Why do people resist receiving AI agents in dating?

A pair of large surveys fielded to active users of a dating platform (N = 2,894 on generative profile features, N = 2,617 on autonomous conversational agents) finds that willingness to deploy one's own AI agent and willingness to receive a counterpart's agent are statistically distinct but highly correlated constructs (ρ = 0.92, ΔBIC = 52), separated by what the paper calls a "delegation asymmetry": the threshold to deploy an agent (θ = −0.38) sits far below the threshold to engage one sent by someone else (θ = +0.32, with full engagement at +1.39), a gap the paper describes as "a robust 0.7 SD asymmetry." Mean deployment propensity exceeds engagement propensity roughly threefold. Under a random-pairing counterfactual built from the stated receptivity scores, only 4 to 13% of directed dyads combine a deployer on one side with a willing receiver on the other, with the imbalance falling along gender lines: male-deployed agents contacting women achieve 8.4% engagement per contact (strict) against 13.4% for female-deployed agents contacting men.

The paper frames this as a four-stage "receptivity audit": an item battery that deliberately covers both the deploying and receiving role within the same respondent, a confirmatory graded-response measurement model with latent regression that locates thresholds on a common scale rather than summing scores, out-of-sample-validated propensities (holding out the target engagement item still yields AUC 0.88 and a 3.1x quartile lift under cross-validation), and a dyadic market simulation that turns those propensities into policy counterfactuals. Two of those counterfactuals quantify the design trade-off directly: a reciprocity rule, letting a user deploy an agent only if their own engagement propensity clears 0.5, excludes 65% of likely deployers and halves interaction volume; routing agent contact only to the top receptivity quartile of receivers instead raises per-contact engagement from 11.6% to 39.4% (strict) while touching just a quarter of receivers.

The excerpt's own framing situates this as a second reciprocity layer on top of existing two-sided matching research — not just mutual interest, as in reciprocal recommendation, but mutual acceptance of the contact medium itself. That layer resembles the asymmetric adoption threshold found where Why do consumers delay adopting agentic search?, and the receive-side routing lever behaves like the search-conditions dependency reported in Do LLM agents achieve fair marketplace outcomes when given ideal search?: both show a two-sided agentic market reaching good outcomes only when a structural condition — ideal search, or receptivity-aware routing — is engineered in rather than assumed. The underlying move of inferring a hidden propensity from a handful of items also recurs in Can a seller infer what a buyer really wants to pay?, though aimed there at price rather than consent. And the receive-side wariness this paper measures as the scarcer resource echoes the privacy and expectation costs found where Does chatbot personalization build trust or expose privacy risks?, suggesting receiver distrust of AI mediation is a recurring cost rather than one specific to dating.

The estimates are stated preferences elicited from concept mockups shown to self-selected respondents on one platform (called Fledge.Love in the paper), not behavior observed under deployed agents, and the paper's own limitations section notes that the direction in which stated and revealed receptivity might diverge is unknown; the English-language subsample is also small (n = 232) and confounded with population composition. The market simulation assumes random dyad formation and static propensities, setting aside homophily in receptivity and the likelihood that receivers update after good or bad agent encounters. This is a preprint, not a peer-reviewed publication, and the 4 to 13% of dyads figure is a counterfactual computed from the fitted model rather than an observed market outcome. What the finding supports, at the strength this evidence allows, is narrower than agents will transform matching: that an agent-mediated contact feature should treat receive-side consent, not deployment opt-in, as the binding constraint on whether a market for agent-mediated communication exists at all.