In endurance sport, we measure almost everything: power, pace, heart rate, sleep, calories, HRV, readiness scores, and the endless stream of metrics delivered by wearables. But despite all this sophistication, one of the most revealing indicators of an athlete’s internal state remains remarkably simple:
What they say about their training.
Recent work from researcher and sport-science engineer Andrea Zignoli brings new clarity to this idea. His study uses advanced transformer-based sentiment-analysis models (the same technology behind modern natural-language processing) to evaluate athletes’ free-text training comments. The outcome is a structured, data-driven look at how an athlete’s own words reflect internal load, emotional wellbeing, and fatigue.
As a coach, the implications feel both intuitive and overdue.
Athletes already tell us how they’re doing – we just haven’t been listening at scale.
Anyone coaching a roster of athletes knows this firsthand: reading training comments is time-consuming, but it’s also one of the most illuminating parts of the job. Numbers tell you what happened. Words tell you how it felt. And that “felt sense” has always been a more reliable predictor of breakdown than anything gleaned from a wearable.
One of Zignoli’s strongest findings is that sentiment in comments aligns closely with an athlete’s self-reported feel – their internal sense of fatigue or readiness. This matters because training is as emotional as it is physiological. When comments shift from descriptive (“smooth,” “controlled,” “strong”) to constrained (“heavy,” “flat,” “unmotivated”), something has changed – often before the metrics show it.
Language captures internal load that traditional metrics miss.
Power and heart rate can’t show frustration with work, creeping mental fatigue, or a fading sense of purpose. Yet athletes describe these states, sometimes directly, sometimes subtly, long before they show up as missed targets or skipped sessions. Shorter comments, sharper tone, or even silence can become their own signals.
Zignoli’s research suggests coaches should treat comments as structured data, not peripheral noise. When tracked over time, emotional shifts form trendlines: rising negativity may signal impending burnout; sustained positivity can indicate readiness for more load. Importantly, sentiment doesn’t replace objective metrics – it enriches them. When physiological and emotional stress climb together, risk increases. When they diverge, it’s time for a conversation.
The takeaway: listening is not optional, it’s part of evidence-based coaching.
You don’t need machine learning to benefit from this insight, but NLP tools make it scalable for larger teams and busy coaches. Even without automation, simply building the habit of reading between the lines can prevent overreaching, preserve motivation, and improve long-term consistency.
I’ve long believed that the best coaches are the best listeners. This research backs that belief with data. The athlete’s voice, expressed in their own words, remains one of our most powerful diagnostic tools. And now, more than ever, we have evidence that what they write may be just as important as what they swim, bike, or run.
