Smash Coach is an AI coach, not a person
Smash Coach has no human coaches. The coach that writes your plan is AI, and nobody at Smash Coach watches you train. That matters here, because the clearest advantage the trials give personal trainers is a person standing next to you while you lift.
So this page keeps every finding, including the ones that favour trainers, says who was measured, and links each study at the foot of the page. None of the studies tested Smash Coach. Where a sentence is our judgement rather than a finding, it says our view.
The useful question is not which one is better. It is which problem you are trying to solve: learning to lift well, turning up, fitting training around a week that keeps moving, getting a straight answer, or dealing with pain. Each has a different best first call, and one of them is neither a trainer nor an app.
What does a trainer do that an app does not?
Three trials gave the same programme to people with and without someone supervising them. The person made the biggest difference on the lift that most needs coaching.
Squat strength: the clearest gap
In a 2025 German trial, trained men and women (79 analysed, about half women, average age about 31) did the same 10-week, three-days-a-week full-body programme in one of three ways: with a trainer at every session, through a coaching app with video demonstrations, progress tracking, reminders and delayed feedback on technique by video, or from a PDF. All three groups got stronger. The supervised group added about 27 kg to its squat, against about 19 kg in both other groups, a significant difference the authors put down to the technical guidance a complex lift needs. Bench press gains were close: about 9, 8 and 8 kg.
Note what the app was: the researchers' own programme on a coaching platform, with videos, reminders and a coach's delayed feedback on technique by video. It was not an AI coach, and nothing in the trial tested one.
A 2023 trial found the same pattern in 36 young men and women who had lifted regularly for at least a year. Eight weeks of one-to-one supervision gave more squat strength and more muscle growth at several sites than the same programme done alone. The size of the squat gap was uncertain, somewhere between about 9 kg and nothing; bench press was similar; isometric knee strength slightly favoured training alone, with a wide margin of error; and the countermovement jump modestly favoured the people who trained alone, a result the authors could not explain.
Heavier loads, sooner
The oldest of the three, from 2000, gave 20 moderately trained men the same 12-week heavy programme with or without a personal trainer. The supervised men raised their training loads faster and finished with a bigger squat and bench press. The abstract reports groups of only 10 and 8 men, so treat it as a small study pointing the same way.
Turning up
Supervision also kept people training. In the German trial, 15 of 17 supervised lifters completed at least 85% of their sessions, against 39 of 48 with the app and 25 of 47 with the PDF. In the 2023 trial, 2 of 22 supervised lifters dropped out against 7 of 23 working alone, and several supervised lifters said the appointment held them accountable. The app kept almost as many people on track as the trainer did, and far more than the PDF.
The caveats
- The German groups were unequal in size, the supervised group was younger, and anyone under 85% attendance was left out of the results, which the authors say may tilt them towards motivated people.
- In the 2023 trial, one author advises a fitness equipment maker and another consults for organisations that employ supervised strength training. The paper declares both.
- In adults over 60, a 2024 review of 34 studies from 30 trials and 2,830 people found no serious adverse events reported with either supervised or unsupervised exercise, and the same attendance, about 81%. Supervision gave small extra gains, and only knee extension strength held up when the authors tested how robust the results were.
What does an AI coach do well?
Answering questions, at any hour
A 2026 Belgian study asked nine practising personal trainers, each qualified to at least European level 4, for the question their clients ask most and their own answer to it. The researchers put the same questions to ChatGPT (version 3.5) on its default settings, once each. Eighteen trainers and nine topic experts then rated the answers without knowing who wrote them, for scientific correctness, how easy it was to understand and how usable it was. ChatGPT scored higher overall on six of the nine questions and lower on none.
The authors call the result exploratory: nine questions, answered in writing, by a version they expect newer ones to beat. It shows a chatbot can explain training clearly. It does not show that one can coach a session.
Writing plans that look right
Plans are harder than answers. Ten coaching experts rated six-week running plans that ChatGPT (version 3.0.1, used on 23 May 2023) wrote for a fictional runner. Quality rose as the chatbot was told more about the runner: the overall plan scored a median of 2, 3 and 4 out of 5, from the barest request to the most detailed. Even the most detailed plan was rated only neutral on health screening, testing and how often to train, and below neutral on how it built volume, and the raters noticed it raised training too quickly. The authors advise against using such plans without an expert coach's feedback.
Strength plans fared much the same. In one of two 2025 studies from the same research group, ten experienced coaches rated muscle and strength plans for a fictional advanced lifter from three chatbot set-ups, including GPT-3.5 through ChatGPT, in April 2024, and judged them moderate overall. Another group compared 12-week programmes from GPT-3.5 and GPT-4.0, written for hypothetical 20-year-olds, with a strength and conditioning textbook. The structure followed recognised principles, three four-week phases each with its own aim, but the authors judged the programmes drafts that likely need fine-tuning by a person.
A chatbot produces plausible answers and plans quickly, and AI coaching is usually there around the clock. Every plan study on this page ended with the same advice: have someone who knows training check the plan against the real person.
None of these studies put an AI coaching app against a personal trainer over weeks of training and measured what happened. The trials compared people with a printed or app-delivered plan, and the AI studies rated answers and plans on paper. Everything beyond that is inference, ours included. For what a general chatbot leaves to you once your week starts moving, see how Smash Coach compares with ChatGPT.
Can you have both?
