AI Chatbots in Fitness — New Study: Trust Follows Relevance

A new AI chatbot fitness study has measured why some people use AI for health and training — and others don't. The core finding: Trust in AI is shaped by attitude, not technology. Those who have experienced AI as useful once trust it. Those who haven't, distrust it. That sounds logical — it is. More interesting is what the study doesn't answer: whether that trust is actually warranted.

What the Study Measured — and What It Didn't

180 participants, various age groups, survey on usage frequency, trust, and satisfaction with AI chatbots for health and fitness. A solid base for a pilot study — but with a catch that's directly relevant for runners, cyclists, and triathletes.

Participants used AI primarily for mental health, symptom checks, and general well-being — not for training planning, periodization, or performance diagnostics. A different world from what most athletes here use AI for. Also: the study measures attitudes and intentions, not whether the AI recommendations were actually correct. Trust and quality were measured separately — that matters in a moment.

For context: the study shows patterns, not laws. The fact that younger women use AI more often doesn't mean they trust it uncritically. The fact that older users are more skeptical doesn't mean their concerns are unfounded. Survey data is always self-reported. Take the numbers as a guide.

The Core Finding: Trust Follows Attitude

Trust in AI and the intention to use it depend on whether you've experienced it as useful. Technology doesn't convince through technical specs — it convinces through a concrete experience that lands. Those who had it trust it. Those who got a generic answer move on.

Most bad AI experiences in training don't happen because the AI is bad — they happen because the question was too open. "Build me a training plan" gets a generic answer — no context, no relevance. "I'm a cyclist, 38 years old, 10 hours training per week, FTP has stagnated for three weeks despite consistent load — what would you check first?" gets something else. Same pattern for runners: instead of "how often should I run?", try "I run three times a week, half-marathon in eight weeks, my pace on long runs has plateaued — which variables would you look at?"

Anyone who's ever tried a new training app knows this: the first few weeks are skepticism, then one recommendation pays off — and you check in more often. Same with AI, except you have to create that first useful experience yourself. How to do that in practice: Create an AI training plan yourself.

Should You Trust AI for Training? Yes and No.

Trust is built through positive experience — that's what the study says. Whether that trust is warranted, it doesn't say.

Language models generate plausible-sounding answers. Not correct — plausible-sounding. On conceptual questions, AI is genuinely strong: "Explain the difference between polarized training and threshold methodology" gets a good answer — because that's knowledge found in hundreds of studies. Basic principles, physiology, method comparisons — AI makes few mistakes here, which is exactly what my AI Fitness Benchmark confirms: strong on the fundamentals, weaker on the nuance.

With personal recommendations, that flips. The more AI doesn't know about you — current load, injury history, sleep, stress outside training — the more it has to guess. And it does that convincingly. A 10% volume rule sounds safe. But it doesn't fit when you're coming back from a break while juggling too much at once.

What LLMs really know about training — and where they stop knowing — is covered in the article AI Training Data: What GPT, Claude, and Gemini Really Know.

What I do myself: when AI gives a specific recommendation, I ask for the reasoning. If a clear explanation comes back, good signal. If it's a platitude, a second source helps more.

Privacy — Legitimate Concerns, Concrete Steps

The study cites privacy and lack of empathy as barriers, especially for older users. That's not a technophobic reflex. Free AI services like ChatGPT or Gemini use inputs as training data by default — questions about symptoms, training volume, or body awareness can flow into model updates. This option can be turned off, but it's rarely mentioned.

For those who want to avoid this entirely: local models. LLMs that run completely on your own device, without the cloud. Ollama makes this possible without any developer background — free, Mac/Windows/Linux. With a model like Llama 3 or Mistral, you can ask questions about periodization, nutrition, and recovery without anything leaving your device. A normal laptop is enough to start.

What this means in practice and which EU regulations apply to AI tools in fitness is covered in the GDPR Guide for Fitness Apps and AI.

Why Women 18–27 Lead — and a Hypothesis

The most frequent users according to the study: women between 18 and 27, mainly for mental health, symptom checks, and well-being. The study doesn't explain why — I have a hypothesis.

AI as a conversation partner for topics that come with barriers in a normal fitness context. Cycle-based training, body weight, mental resilience — topics many don't want to discuss with a trainer, especially if the trainer is male. AI doesn't judge, has no social expectations, is available anytime. What's an advantage for factual questions clearly falls short for real emotional needs.

That's my hypothesis, not a study finding. If you know research on this, feel free to reach out.

Takeaway

What do you take away as an athlete when using AI for your training? Trust is part of it — but with a healthy dose of skepticism and your own evaluation of the answers. Rejecting AI across the board is just as wrong as trusting it blindly.

It makes mistakes. Even with a perfect prompt. The reason is structural: it can only reproduce what's in the training data — and that comes from the internet. Opinions, misconceptions, textbooks, studies. Not everything is correct. Bad research ends up in the data just as much as good research. Misinterpreted studies too. This isn't a problem AI created — it's the old problem of how knowledge has always been passed on.

That's why your own judgment is critical. Those who know the fundamentals of training — periodization, load management, recovery — can evaluate AI answers. Those who don't have to trust blindly. This applies to AI just as much as to trainers, supplement marketing, or any area where you receive recommendations. Foundational knowledge protects you — not from AI specifically, but from bad recommendations in general.

Sources

1. "Exploring User Trust and Adoption of AI Chatbots for Health and Fitness Management Across Age Groups." PubMed PMID: 42174920 (2026).

2. Davis, F.D. (1989): "Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology." MIS Quarterly, 13(3), 319–340.