Mobile marketers spent 2026 sending more push notifications than at any point before it, and getting fewer opens for the effort. That alone would be a familiar story of diminishing returns. What makes this cycle different is a new layer standing between send and open: on-device AI that decides which notifications even surface, independent of anything the marketer configured.
Volume keeps climbing while opens keep shrinking. Airship’s 2026 benchmark report, built from more than 3 billion app users and thousands of apps across 15 verticals, found that push volume at the 90th percentile rose 15.71 percent year over year on iOS and 18.16 percent on Android, while iOS direct open rates fell 13.71 percent over the same period. Opt-in rates, meanwhile, are converging across platforms: the median sits at 48.85 percent on iOS and 53.3 percent on Android, less than five points apart. Marketers are reaching roughly the same share of users and sending them more messages, and a shrinking fraction opens any given one.
A filter marketers do not control now decides what lands. OneSignal’s 2026 State of Customer Engagement report, drawn from benchmarks across 200,000-plus apps and a survey of 500-plus marketers and product teams, found that 48 percent of marketers already believe AI-driven filtering on the device or platform side affects which of their messages get delivered at all. The report frames the fix as sending fewer, better-targeted messages rather than more of them, which cuts against the instinct to compensate for falling open rates with higher volume. Once a system is triaging notifications before a user ever sees the lock screen, open rate stops measuring what marketers think it measures.
Install-era thinking is the wrong lens for a delivery problem. Moburst’s 2026 attribution analysis makes a related point about a different metric: an install means nothing if the user it came from churns on day one, so teams should optimize toward down-funnel events such as trial starts, first purchases, and retention rather than top-of-funnel counts. The same logic applies to push. A send that is filtered before delivery, or opened without ever producing an action, tells a marketing team almost nothing, yet most push dashboards still report sent and opened as if those numbers settle the question.
The fix is measuring the funnel, not the send. Teams that still report push performance as sent versus opened, without a delivered stage and without tying opens to a down-funnel event, cannot tell a filtering problem apart from a creative problem or a timing problem. Breaking volume, delivery, open, and conversion out separately, by vertical and by segment the way Airship’s benchmark data is structured, is the only way to know whether the 2026 jump in send volume is buying anything or just adding noise ahead of an AI layer that increasingly decides who sees what.
Sources referenced in reporting: Airship, Mobile App Push Notification Benchmarks for 2026, OneSignal, 2026 State of Customer Engagement Report, Moburst, Mobile Attribution in 2026: What Marketers Actually Need to Know.