inferwire
/
AI·3 min read

Dating-App Users Want AI Agents to Chat for Them, Not With Them

In a survey of 2,617 users of one dating platform, far more were willing to let an AI agent message matches for them than to talk to someone else's agent.

TL;DR

  • Of 2,617 Fledge.Love dating-app users surveyed, 40.7 percent rated letting their own AI agent message matches above engaging with another's agent; 2.1 percent the reverse [^1].
  • Under the study's model, only about 4 to 13 percent of random pairings would produce an agent contact that the receiver engages with [^1].

Background

Dating apps have started adding generative AI, first as assistance, such as AI-written profile summaries and suggested openers, and now as delegation: agents that hold conversations on a user's behalf [^1]. Such a feature only works if both sides accept it. Every delegated message needs someone willing to receive it, and earlier research found that people who suspect a message was written by AI trust it less [^1].

What happened

The study, by researchers at Fleamily, Inc. and the University of Notre Dame, has been accepted at the WSDM 2027 conference [^1]. Two surveys reached active users of Fledge.Love through an in-app prompt [^1]. The one on agents, fielded from November to December 2025 in Russian and English, showed respondents mockups of an agent that messages their matches on their behalf, with a configurable activity level and tone. It asked how they felt both about using one and about receiving messages from someone else's agent [^1].

Willingness to send and willingness to receive turned out to be closely related but distinct [^1]. Endorsing one's own agent took much less enthusiasm than engaging with a counterpart's, and the average propensity to deploy an agent was roughly three times the propensity to engage with one [^1]. The most common answer pattern was rejecting both; only 11.0 percent fully endorsed both sending and receiving [^1]. Interest was higher among users whose matches rarely turned into conversations, and lower among women and long-time users [^1].

Why it matters

For a platform, the asymmetry means agents would mostly produce messages their recipients say they don't want. In a random-pairing simulation based on the survey answers, only 4.4 to 12.8 percent of directed pairs led to an agent conversation the receiver engaged with [^1]. The imbalance also runs along gender lines: agents deployed by men and contacting women got 8.4 percent engagement per contact, against 13.4 percent for agents deployed by women contacting men, so the heaviest traffic would meet the coldest reception [^1].

The authors also price design options. Letting only users who would themselves engage with agents deploy one excluded 65 percent of likely deployers and halved interaction volume; about a quarter of users want to send agents but not receive them [^1]. Routing agent contacts only to the quarter of users most open to them raised engagement per contact from 11.6 to 39.4 percent [^1]. The results are stated preferences, not observed behavior, and the authors note they can't fully separate the role effect from differences in how the two scenarios were worded [^1].

Practical example

Say your dating app adds an agent that chats with matches while you're busy, and you switch it on to keep conversations from stalling. Going by this survey, the person on the other end is far less likely to welcome a chat with your agent than you were to turn it on, and the gap is widest when the agent writes on a man's behalf to a woman [^1]. Whether your agent's opener lands at all would then depend on the design choices the authors discuss for platforms, such as disclosing agent use, opt-in for receiving agent messages and routing them to people open to them [^1].

Related gear

We recommend this book because it applies ideas from computer science to everyday human decisions, including how to go about choosing a partner.

AdvertisementAmazon

Algorithms to Live By: The Computer Science of Human Decisions

Sources

  1. [1]arXiv — Delegation Asymmetry in Agentic Recommender Systems: Measuring Two-Sided Receptivity in Online Dating