Marketing researchers in Britain, Sweden and Denmark argue that the imitation between people and chatbots has started running in both directions. Models trained on enormous volumes of human-written text learned to imitate human traits convincingly, at least in writing. As the tools spread, the authors say, the influence reverses: chatbots begin shaping user behaviour and gradually make it more like their own. They call the result robotoid humanity, a state in which it becomes easier for a person to count as literate, correct and socially acceptable by matching machine comprehensibility, machine tempo and machine logic. The framework has not been tested in real conditions.
The authors' starting claim is that talking to a chatbot no longer resembles using an interface. It resembles a quasi-social encounter: the user treats the system as a participant that answers, adapts and establishes rules for the exchange.
From there the paper breaks the shift into three stages:
1. The user enters an artificial social reality that prompts them to treat the interaction as meaningful.
2. The chatbot builds a statistical distorted image of the user, fixing their identity in a machine interpretation.
3. The user begins adjusting to that image and adopts the communicative habits attached to it.
Selcen Ozturkcan, associate professor at Linnaeus University and a co-author, described the mechanism as an exchange that hardens through repetition: a person acts, the machine responds, and over many rounds the user internalises the structure of the exchange itself. Machine learning amplifies it, she said, because the system alters its behaviour using the user's own data and imitates the human more precisely each time. The natural human tendency to mirror an interlocutor's manner, on her account, can lead people to hand the machine back its own patterns and become more machine-like in the process.
Inci Toral-Manson, a marketing researcher at the University of Birmingham and a co-author, put it as two-way influence: robots become more like people, and people become more like robots.
The authors think the stakes are not trivial. As people adapt to a reality heavily shaped by AI, they argue, self-perception and even a person's sense of their own worth can change. Jean-Paul De Cros Peronard, associate professor at Aarhus University and a co-author, noted that robots and AI are already an ordinary part of customer service, so companies need to understand how people interact with such systems in order to use the technology effectively while staying ethical.
What the paper is, then, is a vocabulary and a staged diagram. That matters more than the framing around it, because the authors concede the model has not been checked against real behaviour. No sample, no effect, no measurement — a three-box sequence and a name for it. Naming a phenomenon is the cheap part of this kind of work; the expensive part is saying what observation would show the phenomenon is absent, and that sentence is not in the account. Stage two is the weakest link: the distorted image of the user sits inside a system nobody outside the company can inspect, which makes it a plausible metaphor rather than a testable object.
The one part I find convincing is the bit that needs the least theory. Linguistic convergence between conversation partners is ordinary; Ozturkcan is right that people mirror. The novelty is the asymmetry. The machine is also adjusting to the user, so the register both parties are converging on has no fixed location — and it was not arrived at by accident. Machine tempo and machine legibility are not properties of nature. Somebody at a company chose them, tuned them and shipped them.
That is the question the paper walks past. If users are drifting toward the register of the system, the lever sits with whoever sets the register, and the paper's practical advice goes instead to firms wanting to run customer service effectively and ethically. A piece of work that raises the possibility of altered self-worth lands on operations guidance.
Which leaves the evidence problem in an awkward place. The material that would settle whether any of this is happening — years of logs showing how millions of people's writing has changed while talking to these systems — is held by the companies that are themselves the independent variable.