Often, in our view. A 2025 review of 35 studies of coaching delivered through apps and websites to adults, on everyday habits such as physical activity, sleep and stress, found three kinds: human coaches in 18 studies, AI coaches in 13 and hybrids of the two in 4. Athletes and sports coaching were outside its scope. All three kinds were feasible and acceptable to the people using them. The hybrids had not yet delivered on their promise, and participants in two of the hybrid studies said they wanted some human contact alongside the AI.
The authors of the German trial reached a similar place: app-based training combined with periodic in-person support, with the amount of supervision matched to how complex the exercise is.
One way to split the work, in our view:
- Book a trainer for complex lifts such as the squat, especially ones you are new to, and again when you move to a heavier phase. The squat is where supervision showed its clearest gain, in people who already lifted; no trial here tested beginners.
- Let one planner hold the week: your matches, classes and sessions together. If that is an AI coach, tell your trainer what is in it. If it is your trainer, give them every fixture.
- Ask the AI between sessions, and take its plan to someone who knows training if anything in it looks like too much, too soon.
When is a physio the right call?
When something hurts and does not settle. A personal trainer is not a clinician, and neither is an AI coach. Both can work around what you tell them; neither can tell you what is wrong.
- Pain that lingers. The NHS suggests seeing a GP when joint pain has not improved after two weeks of looking after it at home, is getting worse or keeps coming back, or is stopping your normal activities or your sleep. In many areas you can refer yourself to an NHS physiotherapy service without seeing a GP first.
- Choosing a physio privately. In the UK, the NHS advises checking that a physiotherapist is chartered (MCSP after their name) and registered with the Health and Care Professions Council.
- Chest pain or fainting. Stop training. The NHS says to call 999 for sudden chest pain or discomfort that does not go away, and for anyone who faints while exercising. Outside the UK, call your local emergency number.
Chest pain or fainting during exercise is not something to train around, whether the plan comes from a trainer, an app or you. Stop and get help.
What should you choose?
Start with what goes wrong for you, not with the tool. Most people will find themselves in more than one row.
| What goes wrong for you | Who helps first | What the evidence says |
|---|---|---|
| Form on a lift you are learning | A personal trainer | Supervised lifters gained more squat strength than app-guided or PDF-guided ones (Gavanda 2025, 79 trained adults) and than lifters working alone (Coleman 2023, 36 trained young adults). No AI study here watched anyone lift. Both trials studied people who already lifted; none tested beginners. |
| Showing up week after week | A trainer, if you need someone expecting you | 15 of 17 supervised lifters completed at least 85% of sessions, against 39 of 48 with a coaching app and 25 of 47 with a PDF (Gavanda 2025). In adults over 60, attendance was about 81% with or without supervision (Gómez-Redondo 2024). |
| A week that moves around matches and classes | An AI coach that holds the whole week (our view) | Coaches rated chatbot plans better the more the chatbot was told about the runner (Düking 2024). Once you have told it your fixtures, a planner keeps them in every plan it writes; a trainer you see once a week plans from what you tell them in the session. |
| A quick question between sessions | An AI coach or chatbot | Nine personal trainers each gave their most-asked question and their answer; ChatGPT's answer (version 3.5) was rated higher on six of the nine and lower on none (D'hoe 2026). |
| Pain that does not settle | A physio or a GP | The NHS suggests a GP when joint pain has not improved after two weeks of home care, is getting worse or keeps coming back; many areas let you refer yourself to physiotherapy. |
| Wanting a person in your corner | A personal trainer | Satisfaction with supervision was higher with a trainer than with the app (Gavanda 2025), and several supervised lifters said the appointment held them accountable (Coleman 2023). |
Who a personal trainer suits: anyone learning lifts, anyone who wants a person watching every rep and adding weight when they are ready, and anyone who trains best when somebody is expecting them.
Where Smash Coach fits, in our view
Our view is that Smash Coach is the right choice for an athlete who does more than one sport, because it plans one week around every sport and class you do, with Pro or during your trial. It is not a substitute for a person watching you lift, and it is not a clinician.
Coach chat, with Pro or during your trial. The gold circle opens a conversation with the coach that wrote your plan, and it already knows your week: your sessions, your sports, your goals and anything you have reported. A change it proposes arrives on a card, and nothing reaches your plan until you tap Confirm. More in What you can ask your coach.
A plan that changes when life does. With Pro or during your trial, your plan is rewritten when something real changes, such as an injury, a sport or class you add, or a race, and never on a timer. Moving a session does not reshuffle the week. See When your plan changes.
Workout mode walks you through a session one step at a time, counts your sets and your rest, and runs a move done on each side as a left step and then a right step. See Train with workout mode.
Clear illustrations. Exercises are shown as clean illustrations: the movement on a plain background, with nothing extra, so you see what to do at a glance. The team is exploring video and moving images. An illustration shows the movement; it does not watch yours. See exercise illustrations.
Injuries. Report an injury or a fatigue flag, from Cleared to Serious, and with Pro or during your trial your coach rebuilds your plan around what you report. It is a training tool, not a medical service: it will not diagnose anything or suggest treatment, and it tells you to see a physio or a doctor. See Report an injury.
No human coaches. Your coach is AI. Feedback you send from You reaches a person rather than a queue, and a person looks at a session or change that fails the automatic safety check. See Send feedback.
What it lacks, plainly: no human coaches and nobody watching you lift; Garmin is the only watch it works with; HYROX is the race it prepares you for today, with ATHX, DEKA, GYMRACE and running races from 5K to the marathon coming soon; it installs from the website rather than the app stores; and it shows illustrations, not videos yet. If you would rather compare apps than people, see Smash Coach and Edge, Smash Coach and ChatGPT, or everything in Compare.