Why does transforming first-person voice into third-person reduce notification engagement?

This explores why turning a direct, first-person message into a third-person report drains its pull on the reader — and the corpus addresses it not through notification-engagement studies but through the deeper question of what makes an utterance an *address* rather than a *description*. The cleanest lever here is the idea that some messages are speech acts that only function as interpersonal address. The work on why LLMs cannot actually raise alarm Can language models actually raise alarm about threats? argues that alarm isn't a content type you can paraphrase — it's a relationship enacted between an *I* who feels concern and a *you* being summoned to attend. First person carries that summons grammatically; third person dissolves it into a neutral statement about the world. The same words, stripped of the I→you axis, stop *doing* anything to the reader and merely *report* — which is exactly the failure of engagement the question names.

The systematic review of alignment dimensions sharpens why this lands emotionally rather than informationally Do different types of alignment serve different conversational goals?. It distinguishes lexical alignment (which drives task efficiency and comprehension) from emotional and prosodic alignment (which drive relational warmth and trust). A first-person notification recruits the emotional/relational channel — it feels addressed *to you*; a third-person rewrite keeps the lexical content intact but severs the relational channel, producing the 'cold customer-service bot' category error the paper warns about. Engagement is a relational outcome, and third-person voice quietly opts out of the relationship.

There's a civility dimension too. The work on proactive agents How can proactive agents avoid feeling intrusive to users? and on proactivity's efficiency gains Could proactive dialogue make conversations dramatically more efficient? frames an unprompted message as a social intrusion that has to earn its welcome through timing, boundary-respect, and a felt sense that someone is speaking *with* you. First person does some of that social labor for free — it signals an agent taking responsibility for the interruption. Third person reads as an impersonal system announcement, which is easier to dismiss precisely because no one appears to be standing behind it.

The quieter, more surprising thread: voice may also change how *believable* the message feels. The research on self-referential processing Do language models experience consciousness when prompted to self-reflect? shows that first-person, self-referential framing reliably shifts a model into producing committed experiential claims rather than hedged denials — first person pulls *toward* assertion. Relatedly, the 20-questions regeneration test Do large language models actually commit to a single character? shows that a consistent first-person voice is what readers latch onto as a stable speaker to relate to at all; third person offers no character to bond with. So the engagement drop isn't just stylistic — it's the removal of the very thing that makes a reader feel summoned, addressed, and answered-to. The thing you didn't know you wanted to know: 'engagement' here is less about content and more about whether the message constitutes someone speaking to you — and grammatical person is the switch.