ChatGPT will build you a free training plan. Why that is not enough

What ChatGPT and Claude do well

• A one-off plan. Upload FIT files or describe your personal bests and receive a coherent 8–12-week plan. For a first 10K, that may be enough.
• Explanations. They can patiently explain why long runs should be easy and what threshold work means at your level.
• Lab-result interpretation. Upload a PDF and receive a plain-language explanation.

If you need one plan and already have the discipline to manage it, you genuinely may not need anything else.

Where the upload-files-to-a-chat approach breaks down

1. The data is outdated as soon as you upload it. You upload your archive on Monday. You sleep badly on Tuesday, your resting heart rate is five beats higher on Wednesday and your throat hurts on Thursday. The chat plan still describes Monday’s version of you. Training management depends on precisely these daily adjustments.
2. 300 FIT files do not equal analysis. Only a summary or part of the history will fit into a chat. An LLM will not recompute training-load trends across a thousand sessions; it reads what fits into context and generalises. The important details are in the trends: pace at the same heart rate over six weeks, acute-to-chronic load and your response to previous volume peaks.
3. No built-in safety constraints. A chat does not monitor week-to-week load progression. It might prescribe 30% more volume in one answer, then agree that you can still run when you report heavy legs in the next. No mechanism will overrule your enthusiasm and say stop.
4. No persistent memory or disciplined adjustments. Every chat starts from a blank slate. If you become ill, you must explain the context and rebuild the plan yourself. A live system notices that yesterday’s threshold session went poorly and proposes an adjustment: what changes, why and which numbers support it.
5. Lab results without practical consequences. A chat can interpret results, but that interpretation must become a constraint in the plan: an identified limitation remains a red line no session crosses, even when fitness improves and the race is three weeks away. In a chat, it remains buried in message history.
6. No reliable source trail. A chat response is the model’s opinion. In a specialised coaching system, a recommendation should reference a book, chapter or study you can open and verify.

Comparison: five ways to manage your training

Which option suits which athlete

• Your first plan, little data and a goal simply to finish — ChatGPT or Claude is free and may be sufficient.
• You own a watch and want minimal structure — use its app, accepting oddities such as 72-hour recovery after an easy run.
• Experienced, disciplined and well attuned to your body — book-based template plans have worked for decades.
• A major goal and the budget to match — a human coach is unmatched for technique, motivation and race-day observation.
• Plenty of data, no time for analysis and a specific dated goal — a specialised AI coach provides daily decisions based on your complete data, scientific evidence and firm constraints.

From our own experience

The entire build-up to the Moscow Marathon, including a 35:43 10K after a cycling-focused season, used this approach. The greatest value was not the initial plan but two interventions. The coach stopped a threshold series the watch had rated excellent because heart rate stayed flat while pace dropped, indicating accumulated fatigue rather than lost fitness. It also rebuilt the week within a minute after the athlete reported a sore throat. A chat with an uploaded archive cannot do either because those events happened after the upload.

We explain the underlying system, including data, the library of more than 200 books, safety constraints and adjustment discipline, on the How it works page. If you are choosing between the service and a human coach, see our dedicated comparison.

FAQ