The Promise vs the Reality
Every second fitness app now claims to use "AI" to personalise your training. The promise is compelling: a workout plan that adapts to you in real time, like having a personal trainer in your pocket.
The reality, in most cases, is a decision tree. You answer a quiz, it assigns you a pre-built programme, and that's the extent of the "personalisation." Static plans dressed up in AI language.
But actual AI-powered workout planning does exist — and when it's implemented properly, it works. Here's how to tell the difference.
What Real AI Workout Personalisation Looks Like
A genuine AI workout system does more than generate a plan. It updates the plan based on how you're actually performing.
The inputs that matter:
- Your recent training history (weights, reps, sets)
- How hard each set felt (RPE — Rate of Perceived Exertion)
- Your recovery status and training frequency
- Your stated goals and timeline
A real AI uses these inputs to make specific decisions: increase the load on your bench press because you've been hitting RPE 7 for three sessions in a row; reduce your squat volume this week because your RPE data suggests fatigue is accumulating.
This is what coaches do manually. A well-built AI system automates it.
The Evidence for Individualised Training
Sports science strongly supports individualised, auto-regulated training over fixed programmes:
- Athletes who train with RPE-based auto-regulation consistently show better strength gains than those following fixed percentage-based programmes, according to multiple peer-reviewed studies.
- Individual variation in recovery, training response, and fatigue tolerance is significant enough that a plan optimal for one person is suboptimal for another.
- Adherence to a programme is significantly higher when the plan adapts to the individual's lifestyle and performance — fixed plans that don't flex get abandoned faster.
The limitation isn't the concept. It's execution. Most apps don't have the data inputs or the algorithms to deliver on the promise.
What Bad "AI" Workout Apps Get Wrong
No feedback loop. The plan is generated once and never updates. There's no mechanism for the app to learn that the weights it suggested were too easy or too hard.
No RPE or effort tracking. Without knowing how hard your sets actually felt, the app has no signal to work with. It's flying blind.
Generic programming. The "personalisation" is limited to your fitness level and available equipment. Two users with identical profiles get identical plans, regardless of how differently they respond to training.
Overfit to beginners. Most AI workout apps are designed for people starting from zero, where any progressive plan works. They don't hold up for intermediate and advanced lifters who need genuine individualisation.
What Good AI Workout Apps Get Right
The best AI workout platforms share a few features:
Adaptive load progression — weights and volume update each session based on recent performance, not a fixed schedule.
RPE tracking per set — the system knows not just what you lifted, but how hard it felt. This is the key data input for real adaptation.
Long-term learning — the system improves its recommendations the longer you use it, because it has more data about how you specifically respond to training.
Goal-specific programming — distinct approaches for strength, hypertrophy, and endurance, not one-size-fits-all plans.
How Soma Does It
Soma is one of the few apps that takes RPE-based AI programming seriously. Here's how it works in practice:
You log every set with your weight, reps, and RPE. Soma uses this data to assess your performance relative to your targets — and adjusts your next session accordingly.
If you hit your sets at RPE 6 when you were targeting RPE 8, Soma will increase the load next session. If you're consistently above your target RPE, it may suggest a slight deload or extended rest.
Over time, Soma builds a picture of your individual response to training — your strength curve on each exercise, your recovery patterns, the volume you can handle in a week. The longer you use it, the more accurately it can programme for you specifically.
This is meaningfully different from an app that generates you a plan and leaves you to it.
So: Do AI Workout Plans Work?
The honest answer: It depends entirely on the implementation.
A quiz-based plan with no feedback loop? No better than downloading a free programme from Reddit.
An RPE-based adaptive system that learns from your actual performance data? Yes — genuinely. The evidence for individualised, auto-regulated training is solid, and apps that implement it properly can deliver real results.
The test is simple: after a few weeks of use, does your plan look different to someone else's? Does it update based on how your sessions actually went? If not, it's not really AI — it's a fancy template.
If you want to try a system that actually adapts, Soma is free to download on the App Store.