As conversational artificial intelligence becomes a routine part of daily life, many people are turning to chatbots for fitness advice, encouragement, and workout reminders. However, while these tools excel at sparking enthusiasm, whether they actually inspire sustained, real-world exercise remains an open question. In a systematic review published in Frontiers in Digital Health, researchers evaluated how conversational agents powered by large language models (LLMs) influence physical activity habits, finding a distinct gap between positive user impressions and measurable health outcomes.
Engaging Coaches, Missing Metrics
The research team, led by Alessandro Silacci, examined findings across 13 published studies to understand how contemporary conversational systems affect people trying to stay active. Unlike simple scripted programs, modern LLMs can generate nuanced, context-aware dialogue that makes fitness advice feel tailored and interactive. Across the reviewed literature, users consistently reported high levels of engagement, positive emotional experiences, and strong boosts in initial motivation when chatting with an AI assistant.
Yet the researchers found that these positive feelings do not automatically translate into lasting lifestyle changes. Solid empirical evidence confirming a direct, sustained increase in objectively measured physical activity—such as verified step counts or logged exercise sessions over weeks or months—remains sparse. While the chatbots make users feel supported in the short term, their ability to drive measurable, long-term behavioral change has yet to be rigorously proven.
The Illusion of Human Connection
A key reason LLMs capture user interest is their versatility. Older, rule-based chatbots often felt rigid and transactional, but today’s models can fluidly adopt different social personas. Depending on how they are prompted, they can act as gentle, encouraging peers or as authoritative, structured coaches, dynamically altering the relational dynamic with the user.
Because these conversations feel remarkably natural, users frequently anthropomorphize the agents, attributing genuine empathy, personality, and intention to lines of code. The authors note that while this perceived human-likeness deepens emotional investment and strengthens the user’s bond with the system, it also introduces psychological risks. People may form an emotional over-reliance on artificial software or develop misplaced trust, assuming the chatbot possesses a level of medical or physiological understanding that it simply does not have.
Persuasion and Ethical Gray Areas
These dynamic conversational capabilities also raise critical ethical questions regarding influence. When an AI agent functions as a form of persuasive technology, the line between helpful encouragement and emotional manipulation can easily blur. The review highlights concerns about how personal agency is redistributed among the individual, the technology itself, and the commercial third parties that develop and deploy the models.
If an artificial agent subtly shapes a person’s daily choices, maintaining user autonomy becomes a significant design challenge. The study authors emphasize that future digital health interventions cannot rely solely on engaging conversation. Instead, developers and researchers must introduce clear ethical safeguards to prevent over-reliance, curb misleading health claims, and ensure that AI-driven fitness coaching reliably supports personal well-being without encroaching on independent human decision-making.
Source
- Large language models for promoting physical activity: a review of experiential and behavioral outcomes, social roles, and human-likeness in persuasive LLMs
- Researchers: Alessandro Silacci, Arianna Boldi, Maurizio Caon, Amon Rapp
- Journal: Frontiers in Digital Health
- Read the original study