Why Isn't Anyone Responding to My AI Bot?
If you're a solo entrepreneur who's hooked up a chatbot to auto-reply to customer inquiries, or handed sales outreach messages over to AI, you've probably wondered about this at some point: the setup looks solid, the sentences read naturally, so why isn't the response rate climbing the way you expected? The usual troubleshooting move is to polish the prompt further or tweak the message's tone — in other words, to perfect the sending side.
But the point that's easiest to miss lies elsewhere: how the other person reacts once they realize the message wasn't written by a human but by AI. Using AI yourself and having the other party accept AI are two entirely different problems, and if you expand automation without checking the latter, you may be pouring water into a leaky bucket. The paper introduced here puts exactly this question to the test, using dating platforms as its stage.
About the Study
Daria Leshchikova, Valentina V. Kuskova, Dmitry Zaytsev, and Valerii Klimov start from the observation that autonomous AI agents — ones that carry on conversations with a match on the user's behalf — are emerging as a new design paradigm on matching platforms. The catch is that this only works if two separate forms of acceptance hold at once: users must be willing to delegate their own conversations to an agent, and they must also be willing to receive messages sent by someone else's agent. That second condition has barely been examined until now.
The researchers ran two large-scale surveys among actual active users of a major dating platform. One covered 2,894 respondents on AI-generated profile features; the other covered 2,617 respondents on autonomous conversational agents, fielded across two language groups. They then applied a statistical model that infers latent traits from response patterns, quantifying each user's underlying acceptance of AI-agent conversation beyond what the survey answers showed on the surface.
What the Study Found
The first finding was that willingness to send an agent and willingness to receive one are strongly correlated (r = 0.92) yet statistically distinct traits (ΔBIC = 52). In other words, within the same person, these two inclinations don't move in lockstep.
More specifically, the study revealed an asymmetry in delegation. The acceptance threshold needed to deploy one's own agent was -0.38, while the threshold needed to accept a partner's agent was +0.32, and full engagement required +1.39. On average, people were about three times more willing to deploy an agent than to accept one. Plugging these figures into a randomized-matching simulation of real pairing scenarios showed that the combination of one side deploying an agent and the other willingly accepting it occurred in only 4-13% of all directional pairs — with clear gender differences in that directionality as well.
The researchers also tested two design safeguards meant to address this gap. One is a mutual-consent condition, where a conversation only proceeds if both sides agree to use an agent. The other is a routing approach that filters and delivers agent contact based on the recipient's acceptance level. The mutual-consent condition filtered out about two-thirds of attempted deployments, cutting overall conversation volume by more than half. Conversely, routing based on recipient acceptance tripled the engagement rate per contact, and this effect held up even under cross-validation with a held-out target set (AUC = 0.88, a 3.1x lift by quartile).
Impact of Each Design Mechanism
Applying This in Practice
There are two takeaways for solo operators and planners. First, before expanding AI-driven customer response or sales automation, it's worth checking whether the other side is even ready to accept a message they know came from AI. No matter how much you polish the sending side, results won't improve if resistance on the receiving side goes unaddressed. Second, rather than blasting the same AI message to every prospect, it's more efficient to first identify — based on response history or channel characteristics — who is more likely to be receptive to AI contact. That's the same principle behind the study's finding that recipient-acceptance-based routing sharply boosted engagement. It's also worth considering, as a factor that can shape the kind of acceptance this study examines, whether to disclose upfront that a message was AI-written.
Caveats
This study measured acceptance in the specific context of dating platforms, based on self-reported surveys. There's no guarantee the same figures would replicate in actual sales messaging or customer-service settings. It's also worth noting that only partial measurement invariance was confirmed between the two language-group surveys, which should factor into how the results are interpreted. Still, the structural finding that willingness to send and willingness to receive are separate traits offers a takeaway worth keeping in mind for any effort to design communication mediated by AI agents.
