AI can write a running programme in seconds. You can download a beginner's plan for free, get a schedule from your smartwatch or follow a marathon plan you found online. But how much guidance do you actually get when things don't go to plan?
Running has never been more accessible. Whether you're aiming to complete your first 5K or break three hours in a marathon, there's a training plan available to help you get there. You can download the NHS Couch to 5K app, ask ChatGPT to build you a marathon programme, or let your Garmin suggest what you should be doing tomorrow, and it's often convenient, free and a very good starting point.
But there's a distinction worth considering: having a training plan isn't necessarily the same as having a coach. A plan tells you what to do, while a coach helps you understand whether, when and how you should do it. As AI becomes an increasingly common part of how we train, that's a question worth asking at every level of running.
Starting out: is Couch to 5K enough?
The NHS Couch to 5K programme (opens in a new tab) has helped introduce countless people to running. Over nine weeks it takes you from short running intervals to 30 minutes of continuous running, with three runs a week and a rest day between each one. Every run starts with a five-minute warm-up walk and ends with a five-minute cool-down walk, and the app provides audio guidance along the way while encouraging you to go at your own pace. The NHS also advises speaking to a GP before you start if you're worried about your health.
It's a brilliant concept, because it removes much of the uncertainty around getting started and gives people a manageable target to work towards. But consider the people using it. One person might already walk several miles every day, another might have spent the last decade largely sedentary, and someone else might be returning to exercise after an injury or managing a condition that affects their ability to run. They're all beginners, but they're not starting from the same place.
The programme provides general guidance, but it can't assess each person's running history, observe their technique or tell whether they're recovering well between sessions, and it can't personally step in when someone struggles with a particular week. That's not a criticism of Couch to 5K. It's simply the limitation of a programme designed to be accessible to a huge number of people.
There is some research worth considering here. A 2023 study, "Couch-to-5k or Couch to Ouch to Couch!?" Who Takes Part in Beginner Runner Programmes in the UK and Is Non-Completion Linked to Musculoskeletal Injury? (opens in a new tab) (Relph et al.), followed 110 people taking part in a modified nine-week Couch to 5K programme. Participants reported 21 injuries, around 19% of the group, only 27.3% completed the programme, and those with a previous injury were more likely to report a new one. These figures relate to the people in this particular study, rather than the completion rate of the NHS app as a whole.
It's important not to overinterpret those findings. The study was relatively small, and the injuries were self-reported rather than diagnosed. The programme was also modified to add one instructor-led group session a week at a local sports facility, with the other two runs done independently through the NHS app. It doesn't show that the NHS Couch to 5K programme causes injuries.
But it does raise a useful question about how beginner programmes account for differences between participants, and perhaps the more important question isn't whether a beginner can follow a plan, but whether they know what to do when they can't. Should they repeat a week, shorten the running intervals or take another rest day? Is the discomfort they're feeling normal, or something to get checked? Some of those answers are in the programme's general guidance, but reading advice isn't the same as having someone assess your particular circumstances.
The AI coach in your pocket
This is where things get interesting. AI has made tailored training advice more accessible than ever. You can tell ChatGPT your current fitness, target race time, previous performances and weekly availability, ask it to create a complete training programme, then report back after each session with how you felt and your heart rate, and ask it to adjust the following week. In theory, it's a coach available whenever you need one, without the monthly fee.
I decided to put this to the test myself. I set myself the target of running a marathon in under three hours earlier this year (a failed attempt I'll try and do properly next year) and used AI to help structure my training, reporting how I felt before and after sessions, discussing my performances and tracking my progress. There were some obvious benefits. I could ask questions whenever they occurred to me, get explanations of different sessions and talk through how my training was going, and it was considerably cheaper than employing a personal running coach.
As I moved into my 9th or 10th week, I began to realise that having a training plan and knowing how to train aren't necessarily the same thing. During the summer, I had a five-mile tempo session scheduled on a day when temperatures reached 36°C. It was a miserable experience. I struggled through the session, and it affected my enthusiasm for running for a while afterwards. What's interesting is that my AI coach had actually suggested easing back because of the heat. The problem was that I was so focused on the session I'd been given, and the pace I was supposed to hit, that I found it difficult to let go of the target. I knew what the sensible advice was, but I still felt compelled to complete the workout as planned.
Could a human coach have made a difference? Possibly. They might have reinforced the importance of adapting sessions to the conditions, reminded me that missing a target wasn't failure, and challenged my thinking. Of course, a human coach can't guarantee that I'll listen either, but there's something different about having a person who knows you, understands your tendencies and can hold you accountable. The experience taught me that a training plan can be perfectly reasonable on paper and still be followed in an unreasonable way.
