Stanford's AI Coach Outperforms Human Health Experts at Motivating Exercise
A research app called My Heart Counts has found that people respond better to personalized exercise messages generated by AI than to messages written by human health professionals. The finding emerged from a decade-long cardiovascular study at Stanford University School of Medicine that enlisted over 100,000 participants.
The app's AI coach draws on behavioral psychology research to match messages to where each person stands in their readiness to change. Someone who has never exercised gets different messaging than someone already jogging three times a week but at risk of backsliding. The system generates short, encouraging nudges rather than engaging in open-ended conversation.
Participants preferred the AI messages because they were fine-tuned specifically to the transtheoretical model of behavior change-a framework developed in the 1980s to describe how people adopt new behaviors. The AI generated more precisely tailored messages than human experts typically could.
Why This Matters for Healthcare Access
Personalized health coaching from a human professional is expensive, time-limited, and unavailable to most people who need it. An AI system delivering comparable results at scale could change that equation.
Heart disease remains the leading cause of death worldwide. Regular physical activity is one of the most effective ways to prevent it, yet most people struggle to sustain exercise routines. Knowing you should exercise and actually doing it are separated by what the researchers call a canyon that willpower alone rarely bridges.
How the App Works
My Heart Counts launched in 2015 in partnership with Apple and captured data from over 40,000 participants in its first two weeks. The newest version uses a randomized crossover trial to measure whether the AI coaching actually changes behavior.
After a baseline period, participants randomly receive either AI-generated, stage-matched messages or generic step reminders for one week, then switch conditions. Because each person experiences both, they serve as their own control group. This addresses a common problem in digital health research: apps tend to attract people already motivated to change, making it hard to know if the technology works or if it just tracks enthusiastic early adopters.
The primary outcome is deliberately simple: daily step counts.
The app goes beyond step counting. Users complete surveys, perform guided physical tasks, and view a personalized Heart Health dashboard that combines activity data, sleep patterns, and optional clinical biomarkers from Apple Health. The dashboard includes factors beyond the usual cardiovascular risk markers-blood pressure, glucose, and lipids-to show users how sleep and mental health affect heart disease risk.
Trust and Transparency
The researchers distinguish between different types of healthcare AI. Low-risk behavioral coaching carries less downside if the output is imperfect than a diagnostic or treatment recommendation would. AI can also be more consistent than humans and less subject to fatigue or training gaps.
My Heart Counts is not a medical device. It does not diagnose disease, recommend treatments, or replace physician judgment. The AI itself is intentionally constrained-it does not access full health records or unstructured personal data, operating instead on a limited set of structured inputs like activity patterns and survey responses.
The platform's code is fully open-source. Aggregated, anonymized data flows to vetted researchers through governed access pathways rather than being controlled by a single commercial entity.
Building for Scale
The newest version runs on Stanford's Spezi framework, a modular open-source architecture that lets the team update the app without rebuilding from scratch. This should accelerate innovation across cardiovascular research.
Future updates could extend the dashboard to help users reduce risk for other diseases like breast cancer or osteoporosis.
How to Participate
Adults 18 and older in the United States and United Kingdom can download the app, consent to participate, and contribute data to cardiovascular research at Stanford. The study is open to English and Spanish speakers.
The researchers frame their approach as the opposite of gatekept science: "Our goal is increased population health," they say of their open-data philosophy.
Learn more about AI for Healthcare and how Generative AI and LLM systems are being applied to clinical and behavioral health contexts.
Your membership also unlocks:
