a first draft. You: the thing only you noticed, the one change, the tone.
• Check in forms. AI: a first set of questions. You: deleting every question you would never act on.
• Roster triage. AI: who went quiet, whose logging dropped, whose check in is late. You: why, and the conversation.
• Business admin. AI: onboarding messages, FAQ answers, content ideas. You: your voice, your prices, your promises.

The AI half of each line is structure and words. Your half is people and decisions. Most of the hours live in the first half and most of the value lives in the second, so good use of AI moves your time from one to the other.

How much time does AI actually save?

We could not find a study that measured it for coaches, so be suspicious of anyone quoting a confident number. The closest good evidence comes from writing work. In a 2023 experiment published in Science, 453 college educated professionals were given writing tasks from their own fields, and half of them got ChatGPT. That half took 40 percent less time, and their work was rated 18 percent better.

A lot of coaching admin is writing, so here is the arithmetic as an illustration, not a promise. Twenty clients, and a check in reply takes you about 12 minutes: that is four hours a week. If you got the same 40 percent back, you would save about an hour and a half every week without coaching anyone less.

Two catches. The saving only counts if you still read every word, so budget for the reading. And the best use of the time you win back is not twenty more clients on autopilot. It is the human part: a voice memo, a call, a reply that lands the same day. If you want to see what faster admin does to your ceiling, put your new minutes per client into the client capacity calculator.

Where does AI get coaching wrong?

Researchers have started testing this directly: ask ChatGPT for programs or diet advice, then have experts grade what comes back.

• Dergaa and colleagues, Biology of Sport, 2024. Asked for 30 day programs for five hypothetical people, each with a health condition and a goal. The experts found the programs safe but generic, saying ChatGPT "lacked precision in addressing individual health conditions and goals, often prioritizing excessive safety over the effectiveness of training".
• Washif and colleagues, Biology of Sport, 2024. Asked for 12 week resistance programs for a hypothetical man and woman, both 20 and intermediate. The programs were "objectively similar regardless of sex", useful as a draft but likely to need expert adjustment before use.
• Zaleski and colleagues, JMIR Medical Education, 2024. Asked for exercise recommendations for 26 clinical populations. They were 41.2 percent comprehensive and 90.7 percent accurate. The most common error was telling people to get medical clearance before any exercise at all, more caution than needed, which the authors warn can become a barrier to exercising.
• Ponzo and colleagues, Nutrients, 2024. Asked for diet advice for chronic diseases, from sarcopenia to fatty liver disease. It was appropriate in 55.5 to 73.3 percent of answers depending on the disease, with contradictory or inappropriate advice once conditions overlapped.

Every one of these tested a 2023 version of ChatGPT, and models have changed since. But look at the shape of the misses. They are not wild. They are average: safe, generic, a little too cautious, and the same for a man and a woman. It is the hotel breakfast of programming. Nothing is wrong with it, and nothing about it says it was made for you.

The second miss is quieter. A model only knows what it was told. The knee, the allergy, the night shifts, the client who would rather quit than do another burpee: all of it reaches the AI through you, and the draft has to come back through you too.

Five checks before anything AI wrote reaches a client

1. Injuries and pain. Read every movement against every note in the client's file, and name each swap so the client knows it was deliberate.
2. Allergies and medical conditions. Overlapping conditions are exactly where the nutrition study saw the advice go wrong. Anything medical (diabetes, pregnancy, an eating disorder history, a new pain that is not settling) goes to the right professional, not to a better prompt.
3. The arithmetic. Add up the meal plan yourself, or use a tool that computes it from real food data. A model can write "2,000 calories" at the top of a plan with exactly the same confidence whether the meals underneath add up to it or not.
4. The client's actual week. Sessions that fit their schedule, equipment they own, and exercises they can do and do not hate.
5. The voice. Would the client recognize it as you? If they have never heard you say "I hope this message finds you well", they should not start hearing it now.

Do clients mind if their coach uses AI?

Here is the uncomfortable study. In 2024, researchers publishing in PNAS found that AI written messages made recipients feel more heard than messages written by people, and that AI was better at detecting emotions. Then the twist: recipients "felt less heard when they realized that a message came from AI".

Both halves matter. AI can genuinely help you write a warmer, more careful reply. The moment a client suspects nobody actually read their check in, the warmth flips. So:

• Say it once, plainly, at onboarding. Something like: "I use an AI assistant to help draft programs and summarize your check ins. I read everything, I make every decision, and every message from me is written or edited by me." Then make that sentence true every week.
• Never send a draft you have not changed. Add one thing only you could know: the 6 AM session they squeezed in before a flight, the dumbbell that finally went up a size.
• Keep the big moments human. The first week, a bad week, an injury, a client who wants to stop. Those get a voice memo or a call, not a draft.

AI saves the most time on the parts clients notice least, so spend what it saves on the parts they notice most. Our guide to keeping coaching clients covers what those are and the warning signs to scan for each week.