Can AI actually write a good training plan?
It's easy to dismiss AI-generated plans as generic schedules assembled from the internet, but the research suggests the picture is more complicated. In ChatGPT Generated Training Plans for Runners are not Rated Optimal by Coaching Experts, but Increase in Quality with Additional Input Information (opens in a new tab) (Düking et al., 2024), researchers asked ChatGPT to create six-week running plans using three different levels of information about the runner, and coaching experts rated each plan against 22 quality criteria.
The more information supplied, the better the plans were rated, but even the most detailed version wasn't rated optimal across all the criteria. The researchers advised against using ChatGPT-generated plans without feedback from an expert coach, rather than treating them as a replacement for professional guidance. That's an important distinction. AI can produce a programme that looks convincing, follows familiar training principles and includes a sensible mix of sessions, but even when a plan scores well against recognised training criteria, that doesn't necessarily mean it's appropriate for the person following it.
A small 2025 pilot study, Acceptance and trust in AI-generated exercise plans among recreational athletes and quality evaluation by experienced coaches (opens in a new tab) (Wachholz et al.), offered a more positive perspective. Recreational athletes were generally open to trusting an AI-generated plan, and the experienced coaches involved rated it favourably, although most of them could still tell which plan had been written by AI. It suggests AI is capable of producing credible programmes, but a pilot study of plan quality isn't the same as evidence of equivalent long-term performance, injury rates or coaching outcomes. The technology is improving. The question is whether the coaching experience is improving at the same rate.
What does a human coach bring?
I recently spoke with a performance coach and physiotherapist about training plans and the guidance people receive when following them. He described a pattern he sees regularly: people who haven't been given much education about running, or the demands it places on the body, taking on more than they're ready for. Some buy a new pair of trainers and head straight out with little idea of how to build up gradually. Others commit to a plan and feel they have to complete every session as written, even when their body is telling them otherwise. Many of the people who end up in his clinic with injuries got there not through a lack of effort, but through a lack of guidance.
It's an interesting perspective, because it highlights a problem that isn't unique to AI. Someone might download a marathon programme, follow their smartwatch's recommendations or ask an AI tool to write a schedule, and in each case end up following instructions without fully understanding how to adapt them to their own circumstances.
A human coach can offer something a standardised plan generally can't: individual assessment and ongoing judgement. They can ask how you're feeling, look at your recent training, consider your injury history and help decide whether a session should go ahead as planned. But there's another important difference: a coach can actually watch you run.
Running technique is something we don't always give enough attention to. A coach might spot an aspect of your posture, stride or movement that could be improved, then work with you on making small adjustments. That could help you run more efficiently, improve your running economy and potentially reduce unnecessary strain. Even something relatively subtle might make your easy miles feel a little easier, leaving you better placed to tackle the harder sessions. Of course, technique isn't a magic solution to injury prevention, and what works for one runner won't necessarily work for another. But having someone experienced observe you and offer feedback is something a training plan or smartwatch simply can't replicate.
They can also challenge you, sometimes by encouraging you to do more, and other times by telling you to back off even when you're convinced you should push harder. Runners can become very attached to their training schedules, treating missed sessions as failures and prescribed paces as targets that must be achieved. A good coach should help put those things into perspective.
There are limitations to human coaching too. Coaches vary in experience and quality, they aren't available around the clock, and employing one can be expensive. For many recreational runners, the cost alone makes a downloadable plan, an app or an AI alternative much more appealing, and that's a perfectly understandable decision.
Memory, data and the runner's catch-22
There are other questions worth considering before handing your training over to an AI tool. One of my frustrations was continuity. At times, information from earlier conversations or sessions months before wasn't remembered, and that matters, because good training advice should be based on more than the session you've just completed. Your recent mileage, previous injuries, how you respond to harder workouts, your recovery and your wider training history all contribute to the bigger picture. If you're repeatedly explaining the same information, or the tool can't see your training history, how personalised can its recommendations really be?
Dedicated coaching apps are a different proposition from a plan you've put together with a chatbot on your laptop, although they don't all work in the same way. Runna, for example, says its training plans are designed by coaches rather than created by AI (opens in a new tab). It works with watches including Garmin and Apple Watch, and uses an algorithm that can change your pace targets if you repeatedly miss or beat them. Its head coach has said that "no runner is the same, so we don't think people should follow a one-size-fits-all plan" (opens in a new tab), although that's the company explaining its approach rather than independent evidence of how it compares with a human coach. Apps like this can close some of the gaps I ran into.
