Users Reject AI Agents That Speak on Behalf of Match Prospects

TL;DR

• Researchers measured user acceptance of autonomous AI conversation agents across dating platforms, discovering a strong two-sided delegation asymmetry.
• While individuals comfortably deploy AI agents to filter potential matches, they reject receiving AI-generated communications from prospective human partners.

Background

Matching platforms increasingly test autonomous language agents that converse on behalf of users. These automated representatives screen potential partners, initiate icebreaker messages, and coordinate schedules. However, algorithmic matching depends on mutual trust. If a user discovers that an incoming message originated from an artificial intelligence agent rather than a human, the perceived authenticity of the match collapses. Understanding how humans perceive dual-sided automated mediation is essential before platforms mandate agentic communication.

What happened

A team of researchers conducted two large-scale empirical surveys involving 2,894 active users on a major online dating platform to evaluate receptivity toward conversational software agents [^1]. The study measured two distinct operational dimensions: outbound delegation (a user employing an agent to message others) and inbound receptivity (a user receiving messages drafted or sent by someone else's agent) [^1].

The empirical findings exposed a pronounced preference gap that researchers termed delegation asymmetry [^1]. While a substantial portion of respondents expressed willingness to delegate repetitive messaging and initial candidate filtering to autonomous software, an overwhelming majority rejected receiving agent-driven outreach from other participants [^1]. Participants viewed inbound automated messages as deceitful, low-effort, and violating implicit social norms of human courtship [^1].

The researchers noted that this asymmetry poses a structural adoption bottleneck for agentic recommender systems [^1]. Standard single-agent interaction frameworks fail when deployed in bilateral social environments [^2]. When both participants in a potential match deploy autonomous agents, communication devolves into machine-to-machine context negotiation, stripping human intent from the interaction pipeline [^1]. Furthermore, disclosure requirements did not solve the issue; informing users that an incoming text was agentic reduced engagement further [^1].

Why it matters

This study highlights a fundamental friction point in consumer agentic software: the asymmetry between personal convenience and interpersonal expectations. Software developers frequently design artificial intelligence assistants under the assumption that efficiency benefits both ends of a communication channel. In reality, human relationships rely on proof of effort. When one party automates conversational effort, the perceived value of the interaction drops significantly for the recipient.

For product managers building multi-sided platforms, delegation asymmetry complicates feature rollout strategies. If platform operators deploy autonomous conversational agents without accounting for recipient friction, overall user retention declines. Users who suspect they are conversing with synthetic personas disengage from the ecosystem entirely. This dynamic forces matching services to rethink agent boundaries, shifting software roles from active conversational proxies toward passive, background recommendation tools.

The findings also provide critical lessons for enterprise messaging and automated workflow systems. While automated scheduling agents and customer outreach bots operate smoothly in transactional corporate settings, bilateral trust-sensitive domains—such as executive recruiting, investor relations, and high-stakes sales—face similar social backlash. Replacing human effort with language model delegation risks signaling indifference, ultimately damaging brand credibility and trust.

System architects must therefore rethink how autonomous agents integrate into human networks. Rather than allowing software agents to impersonate human voice directly, platforms need clear boundaries that preserve human agency. Using artificial intelligence to summarize profiles, highlight shared interests, or suggest meeting slots creates tangible utility without triggering the uncanny valley of automated conversation.

Practical example

Imagine a user named Alex logging into a dating app on a Tuesday morning. The app offers a new automated assistant that promises to save time by chatting with potential matches automatically.

Alex enables the feature. The agent reviews profiles, sends custom greetings, and answers initial screening questions on Alex's behalf while Alex is busy at work. Alex enjoys the time saved.

An hour later, Alex receives a thoughtful message from a prospective match named Sam. Alex opens the chat, excited by the personal tone and specific questions about shared hobbies.

However, a small platform tag reveals that Sam did not write the message. Sam's AI agent generated and sent the text automatically. Alex instantly feels misled, loses interest in meeting Sam, and closes the application. The perceived lack of genuine human effort turns what seemed like a helpful feature into a reason to leave the platform.

Related gear

We recommend this book because it explores how computational algorithms and delegation dynamics govern human decision-making and social interactions.