Is it safe to put client data into an AI tool?

Treat anything you type into a consumer chatbot like a postcard: fine for the weather, wrong for someone's medical history. Google's own privacy page for its Gemini app says human reviewers review some of the data it collects, and puts it plainly: "Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services". That is a responsible thing for Google to say, and a good rule for you.

Four habits:

1. Strip identifiers. First name or initials at most. No contact details, no photos, no dates of birth.
2. Check the data settings before you paste anything in, not after.
3. Tell clients, and get their consent. One line in your intake about which tools you use and for what.
4. Ask vendors in writing whether your clients' data is used to train their models, and prefer the ones that say no.

For the record, here is our answer. When you use Bryksy, the relevant client data goes to our AI providers to produce your answer, and under their commercial API terms they do not train their models on it. Our privacy policy names each provider.

Who is responsible when the AI gets it wrong?

Assume you are. The client hired you, and the program went out under your name. "The chatbot wrote it" is not an excuse any client will accept, least of all one with a sore knee.

AI changes none of your obligations. The health intake still happens (we suggest starting with the PAR Q+, as covered in our check in template guide), anything you would have referred out still gets referred out, and a short note of what you changed in each draft and why takes thirty seconds. For the legal detail, ask your insurer and your certifying body, not a chatbot.

A routine that saves hours without losing the client

Brief, draft, check, add, send. Five steps, and the third one is not optional.

1. Brief it like a new assistant on their first day. The more context it gets, the less average the draft. For example:

   Draft a 12 week program for a 41 year old intermediate lifter, three days a week (Monday, Wednesday and Friday, 50 minutes each), in a home gym with dumbbells up to 25 kg, a bench and a pull up bar. Goal: stronger, and about 5 kg lighter. Left knee: no deep loaded knee bends, no jumping. Hates burpees. Last block: goblet squat, 20 kg for 3 sets of 10. Progress between 4 week blocks and name every swap you make for the knee.

2. Let it draft one thing at a time. The program, then the meal plan. Never "sort out everything for Sarah".

3. Run the five checks. Every time, including the hundredth.

4. Add the line only you could write. For check ins, try asking the AI for a three line summary (wins, problems, one question to ask) and writing the reply yourself. It can be faster than editing a full draft, and it sounds like you because it is you.

5. Send it under your name, then spend the time you saved on the people part.

Start with one job, not six. Check in summaries make a good first one: low risk, high volume, and two weeks is enough to know whether it is saving you anything.

What not to hand to AI at all

• The first message to a new client.
• Replies to bad news: an injury, a loss, a client who wants to stop.
• Anything about body image or an eating disorder history.
• Medical questions, which belong to a doctor, a physiotherapist or a dietitian.
• The celebration when someone hits a big goal. That one is yours.

How Bryksy does this inside BryksBuilt

Full disclosure, this is our product. Bryksy is the AI assistant built into the BryksBuilt coach portal and coach app (your conversations sync between the two), and we built it around the rules above: it drafts, you decide.

• It works from your roster, not the internet. Bryksy reads your clients' logged training, food, check ins and weight, and a roster briefing sorts who needs attention, who had a win, where check ins stand and who has gone quiet. It is built to cite: numbers about a client come with source chips that open the log they came from.
• It drafts, you approve. Programs, meal plans, check in forms, check in replies, messages and welcome packages. Nothing it drafts is saved or sent until you approve it, you can edit a message before it goes, and Bryksy never messages a client on its own.
• Programs start from your exercise library. Drafts prefer the exercises in your library, so clients get the demo videos. In the portal's program generator, picking a client fills in the injuries they shared in the app, and Bryksy is told to swap out anything that loads them.
• Meal plans are filtered before the AI sees them. In the portal's meal plan generator, picking a client ticks the allergies they gave the app (you can see and change them), recipes containing a ticked allergen are taken off the recipe list before the AI picks from it, and the macros on library meals come from the recipes themselves, not from the AI doing arithmetic.
• Macro targets start from the client's own numbers. Bryksy proposes targets from the maintenance BryksBuilt works out from the client's own answers, and approving a target 1,000 calories or more below maintenance, 600 or more above it, or under 1,200 a day makes you confirm first.
• Client words are treated as data. Bryksy is told to treat anything a client wrote as information to read, never as instructions to follow, and nothing it drafts reaches anyone without your approval anyway.

Bryksy is included on every paid coach plan. Trainer Starter is $22.99 USD a month for up to 25 clients with 250 Bryksy messages a month, Growth includes 500, and Unlimited has no monthly message cap. Each plan also has a monthly allowance of full drafts (programs, meal plans, recipes and check in forms: 20 on Starter, 50 on Growth, 150 on Unlimited) and a fair use daily limit. Trainer Free covers up to 5 clients, free forever, with check ins, programs, meal plans and messaging, but not Bryksy. If you are weighing platforms, our ranking of personal trainer software compares what five of them cost at 5, 25 and 50 clients.