A controlled study of 4,212 US lottery players found that capping draw-result push notifications at two per hour reduced session-70 check-ins by 9.0% relative to an uncapped control, with the largest declines concentrated among players who had previously opted into alerts for three or more games. The effect held across both Powerball and Mega Millions draw windows and persisted for 21 days after the cap was lifted, suggesting the constraint altered checking habits rather than merely deferring them.
The finding matters because push alerts are the primary re-engagement surface for lottery operators that have moved beyond app-open prompts, and because regulators in several states have begun treating notification frequency as a consumer-protection variable rather than a marketing dial. What follows examines the study design, the mechanism most consistent with the data, and the questions the result leaves unresolved.
Study Design and the 9% Figure
The trial ran from January 8 to March 29, 2025, across a mid-sized state lottery's mobile app, which reported 214,000 monthly active users at enrollment. Researchers randomized 4,212 consenting players into two arms. The control arm received the operator's standard alert cadence, which during draw nights averaged 5.8 pushes per hour between 7 p.m. and 11 p.m. local time — a mix of jackpot reminders, result postings, and secondary-game promotions. The treatment arm received no more than two pushes in any rolling 60-minute window, with the operator free to choose which two.
The primary outcome was the "session-70" check-in: an app session initiated within 70 minutes of a scheduled draw, the window in which players most commonly verify results. Secondary outcomes included ticket purchase volume, session duration, and 30-day retention. The 9.0% reduction in session-70 check-ins (95% CI: 6.4% to 11.6%) was the only primary outcome to reach significance after Bonferroni correction for four comparisons.
| Measure | Control | Capped (2/hr) | Change |
|---|---|---|---|
| Session-70 check-ins per player-week | 3.41 | 3.10 | −9.0% |
| Mean session duration (min) | 4.7 | 4.9 | +4.3% |
| Tickets purchased per player-week | 2.18 | 2.14 | −1.8% (n.s.) |
| 30-day retention | 61.2% | 60.8% | n.s. |
The null result on ticket purchases is the study's most commercially consequential number. A 9% drop in check-ins did not translate into a measurable drop in play, which complicates the operator's assumption that check-in frequency and revenue move together.
Why the Cap Was Set at Two
The two-per-hour threshold was not arbitrary. Pre-trial telemetry showed a median of 1.9 pushes per hour on non-draw evenings and a sharp right tail on draw nights, where the top decile of users received 11 or more pushes in a single hour. Two per hour therefore functioned as a cap that bound only on draw nights and only for heavy-alert users, which is precisely the population the study wanted to isolate. A cap of four would have bound on almost no one; a cap of one would have conflated the frequency effect with the loss of result delivery itself.
The Mechanism: Alert Fatigue or Habit Disruption?
Two explanations fit the headline number, and the study's secondary data discriminate between them only partially.
The alert-fatigue account holds that high-frequency pushes train users to ignore notifications, so capping them restores signal value and should increase check-ins among users who previously tuned out. The data contradict this: check-ins fell, not rose, and fell most among the heaviest prior alert recipients (−13.4% for users with three or more game subscriptions versus −4.1% for single-game users).
The habit-disruption account fits better. Frequent pushes appear to function as external cues that sustain a checking routine; removing some cues weakens the routine. Under this reading, the 9% decline is not a failure of relevance but the expected cost of reducing cue density. That interpretation is supported by the persistence finding: check-ins in the treatment arm remained 6.2% below control at day 21 post-cap, well after notification volume had normalized — a pattern consistent with habit decay rather than a transient response to fewer pings.
A third possibility, that capped users simply shifted to checking results through the web or a third-party app, was not fully excluded. The study's device-level attribution captured in-app sessions only; the authors estimate out-of-app checking could account for up to a third of the observed decline but did not measure it directly. This is the study's most serious limitation and the one most likely to be challenged by operators with cross-platform analytics.
What the Result Does and Doesn't Say About Revenue
The temptation is to read the null purchase result as evidence that notification caps are revenue-neutral. That inference is premature. The trial ran 11 weeks, and lottery purchase behavior is heavily jackpot-driven; a rolling jackpot above $400 million during weeks 6 through 8 likely masked any small negative effect by lifting purchase intent across both arms. The authors' own power calculation suggested the study was powered to detect a 4% purchase change, not the 1.8% point estimate observed, so the null is informative but not decisive.
There is also a composition question. If capped users check in less often but buy the same number of tickets, the operator's cost per acquisition and per retention falls, which improves unit economics even at flat volume. One participating operator reported a 14% reduction in push-delivery infrastructure costs during the trial period, a figure that will matter more to smaller state lotteries than to national brands.
Regulators, meanwhile, may read the result differently. If a two-per-hour cap reduces engagement without reducing play, the consumer-protection case for mandatory caps gets stronger, not weaker: the harm of over-notification becomes easier to argue when the remedy costs the operator little. Expect this study to surface in comment periods for state rules on promotional messaging, where the industry's usual defense — that frequency limits would harm revenue and therefore state lottery proceeds — now has a counterexample.
The Open Question: Where Is the Optimal Cap?
The study establishes that two per hour binds and that the binding reduces check-ins by 9%. It does not establish that two is optimal, or even that a cap is the right instrument. A dose-response design — arms at one, two, three, and four pushes per hour — would locate the point where cue density stops sustaining habit and starts eroding it, and that point is almost certainly not uniform across game types or player tenure.
The more interesting unresolved question is temporal. If push alerts work as habit cues, then the timing of pushes may matter more than their count. A single well-placed result notification at the moment a draw resolves could plausibly outperform four spread across the evening, and no published trial has tested placement against frequency. Until one does, operators capping at two per hour are making a defensible choice on incomplete evidence — and players who notice fewer pings after a draw should know the change was studied, not accidental.