Repeated interaction with AI-powered customer service agents may push consumers to reshape how they present themselves, adopting the patterns those systems most readily process, according to a paper published this month in the journal AI & Society.
The authors define robotoid humanness as a drift in which consumers come to experience themselves as most fluent, correct, or socially viable when they become compatible with machine legibility, machine pacing, and machine logic. They describe a three-stage mechanism: entry into a synthetic social reality that invites commitment; computational identity capture, in which the system returns a reduced profile back to the consumer as personalized recognition; and adaptive self-adjustment, in which users shift their self-presentation toward what the system can parse and reward.
Ozturkcan said the consumer acts, the robot responds, and over repeated exposure the consumer internalizes the exchange. Toral-Manson said the research examines a bidirectional influence in which robots become more like people and people become more like robots.
The paper extends the argument beyond recommendation and personalization engines to large language model chatbots, identifying three features it says produce comparable dynamics: persistent memory architectures that carry behavioral profiles across sessions, reinforcement learning from human feedback that favors affirming and legible responses, and embedding-based vectorization that renders user inputs as numerical representations.
The study is theoretical and reports no experimental data. Its authors state that no datasets were generated or analyzed, and they set out three testable propositions for later longitudinal and experimental work.
The authors say the drift is not inevitable. They write that effects should be stronger where anthropomorphic cues are strong, where consumers have lower self-concept clarity, and in service contexts that reward compliance and speed, including healthcare and formal service bureaucracies. Design regimes that build in friction and contestability, they write, may interrupt the loop.
