Are you starting to sound like the chatbots you talk to?

This article was featured in the September 4, 2026, edition of the IBM Think newsletter. Get it in your inbox.

Ever argued with a customer service bot until you won? The bot stays relentlessly cheerful, because it’s a bot, and that’s what bots do. The problems start when it can’t handle whatever is weird or specifically human about the situation, so you, the human, adapt. (Well, you usually get a little angry first, and demand to speak to a fellow human, and when that doesn’t work then you adapt.) The niceties and the backstory go, and you restate what you need in Short. Clear. Sentences. Until the dang thing finally coughs up your refund.

A team of researchers has put some thought into how we’re modifying the way we communicate to get through to these bots. In a new paper published in AI & Society, Selcen Ozturkcan of Linnaeus University and co-authors at Aarhus University and the University of Birmingham argue that the influence between people and AI runs in both directions. The systems are engineered to sound more human. And we, over repeated exchanges, also start adapting ourselves to what the systems can recognize and reward. The researchers call the result “robotoid humanness”—a state in which we start speaking and writing in ways a machine can easily parse, because when we do we get rewarded for it.

Ozturkcan describes the process as a feedback loop. “The consumer acts, the robot responds, and with repeated exposure, in time the consumer internalises the exchange,” she said in a press release announcing the paper. AI can accelerate that loop by adapting to user input in real time, and “the innate instinct to mirror can lead to people returning the robot’s communicative behaviours, making them more robot-like.” Plus, if it gets you the refund, it starts to feel like the right way to talk. Which is a little unnerving.

Quick caveat: this paper isn’t a study. The authors didn’t measure this actually happening, although they do offer testable propositions for future researchers. For now robotoid humanness is just a hypothesis with a rad name that sounds straight out of an Isaac Asimov novel.

Still, the underlying instinct isn’t hard to recognize. We already know people bend their language toward whatever gets results from AI—just look at the phenomenon of resumes custom-designed to get past AI screeners, or those weird one-sentence paragraphs everyone uses to appeal to the LinkedIn algorithm gods. If a system keeps rewarding one way of communicating, people have a funny habit of continuing to communicate that way.

So the next time you’re trying to charm a bot into doing what you want, it’s worth a little self-reflection on what you said to get there. Or you can ignore all that and just feed the thing absolute nonsense and watch it try to make sense of it all. Unlike a person, a bot doesn’t seem to mind. (Well, yet.)