A small study published in Diabetes Care, the journal of the American Diabetes Association, shows that personalized text messages with positive and negative feedback are more effective than texts that are simply reminders.

The study was conducted by researchers at the Israeli Institute of Technology and one researcher was affiliated with Microsoft Research. They split 27 patients into a seven-patient control group and a 20-patient intervention group, both consisting of patients with Type 2 diabetes who did not already perform regular physical activity. Both groups received smartphones with built-in activity tracking sensors and received regular text messages.

The control group received weekly reminders to exercise. The intervention group received messages from an app with a learning algorithm. The app alternated between negative feedback, positive feedback, and positive feedback with a social component, choosing which type of message to send based on how effective it had been previously in getting the user to move.

The intervention group saw a small positive change in activity, whereas the control group did not. In addition, participants were surveyed and the intervention group reported that the messages helped them maintain or increase activity levels, while the control group reported no effect.

"The learning algorithm improved gradually in predicting which messages would lead participants to exercise," the study authors wrote. "On average, the best daily message was a positive-feedback message with a social component (average improvement of 8.8 percent in activity in the day following such a message), and the best consecutive messages were a positive social message after a negative-feedback message (42.7 percent improvement). The least effective message was a positivefeedback message without social reference (9.9 percent reduction), and the least effective consecutive messages were a negative-feedback message after a positive social message (-61.4 percent)."

The study follows a trend in text message studies to explore not only whether text messages are effective for changing health behaviors, but which types of text message are most effective, particularly exploring tailored text messages vs stock messages. Having a computerized algorithm do the tailoring is also important, because it makes tailored text messages more scalable to large populations.