2x Retention: How Particle Added AI Audio to a News App and Doubled User Engagement
Particle's Challenge
Particle is an AI-powered news app that curates and synthesises personalised newsfeeds for its users. The core product worked well for readers who had uninterrupted screen time, but the team identified a structural engagement gap: users commuting, cooking, or exercising could not read — and without an audio alternative, those moments were simply lost. Adding a "Listen to the News" feature required a voice layer that would not undermine the app's credibility. News is a trust-sensitive product; a synthetic voice that sounded robotic or unnatural would raise immediate doubts about content quality. The team needed AI-generated narration that users would accept without friction, and that could be delivered in real time against a continuously updating personalised feed. Existing text-to-speech APIs produced outputs that fell short of this bar — the gap between AI voice quality and genuine human delivery was audible enough to create hesitation.
How ElevenLabs Solved It
Particle integrated ElevenLabs' voice library and API to power the "Listen to the News" feature. ElevenLabs' natural-sounding voices — designed to produce output indistinguishable from professional narration — handled real-time audio generation of the personalised newsfeed without pre-recording or manual production steps. The API's low-latency architecture enabled on-demand audio generation as users navigated articles, keeping up with a continuously refreshing feed. The team selected voices from the ElevenLabs library that matched Particle's editorial tone and user demographic.
The Outcome for Particle
Users who engaged with the "Listen to the News" audio feature retained at twice the rate of users who did not. The audio narration quality cleared the credibility bar that mattered most for a news product: the company's Co-Founder and CEO Sara Beykpour noted that when the feature launched, users were "blown away that the voices are AI" — meaning the quality was high enough to produce surprise rather than friction. The 2x retention differential demonstrates that audio access during non-screen moments represents genuine incremental engagement, not just a format preference.
What Particle Learned
Particle's retention result is a clean demonstration of what happens when a product reaches users in contexts its primary format cannot. A news app that requires screen attention is functionally absent during the significant portion of a user's day when their hands and eyes are occupied. Adding a credible audio layer does not cannibalise reading — it fills the gap. The 2x retention figure is instructive because it reveals the direction of causality: it is not that audio users were already more engaged users who happened to try the feature; it is that the audio feature itself extended the product's reach into time slots that previously generated no engagement at all. For digital media and content platforms, this pattern recurs wherever the primary content format has an accessibility constraint — audio is the unlock for the commute, the kitchen, and the gym. The barrier to implementing it is no longer technical capability but voice quality, and that bar has moved.
- Retention lifts from audio features are additive, not substitutive — they capture engagement in time windows where the primary format cannot compete, rather than cannibalising existing reading or viewing behaviour.
- Voice quality is a trust proxy in news. A synthetic voice that sounds robotic undermines the editorial credibility of the content it reads; the threshold for acceptable AI narration in a news context is higher than in utility applications.
- Real-time on-demand audio generation against a personalised feed is a different technical problem from pre-produced audio content — it requires a low-latency API, not a studio pipeline. AI voice is the only format that makes personalised audio economically viable.
See how ElevenLabs can help your business address similar challenges.
Affiliate link — we may earn a commission at no cost to you.