Google introduced an AI-powered personal health coach built with Gemini models on October 27, 2025. In an announcement by Shwetak Patel and Florence Thng, the company said a public preview starts the next day for eligible US-based Fitbit Premium Android users, with iOS to follow. The coach runs inside the Fitbit app.
Availability and how to join
- Public preview begins the day after the announcement and rolls out over the following week. Details and sign-up are available on the Health coach public preview page.
- Eligibility at launch: US-based Fitbit Premium subscribers on Android. iOS expansion is planned.
- Participants must consent to share Fitbit data to receive personalized insights.
- Google states the coach is not a medical device and is intended for general wellness and fitness only.
- Users can share feedback in the Fitbit community forum.
How it works
The coach is an AI-first experience spanning Fitbit, Google Research, and Google DeepMind. It uses a multi-agent framework to coordinate:
- A conversational agent to handle user interactions.
- A data science agent to retrieve and analyze user data.
- A domain expert to generate fitness and wellness plans.
The system performs numerical reasoning on physiological time series, building on prior work such as PH-LLM. Guidance is grounded in established coaching and fitness frameworks, including weekly cardio load recommendations documented in Google's coaching and fitness frameworks. The coach addresses sleep, fitness, and general health questions using personal baselines and population-level statistics where applicable.
Research, evaluation, and oversight
- Expert input includes Google's Consumer Health Advisory Panel and professional fitness coaches.
- User feedback is informed by large-scale, consented Fitbit Labs studies, including the Sleep and Symptom Checker Labs Labs.
- Outputs are evaluated with the SHARP evaluation framework measuring safety, helpfulness, accuracy, relevance, and personalization. Google cites more than 1 million human annotations and over 100,000 evaluation hours across relevant clinical and fitness domains.
Background
Google describes development as iterative and science-based, with ongoing evaluation and updates. The company details the end-to-end approach and technical design in its technical overview, including how real-world performance and autoraters inform continuous improvements alongside SHARP.
Sources
- Google Research technical overview: How we are building the personal health coach
- Public preview details: Health coach public preview
- Model documentation: Gemini models
- Background on time-series reasoning: PH-LLM
- System architecture: multi-agent framework
- Fitness methodology: coaching and fitness frameworks
- Expert oversight: Consumer Health Advisory Panel
- User studies: Labs
- Evaluation methodology: SHARP evaluation framework
- Eligibility context: Fitbit Premium
- Feedback channel: community forum
.svg)
.svg)
.svg)