AI Training Plans and Advice - Watch Out for these Pitfalls

Jun 8
6 min read

AI is everywhere. It is becoming increasingly common for athletes to go to the likes of different Ai platforms for advice on all sorts of topics related to their training and racing. While other training apps are creating ‘personalised’ training plans solely based on AI models. This is not necessarily a bad thing as it has lowered the barrier for many people seeking information and training structure.

However, it has also raises some important questions and potential pitfalls to be aware of.

Through the consulting service I offer for independent athletes not wanting a full time coach but still wanting advice and guidance for their training, I have seen a surge in athletes arriving with AI generated training plans and wanting my advice on them.

Here are some key things to watch out for.

Poor inputs create poor outputs.

One of the interesting paradoxes with information and training plans from AI is that to write a good quality prompt to get a good quality answer you need a certain level of knowledge about training in the first place.

At that point, if you already have a strong understanding of training principles, you could probably write a decent programme yourself.

In addition to this if you are using an AI powered training app to prescribe your training it is extremally important that the data that is feeding back into this model is high quality. For example, if you are relying on heart rate data from a wrist based heart rate monitor which are notoriously inaccurate depending on strap tightness, movement, sweat or muscle tension, then the AI generate plan based on your data is going to under or over cook you.

Another example of this is your training zones. Getting these set correctly is really important for an AI training model but it is very common to find athletes using the default training zones in these apps or on their training devices. The same goes for good quality sleep data and HRV which while many devices calm to measure accurately they themselves rely on algorithms and models in their data outputs.

Again, poor inputs create poor outputs.

How do you know if the plan is actually good?

This is probably the bigger issue.

Once AI gives you a training plan or recommendation, how do you know if it is actually suited to your needs, situation or if the advice is actually any good?

AI models are very good at sounding confident. That however does not mean they are correct.

The ‘black box’ nature of AI means that we often do not understand how it came up with its answer. There are many cases of AI hallucinations in which it makes up information, creates falsehoods or misinterprets words out of context. In many cases AI developers do not fully understand why AI responds in certain ways.

Asking the same question on different AI platforms, models or even when using a free vs. premium edition results in significantly different information being produced.

There is also a bigger long-term question worth thinking about. As more AI generated content floods the internet, AI systems will increasingly train on and learn from information that was produced by AI. Over time, will this create a feedback loop where certain ideas become reinforced simply because they are repeated frequently online or preferred by the AI models or people reading them liked the response which reinforces the AI model?

So always make sure you ask for references and check the sources to ensure you are getting good information.

Oh, and don’t forget to think for yourself.

When it comes to AI created training plans I commonly see

Unrealistic loads and load progressions. This is potentially from an algorithmic bias or availability to research and or articles about elite athletes informing the AI answers? You need to be very specific about the details of you (the athlete).
I have also seen some very ‘interesting’ exercise selections in strength plans that do not match the individuals training experience or goals which can significantly increase the risk of injury.
There also seems to be a bias towards high intensity sessions with AI plans without much attention put into the aerobic base or how these higher intensity sessions are ramped up.

When planning for mainstream events that have significant research behind them and lot of generic training plans on line such as marathon running, cycling and triathlon the plans don’t stack up too badly.

However, I have noticed the biggest gaps are when people are training for or seeking information for less mainstream events where there has not been specific research or content published about them such as multisport, adventure racing, ultra endurance cycling events and ultra running. AI seems to be good at providing general ideas and training principles but seems to struggle to bring these into a cohesive real world plan.

In training, nuance matters, details matter and most importantly context matters.

Again, this comes back to knowledge and experience. To judge whether an AI generated plan is good or how to then shape this information into your real world situation, you still need a certain level of knowledge and experience.

The plan is only one part of the puzzle

Interestingly, most AI plans I have reviewed are often “good enough” for introductory level athletes. At the end of the day most types of training work for the first 4-6 weeks, especially for beginners who have large capacity for adaptation.

The irony is that beginner athletes are usually the ones who need the most guidance not necessarily in the shape of a well formatted programme but in terms of support and accountability because they do not yet know the general training rhythm, how to interpret the fatigue signals from their bodies, how to progress or adapt training sessions based on the curve balls that get thrown at you in day to day life.

AI can make a nice looking spreadsheet (and I do love a good spreadsheet) but it is important to remember that life is not a spreadsheet.

Life is messy and complicated and at the end of the day the training process is about navigating this rocky road to your goal not just numbers on a spreadsheet.

I have never been part of an athletes build up at any level that has gone 100% according to the original plan. There are always adjustments to be made, changes in day to day life, illness, injury, sick kids, meetings, and unforeseen obstacles always pop up, we will call it LIFE.

A training programme by itself is not what creates performance improvement. Performance improvements come in the doing of the programme and how this is implemented in the real world.

Good coaching is not just about writing sessions. It is about understanding when to push, when to back off, when hidden life stress is starting to pile up or how to use creative training approaches to unique situations.

AI can analyse data, identify trends and make pretty spreadsheets but sport is still deeply human.

Sometimes the best training adjustment is not changing a set of intervals or the latest and greatest sauna protocol. It is helping an athlete navigate motivation, confidence or lifestyle pressures. These are areas where lived experience, communication, and relationships still matter far more than algorithms.

This is the Art of Coaching.

Final thoughts

AI is here to stay, and I believe it will become an increasingly useful tool for coaches and athletes alike.

The key is understanding that a training plan is only as good as the context, interpretation and decision making behind it. The real challenge is not generating a programme, it is knowing whether it is appropriate for you, how to adapt it when life gets in the way and how to turn it into consistent progress over weeks, months and years.

If you enjoy using AI to help guide your training, that's great, use it. Ask questions, explore ideas, crunch challenge your thinking. But don't forget to think for yourself.

If you have an AI generated training plan, a plan from an app or something you have created yourself and want a second option or sounding board to ounce ideas off I can help. We can dig into it and identify potential issues, answer your questions and shape it into something that better suits your goals, experience and real world circumstances.

AI can provide information just like research papers, text books, chats with training partners, articles and Google searches have always done.