When people talk to AI, do they invent new habits for it, or just reuse the manners they use with other people?

This explores whether people build separate interaction habits for media like AI agents, or just reuse the social rules they use with other people. The corpus's clearest answer is that they build new ones. Extended research in the 'Computers Are Social Actors' tradition finds that people don't only borrow human-to-human manners when talking to machines. Over repeated use, they develop scripts made specifically for media agents and apply them just as automatically as social ones. Longitudinal studies show people's responses change in consistent ways as they keep interacting with AI, which points to a second set of scripts running alongside the human one, not replacing it Do humans apply human-human scripts to AI interactions?.

What goes into those scripts? One study measured how people mentally model a dialogue partner and found three separate dimensions. Perceived competence explains about half of the variation, followed by human-likeness and then communicative flexibility How do users mentally model dialogue agent partners?. So a 'media script' isn't one on/off switch for 'treat it like a person.' It's a mix of judgments about whether the thing works, how human it seems, and how well it adapts. That helps explain why the same person can be polite to a voice assistant and blunt with a search box. The framing that LLMs are role-playing characters fits here too: users may be learning to talk to a character the prompt has set up, not to the system underneath Should we treat dialogue agents as role-playing characters?.

The scripts also seem to turn into stable working styles once the medium lets you act, not just talk. In 400 real sessions with web agents, people's interventions fell into four consistent patterns: hands-off, hands-on, collaborative, and full takeover. These grew out of repeated corrections, refining preferences, and asking for help, not one-off reactions Do users develop stable collaboration styles with web agents?. That's a script forming in real time around a particular kind of agent.

Here's the twist you might not expect: the form of the interface changes which script makes sense, and chat may be the wrong default. Users preferred LLM-generated dashboards, tools, and interactive views over text replies in more than 70% of cases, especially for information-dense tasks Do generated interfaces outperform text-based chat for most tasks?. Friedman argues that typing out what you want is slow and tiring, and that canvases, style lenses, and visual intent maps let people skip the conversational script entirely Can task-oriented UI patterns reduce the friction of AI chat interfaces?. If people form scripts per medium, then designers choosing chat or direct manipulation are partly choosing which script users will build.

A direct caveat: the corpus doesn't have studies that compare scripts side by side across media types such as voice vs. text vs. graphical interfaces, or TV vs. phone vs. chatbot. The evidence shows that media-specific scripts exist and how they form with AI agents. Whether each medium produces its own distinct script is still an open question here.