What the first answer said: People are about three times more willing to send messages through their own AI than to receive someone else's.
What checking found: That misreads the paper. Its 3.1 times figure is the engagement lift when agent messages are routed to the most receptive users. The 4–13% pair figure is right.
The finding
Using AI to write your own messages and receiving messages from someone else's AI are two different things. A study surveyed 2,617 users of a large dating platform in late 2025 about AI agents that act for them.
Willingness to send through an agent ran well ahead of willingness to receive from one. Under random pairing, only 4–13% of sender and receiver pairs combined both. Routing agent contacts to the most receptive users lifted engagement per contact about 3.1 times.
So agent messaging only works for a small, self-selected group. For everyone else, a message from someone's AI is unwelcome.
Why it mattered to the design
The operating law says members decide, AI suggests and moderates, and the platform never sends messages as a member.
In practice, AI may draft a reply inside Plus. The member reads it, edits it if they like, and sends it themselves. Meetup acceptance, face reveals and any consent always need the member's own tap; consent is never delegated to AI. AI never impersonates a member or acts as a romantic companion.
Earlier drafts left room for labeled AI-sent logistics messages in a later phase. Plan v2 removed that, and agents creating accounts or uploading photos was dropped outright.
How it was checked
AI research agents produced the first answer; it was then checked against primary sources. The paper on arXiv was opened. It matched the 2,617 sample, the fieldwork dates and the 4–13% pair figure. The research summary had turned a 3.1 times engagement lift from routing into a claim that people are three times more willing to send than receive. The paper does not say that, so the finding was corrected.
Sources
- Delegation Asymmetry in Agentic Recommender Systems: Measuring Two-Sided Receptivity in Online Dating, arXiv (Leshchikova, Kuskova, Zaytsev, Klimov), published 2026-08. Checked 2026-09-28.