What Apple and Google are doing to your push notifications

Apple and Google are redefining the landscape of push notifications, evolving them from simple delivery mechanisms to heavily intermediated platforms where their AI-driven systems actively parse, summarize, and prioritize content. This transformation fundamentally alters how messages reach users' devices, posing significant challenges for developers and marketers who previously had more direct control over their communication.

• Push notifications were initially conceived to address battery drain on mobile devices, centralizing message delivery through proprietary services like Apple Push Notification Service (APNs) and Firebase Cloud Messaging (FCM).
• Over the past 15 years, platform intervention has escalated from basic on/off toggles to sophisticated features like Android's notification channels (2017) and iOS's Focus modes and Scheduled Summary (2021), offering users finer control.
• A significant turning point occurred with Android 13 (2022), which made notification permissions an explicit runtime grant, leading to a predictable drop in user opt-in rates.
• Platforms justify this control by framing user attention as a scarce resource they are obligated to defend, simultaneously protecting their ecosystem and showcasing their advanced AI capabilities.
• This trajectory closely mirrors the evolution of email, where providers like Gmail introduced categorization (Promotions tab) and privacy measures (Mail Privacy Protection), forcing senders to contend with hidden classifiers and stricter authentication standards.
• Unlike email's open, federated protocols, push notification channels are entirely controlled by Apple and Google, meaning tokens can be invalidated at will and marketers cannot retain user subscription lists independently.
• On-device AI models, such as Apple Intelligence and Gemini Nano, are now at the core of this editorial process, locally summarizing and prioritizing notifications based on learned user behavior and platform-defined relevance.
• These models utilize specialized adapters for various tasks, and platform adjustments, like Apple's response to backlash over news summarization errors, demonstrate the dynamic nature of these AI-driven features.
• Marketers now have severely limited visibility into the fate of their notifications; traditional metrics like open and click rates become increasingly unreliable as messages are subject to unseen summarization, suppression, or deprioritization.
• Empirical research consistently shows users prefer transactional and conversational notifications over promotional content, highlighting that excessive volume negatively impacts user permission.
• The author advises marketers to strategically reserve push for critical, user-initiated alerts, personalize messages extensively, request permissions contextually, prioritize factual content in their copy, and shift communication weight to owned in-app surfaces that bypass platform editing.
• Future developments suggest an even more agentic role for AI, with systems like Siri and Gemini potentially acting on notifications on behalf of the user, further diminishing the relevance of traditional click-through metrics.

In conclusion, the article stresses that marketers must adapt to a new paradigm where they negotiate with an unseen, uninspectable AI editor that fundamentally prioritizes the user's experience over the sender's intent. Long-term success will depend on delivering highly relevant, user-centric messages through push for re-engagement, while leveraging owned in-app channels for broader communication where platform intermediation is absent.