Notifications are the lowest-common-denominator behavior change tool in health apps. Build a reminder. Pick a time. Send it daily. The assumption is that more pings mean more engagement.
The actual evidence is less comfortable.
A micro-randomized trial published in JMIR mHealth and uHealth by Bell, Garnett, Bao, and colleagues tested exactly how push notifications affect engagement in a behavior change app, using a study design specifically built to capture the effect of each individual notification decision rather than averaging across weeks of exposure. The results surface a specific and underappreciated problem: standard messages habituate. Novel messages don't, but they're almost never what health apps send.
How the study worked
The micro-randomized trial (MRT) is a relatively recent design in digital health research. Rather than randomizing users once to a condition and following them for weeks, MRTs randomize each user multiple times across the study period, allowing researchers to isolate the causal effect of a specific intervention delivered at a specific moment. This makes MRTs particularly suited to studying push notifications, where the relevant question isn't "do notifications work in general" but "does this notification, sent now, to this user, do anything."
The study enrolled 566 users of a real-world behavior change app. Sixty percent (n = 349) were assigned to the main MRT arm, with the remainder split between a standard notification control arm (n = 120) and a no-notification control arm (n = 97). The primary outcome was app opening within one hour of notification delivery.
Notifications were classified as either standard (routine reminder messages the app already used) or novel (new messages targeting perceived usefulness of the app, written specifically for the study).
What notifications actually do, immediately
The near-term effect of any notification was large. Receiving a message increased the probability of opening the app 3.5-fold within one hour compared to receiving nothing (RR 3.52, 95% CI 2.91–4.25). Standard messages produced a 3.66-fold increase (95% CI 2.99–4.48). Novel messages produced a 3.39-fold increase (95% CI 2.77–4.13).
Those numbers look strong. Notifications clearly work in the sense of generating immediate opens. This is why product teams keep shipping them.
But the study was designed to catch what happens next, and that's where it gets interesting.
The habituation finding
The key moderator analysis asked: does receiving a notification yesterday change the effect of today's notification?
For standard messages, yes. When a user had received a standard notification the day before, the near-term effect of today's standard notification dropped by 11% (RR 0.89, 95% CI 0.60–1.31). The confidence interval is wide enough that this finding isn't statistically definitive at conventional thresholds, but the direction is clear and the mechanism is theoretically coherent.
For novel messages, no. When a user had received a notification the day before, the near-term effect of today's novel message remained essentially stable (RR 1.01, 95% CI 0.68–1.52).
The researchers also tested a second moderator: whether the user had already engaged with the app the previous evening. Here the pattern reversed. Users who had already opened the app didn't benefit from a novel notification (RR 0.80, 95% CI 0.55–1.17), but standard messages were barely affected (RR 0.96, 95% CI 0.65–1.42). In other words, novel messages work best on users who haven't been engaging, and standard messages hold up better for users who have.
The long-term problem neither type solved
Here's the finding that should generate the most discomfort for product teams. Despite the real near-term effect of notifications, the study found no significant difference in time to disengagement across the three trial arms. The log-rank test returned chi-squared = 1.7, p = .42. Median disengagement from the app occurred at day 11 for both notification arms. For the no-notification group, it was day 7.
Notifications bought about four days of additional engagement at the median. That's something. But it's a long way from the sustained behavior change apps are meant to produce. The study wasn't designed to test long-term engagement optimization, and the authors are careful to frame this as a starting point rather than a verdict. The MRT is explicitly meant to generate evidence that informs adaptive notification policies, not to conclude that notifications are useless.
Still, the implication is plain: a static daily reminder policy, regardless of whether the message is standard or novel, appears insufficient to drive durable engagement on its own.
What the design literature says about why this happens
Habituation to repeated stimuli is one of the most reliably documented phenomena in behavioral neuroscience. Orienting responses, the neurological reflexes that direct attention toward new inputs, diminish with exposure to familiar stimuli. A notification that arrives at the same time every day, with the same framing, stops being processed as a signal worth acting on. It becomes background noise.