But underneath, the same limits apply: an app only knows what it can measure or what you tell it, and it can't watch you run, examine a sore calf or notice that something about you seems off.
There's also more data than ever to make sense of. Your watch might give you a sleep score, heart rate variability, a training readiness figure and a recommended recovery time before you've even put your shoes on. Used well, that information can help. But it can also cloud your judgement, whether that's a low score making you feel worse than you actually do, or a good one encouraging you to run through something you shouldn't.
And in the end, the decision still rests with you, which is where the catch-22 comes in. We're told to listen to our bodies, yet some of what matters most is exactly what a runner can't always judge on their own. Is that niggle harmless or the start of something more serious? Is the fatigue a normal part of training, or a sign you're overdoing it? Is your technique the reason something keeps hurting? The apps can give you more information, but they can't always tell you what it means for you. That's where an experienced coach can help with training decisions and technique, while a physiotherapist can assess problems that need clinical attention.
Privacy and responsibility
Whichever tool you choose, it's worth understanding exactly what information it keeps and how it's used. Running data can be surprisingly personal, potentially including your weight, heart rate, injuries, medical information, location and daily routine. Some of this information, particularly information relating to your health, may qualify as special category data under UK GDPR and receive additional protection. That doesn't mean you can't share it with an AI service, but it's worth reading the provider's privacy policy to understand what happens to your information, how long it's kept and whether it's used for anything else.
There's also the question of responsibility. If an AI-generated programme contributes to an injury, who is responsible? What happens if a paid coaching service gives unsuitable advice, and does the provider have appropriate insurance? These questions don't have a single straightforward answer, because liability depends on the service provided, the circumstances and the contractual and insurance arrangements in place. Professional coaches may hold public liability or professional indemnity insurance, although the type and extent of cover varies, while an AI-generated plan doesn't automatically come with the same professional relationship or protections. It's another reason to understand what you're actually buying when you sign up to a coaching service.
Do you need a human coach?
I don't think the answer is necessarily yes. If you're relatively new to running, a well-designed beginner programme can be a fantastic way to get started, and if you're an experienced runner who understands training principles, an online marathon plan might give you all the structure you need. AI can be a useful tool for explaining sessions, organising training, discussing performance data and helping you understand the principles behind your programme. But I think we need to be realistic about its limitations.
My own experience has made me reconsider how I approach training for a 2027 marathon. Rather than relying entirely on an AI-generated programme, I'm considering a more collaborative approach.
I plan to use a performance lab to assess my running technique and identify areas where I might be able to improve. I'd then use AI to help structure my training, review sessions and monitor progress, while bringing in a human coach to assess how things are developing, probably through monthly reviews or whenever I feel additional guidance is needed. The coach could look at my training load, recovery and performance, as well as any feedback from the performance lab, and help decide whether the programme needs adjusting.
The idea is to combine the accessibility and convenience of AI with the experience and judgement of a human coach, supported by a more detailed understanding of how I actually run. It might offer a sensible middle ground between the two, particularly when the cost of regular coaching is a consideration. Not everyone can afford weekly sessions, but occasional professional input, alongside a proper assessment of your running, could help you make better decisions and get more from your training.
It's an approach I'm interested in testing, rather than something I'm suggesting everyone needs to do. The important thing is that the plan remains a guide, not a set of instructions to be followed regardless of what happens.
A plan is a starting point, not a guarantee
I wouldn't want anyone to read this and decide they need a personal coach before they can start running. That would be entirely the wrong message. Training plans have made running more accessible, and AI has the potential to make useful guidance available to even more people, which is something worth embracing.
But whether you're downloading Couch to 5K, following a Garmin training plan or asking ChatGPT to prepare you for your first marathon, it's worth remembering that a programme doesn't necessarily understand you as well as you might think. Learn when to adapt, understand the difference between discomfort and pain, and don't treat every pace target as a commandment. If something doesn't feel right, don't be afraid to stop, seek advice or change the plan.
A training plan can tell you what to do. Knowing when not to follow it might be the most important lesson of all.
How do you approach your running training? Do you follow a downloadable plan, use an app or AI, or work with a human coach? And do you feel you get enough guidance along the way?
.png?width=2560&quality=70&format=webp&resize=contain)


.jpg?width=2560&quality=70&format=webp&resize=contain)