AI in Fitbit modestly increases step counts, with no established advantage over standard reminders at two months

Personalizing notifications with artificial intelligence may help people walk more, but the gain remains modest. In the PEARL randomized trial, conducted among Fitbit users, adaptive reminders produced a difference of 210 daily steps compared with no reminders at two months. However, their superiority over fixed or random reminders was not statistically established at that point.

Fitness tracker and phone illustrating step tracking and physical activity reminders.
Gadgets & TechEditorial illustration, not study data.

Four strategies compared in the Fitbit app

The researchers randomly assigned 13,463 users to four groups: no reminders, reminders with random content and timing, fixed reminders based on an initial questionnaire, or reminders personalized through reinforcement learning.

The last approach adapted the content and timing of messages. It drew on a behavioral framework distinguishing capabilities, opportunities to act, and motivation. The primary objective was to compare changes in average daily step counts between baseline and two months.

The primary analysis included 7,711 participants, with an average age of 42.1 years, of whom 86.3% were women. The provided abstract does not explain why not all randomized participants were included in this analysis.

An early advantage that does not hold up against all reminders

At one month, the group using the adaptive algorithm showed a statistically significant advantage over all three other groups: 296 additional daily steps compared with the no-reminder group, 218 compared with random reminders, and 238 compared with fixed reminders.

At two months, the study’s prespecified primary time point, the difference remained statistically significant compared with no reminders, at 210 additional steps per day. However, the differences compared with random and fixed reminders were no longer statistically significant. This does not demonstrate that these strategies are equivalent, but it prevents any conclusion that personalization offers a lasting advantage.

A finding about walking, not proof of clinical benefit

The trial supports the feasibility of automated personalization at scale and suggests a modest effect on walking over two months. It does not establish that this effect persists beyond that period or directly improves health.

The participants were volunteer Fitbit users, mostly women. The finding therefore does not automatically generalize to the entire population, other apps, or people who do not use a connected device.

Limitations to keep in mind

Only the PubMed abstract was consulted, not the full text. Follow-up was limited to two months, and the primary analysis included only some of the randomized participants. The reasons for this discrepancy, funding, and conflicts of interest were not verified.

What we can take away

Adaptive reminders outperformed no reminders at two months, but their added value over simpler reminders remains uncertain at that point.

Sources and provenance

Partager cet article

X f in

An error or a source to report? Contact the editorial team.