The digital health literature has been catching up to this for several years. A 2024 systematic review in JMIR on digital behavior change designs for habit formation identified cue variability as an underexplored mechanism: fixed cues can form associations, but they can also become invisible through overexposure. The analogy to notifications is direct.
The May 2026 umbrella review by Boumparis and colleagues in JMIR Mental Health found that across 20 systematic reviews covering 224,135 adolescent participants, the behavior change technique most consistently associated with successful digital interventions was "social support (unspecified)", not reminders. Reminders appear as infrastructure, not as the active ingredient. They sustain engagement long enough for other mechanisms to do work, but they're not the mechanism.
The 42-technique problem
The umbrella review surfaces a related design problem that compounds the notification issue. Across studies reviewed, 42 of the 93 documented techniques in the BCT Taxonomy v1 were absent from any of the analyzed apps. The most overrepresented BCTs were goal setting, self-monitoring, and feedback, each appearing in the majority of products. The least used included identity-based techniques, comparison of outcomes, and antecedents.
Notifications almost always belong to the "cue" and "scheduled consequences" categories. They're well-established, easy to implement, and familiar to product teams. The techniques that appear most absent are exactly the ones that require deeper personalization and context sensitivity: understanding who the user is and what situations are most likely to shift their behavior, rather than pinging them at 9 AM because that's when the cron job runs.
This isn't an abstract concern. If notifications are the dominant delivery mechanism and habituation is eroding their effectiveness over time, the question for product design isn't just "how do we make better notifications." It's "what do we use when the notifications stop working."
Specific design directions the research supports
The micro-randomized trial and the broader habit formation literature together point at a few concrete changes worth considering.
Vary message content more aggressively. The novel vs. standard distinction in the Bell et al. study isn't just about novelty as entertainment. It's about maintaining the signal-to-noise ratio of the notification channel. Teams that write five notification variants and rotate through them are doing something better than teams that send one. But even rotation can habituate. The research suggests that messages addressing why the app is useful (utility framing) may wear out differently than pure behavioral reminders.
Use notification suppression for already-engaged users. The study found that novel messages delivered to users who had already engaged the previous evening showed reduced effectiveness. Some notification platforms support this kind of context-aware suppression: don't send if the user already completed the target behavior today. Not doing this isn't neutral. It's actively diluting the channel.
Don't conflate opens with outcomes. The 3.5-fold increase in app opens is real but narrow. An open without the relevant behavior change doesn't count. Product teams that optimize notification copy for click-through rates are optimizing a proxy that may not map to the outcomes they actually care about.
Treat notifications as one layer of a behavior change stack, not the stack itself. The broader research consistently positions reminders as infrastructure rather than mechanism. A product relying on notifications as its primary engagement driver is building on a foundation the evidence says will erode.
The adaptive notification policy question
The Bell et al. study was designed explicitly to generate data for an adaptive notification policy, one where the system learns, at an individual level, when to send, what to send, and when to hold off. The micro-randomized trial design is well-suited to this because it generates per-decision causal estimates rather than averaged treatment effects.
This is where the research is pointing. Not "what notification works on average" but "what notification works for this user, given what they did yesterday, and whether they've been engaging." That's a harder product problem than deploying a daily reminder, but the evidence increasingly suggests the daily reminder is a temporary solution with a predictable expiration date.
The 11% habituation effect for standard messages isn't catastrophic in isolation. Stacked across weeks, for a user who has already received the same reminder sixty times, it probably is.
Paper reference: Bell, L., Garnett, C., Bao, Y., Cheng, Z., Qian, T., Perski, O., Potts, H. W. W., & Williamson, E. (2023). How Notifications Affect Engagement With a Behavior Change App: Results From a Micro-Randomized Trial. JMIR mHealth and uHealth. https://doi.org/10.2196/38342 | PMC: PMC10337295
Related: Boumparis, N., et al. (2026). Behavior Change Techniques in Digital Health Interventions for Promoting Adolescent Health Behaviors: Systematic Umbrella Review. JMIR Mental Health. https://doi.org/10.2196/84754
Related: Digital Behavior Change Intervention Designs for Habit Formation: Systematic Review. JMIR (2024). https://doi.org/10.2196/54375