Two months ago, I uploaded more than a decade’s worth of running data into the artificial intelligence chatbot Claude and asked for help writing a half-marathon training plan. “The good news,” it told me, “is your engine is enormous.”
This is the kind of flattery AI is known for – I’m not even in my run club’s fast group. Still, watching Claude scan the GPS files I had logged to the fitness app Strava and identify the highlights – a marathon personal best, a rugged ultradistance trail run – felt nice.
Then things took a turn. “The honest news,” it continued, “is where you’re starting from right now.” My mileage had fallen off a cliff. A standing Friday morning run with a friend was my only consistent workout. I was risking injury if I tried serious training too soon.
The reality check stung, but it also made me think the AI might be worth listening to. It then asked about my goals, my current fitness and workouts I’d enjoyed. Minutes later, I had a custom plan based loosely on the teachings of running coach Jack Daniels. Soon I was checking in with my AI coach after workouts and even adjusting my pace midrun to earn its approval.
AI has rapidly flooded the fitness world. One industry survey from last December found that two-thirds of gymgoers had used AI-powered fitness software in 2025. In 2024, Strava added an AI workout summary for subscribers. The fitness app has also acquired the automated coaching program Runna, which uses some AI to modify human-written plans. Last year, Peloton introduced an AI system that can count reps and give feedback on form using a built-in camera.
But many people are simply asking general-purpose AI models to coach them toward their next goal. While dedicated fitness apps tend to offer a narrower set of features, Claude or ChatGPT can be more flexible and will at least try to answer almost any training-related question.
Personal bests
For more experienced athletes, AI can act as an assistant to structure and refine their own ideas for training.
Daylen Yang, 30, a software engineer in San Francisco, is serious about fitness technology. His personal website displays his maximum heart rate from his last workout, his yearly cycling and running mileage, and how many hours of sleep he got the previous night.
He first used ChatGPT for coaching last year while preparing for a half Ironman triathlon, which consisted of a 1.2-mile (1.9 kilometres) swim, a 56-mile (90 kilometres) cycling time trial and a 13.1-mile (21 kilometres) run. He asked the chatbot how to improve his personal best by nearly 30 minutes, and it produced a viable plan immediately, he said. On race day, despite brutally hot conditions in the Utah desert, he met his goal.
The technology didn’t always work seamlessly. When Yang came back to ChatGPT to plan a fall marathon, the weekly mileage totals didn’t add up correctly. (AI often struggles with simple math.) Once he ironed that out, Yang said, the coaching proved its worth. Beyond the training plan, the model offered guidance on pacing, postrun aches and midrace nutrition.
It also kept his ego in check. When he floated an aggressive goal, the AI returned with something more attainable. Nine weeks later, he came within eight seconds of that target.
Some athletes are also turning to AI coaching for strength training. Victoria Boyd, a weightlifter in Las Vegas, went to ChatGPT for advice after a knee surgery had cut her maximum deadlift to 135 pounds (61kg) from 335 pounds (151kg). She was determined to build it back. Boyd, 44, had gotten traditional coaching in the past, and she thought the AI plan looked reasonable: It proposed well-structured workouts that became progressively more difficult.
She kept a running dialogue with ChatGPT to talk through how each session felt and to log her nutrition. She came to appreciate the validation and the tough love after hard workouts. “It’s like: ‘You’re not at your protein goal. You need to drink a shake right now,’” she said.
For Boyd, the numbers spoke for themselves – she is now as strong as she was before her surgery.
The human element
For some athletes who use AI to train, experience with human coaching has shaped how they relate to the technology.
Long before Chris Doenlen, 38, started working at Anthropic, the AI company that makes Claude, he earned his living as a strength coach and personal trainer. He now uses the AI model to help him train for long-distance cycling in the Cascade mountain range near his home in northern Washington.
The training plan Claude made was reasonable, Doenlen said, and similar to what a human might come up with after careful research. But he was also aware of what it was missing. Good trainers rely on context and nonverbal cues, he said. The AI, on the other hand, “is just going off what it has from you – it exists in a pure vacuum.”
For Jon Mott, a running coach from Lakeland, Florida, that human relationship is central to his work. But while training for a half-marathon last fall, he decided to see what he was up against. He employs six coaches who work with some 200 athletes, but he gave the job of his own training to ChatGPT.
Mott, 39, is a three-time qualifier for the US Olympic trials in the marathon, with a personal best of 2 hours, 17 minutes, or roughly 5 minutes, 15 seconds per mile. The AI put too much weight on that data point and prescribed workouts that he “could not touch,” he said.
The model then suggested much more prerace rest than he was used to. He felt sluggish from the gun and missed his goal by more than four minutes.
Even so, he said he doesn’t fully dismiss the technology because he sees how affordable, accessible basic coaching can have real value for many people.
“I want to hate on it 100%,” he said, “but I can’t.” – ©2026 The New York Times Company
This article originally appeared in The New York Times.
