You want to run a marathon. You’re a casual runner, but you’re ready to get serious. Where do you start?
Creating a training plan can feel overwhelming, so you open ChatGPT, type in a prompt, and suddenly have a 16-week marathon plan.
Easy, right?
Maybe it’s not that simple.
The Research
What does the research say? Honestly, not much.
Large language models (LLMs), like ChatGPT, are relatively new. So far, I have not found any peer-reviewed randomized controlled trials showing that AI-generated running or triathlon plans improve race performance, reduce injuries, or improve long-term adherence. The available research consists mostly of descriptive studies, expert evaluations, and surveys.
A 2026 study by Montaruli et al. examined marathon plans generated by several LLMs and compared them with established marathon-training principles. Most plans included mileage progression, tapering, and a high proportion of low-intensity running.
However, the researchers also found important inconsistencies. Some plans omitted weekly mileage totals, failed to clearly differentiate between intermediate and advanced runners, or provided inconsistent pacing recommendations, particularly for advanced athletes. Montaruli et al., 2026
In a 2024 study, experienced coaches evaluated ChatGPT-generated six-week running plans created from prompts containing different levels of detail. More detailed prompts generally produced better plans, but even the most detailed plan was not rated optimal.
The coaches identified problems with health screening, testing procedures, monitoring contextual factors, progression of training frequency, and progression of training volume. The authors advised against using ChatGPT-generated training plans without expert review. Düking et al., 2024
It seems AI can mimic the structure of a training plan. But can it keep up with the messy reality of training?
What AI Can’t Do
An LLM can produce a response to a prompt. What it does not reliably do is gather all the information a coach needs before prescribing and adjusting training.
If a plan increases volume too quickly, a coach might ask about fatigue, recovery, nutrition, injury history, schedule, and life stress before making a change. In the Düking et al. study, ChatGPT did not ask the kinds of follow-up questions a coach would typically use to refine a plan. Düking et al., 2024
AI relies heavily on the user to provide the right information. Not every athlete—especially a novice—knows which details matter.
There’s another issue: we generally do not know exactly how commercial LLMs weigh the information behind their responses. Their training data may include academic research, magazine articles, blogs, and other sources. As a result, a recommendation may sound convincing without being based on the strongest available evidence.
Finally, a basic AI-generated plan is usually static. Unless you return to the tool with updated information, it does not automatically know about missed sessions, soreness, fatigue, illness, poor sleep, or changing life demands.
Some training apps include adaptive features, but a one-time prompt or downloaded plan does not automatically interpret your wearable data or adjust your training in real time.
That distinction matters.
What Does a Coach Do That AI Can’t?
The true value of an endurance coach is individualization beyond the template.
Our intake process is tailored to the athlete’s history. We identify the information that matters before building the plan instead of relying on a generic prompt.
We design progressions and regressions based on how the athlete responds—not simply on average patterns. We consider wearable data, when appropriate, alongside fatigue, life stress, work schedules, family obligations, and recovery.
That allows us to prioritize the most important workouts and make the plan fit the athlete’s actual life.
Coaches can also help identify early warning signs of injury or overtraining. If an athlete is injured, a coach can work with the athlete and relevant healthcare professionals to develop a graded return-to-running plan that is flexible and individualized.
And coaching provides something AI cannot: accountability, context, encouragement, and judgment.
AI can generate a plan.
A coach helps you understand when to follow it, when to modify it, and when to stop.
How Can I Help?
If you’re tempted to try an AI-generated plan, I get it. It’s convenient, inexpensive, and available instantly.
But endurance sport is not easy, and training consistently takes time. Why cut corners with the part of the process that determines how you spend that time?
Until stronger evidence exists, athletes using AI-generated plans are effectively testing an evolving tool against their own bodies and race goals. Current research has not shown that these plans produce better outcomes than personalized coaching—or that they are equally effective and safe for every athlete.
Before you type that AI prompt, ask yourself what you want from your training and whether an AI-generated plan can truly help you achieve your best outcome.
If you want a plan built around your goals, schedule, training history, and real life, send me a DM or book a free coaching call to see whether personalized coaching is right for you.
Sources
- Montaruli G, et al. (2026). “Artificial intelligence-generated marathon training programs: reliable tools in exercise prescription for athletic performance?” British Medical Bulletin, 157(1), ldag010. DOI | PubMed
- Düking P, et al. (2024). “ChatGPT Generated Training Plans for Runners are not Rated Optimal by Coaching Experts, but Increase in Quality with Additional Input Information.” Journal of Sports Science & Medicine, 23, 56–72. Full text | DOI
