Does AI life coaching work?
Against human coaches
Two trials have compared an AI coach with human coaches. The first was two matching ten-month trials with undergraduate students: a chatbot coach and human coaches both helped people reach their goals more than no coaching did, and at the end the chatbot was as effective as the human coaches (Terblanche and colleagues, 2022). That chatbot was a narrow expert system built for one task, not a large language model.
The second put the two head to head. 114 people in executive coaching got either accredited human coaches or automated AI coaches, and the effects on goals, motivation, resilience and wellbeing were clear only for the human coaching (de Haan, Terblanche & Nowack, 2026). Two of its authors also ran the first study.
Chatbots and everyday habits
A review of 19 trials of chatbots for activity, diet and sleep found small to moderate effects: people moved more (SMD = .28 for total physical activity), ate more fruit and vegetables, and slept longer and better (Singh and colleagues, 2023). Fourteen of the 19 trials were rated low quality, and all were published before September 2022, before chatbots like ChatGPT.
Coaches built on large language models
Trials of coaches built on a large language model, as Forvel’s is, are recent and short. In one, 543 adults set a goal with one of several chatbots in a single session; guidance helped, and feedback on the goal helped most, but the study measured how well the goals were set, not whether anyone reached them (Schimpf and colleagues, 2026). In another, brief conversations with a GPT-4 chatbot using motivational interviewing raised people’s readiness to change (Meyer & Elsweiler, 2025), which is not the same as changing.
What isn’t known
Whether a coach built on a language model helps people reach their goals over months: no trial we could find has tested it. No study has tested Forvel either. The one head-to-head trial favoured human coaches, and this page won’t pretend otherwise.
What Forvel does with that
Forvel doesn’t rest on the AI coaching research. It is built on tools with stronger evidence: writing down what you did, which changed behaviour across 138 randomised trials, and if-then plans with a fallback for when something gets in the way. The coach is there to help you pick one thing and keep it, and every figure on the site comes with its source.
Sources
- Terblanche, Molyn, de Haan & Nilsson (2022). PLOS ONE, 17(6), e0270255.
- de Haan, Terblanche & Nowack (2026). Human Resource Development International, 29(4), 754–783.
- Singh, Olds, Brinsley, Dumuid, Virgara, Matricciani, Watson, Szeto, Eglitis, Miatke, Simpson, Vandelanotte & Maher (2023). npj Digital Medicine, 6, 118.
- Schimpf, Maier, Wyrowski, Christoforakos, Feuerriegel & Bohné (2026). Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, 1–24.
- Meyer & Elsweiler (2025). International Journal of Human-Computer Studies, 200, 103514.