Chatbots feel human partly because designers build in human cues on purpose — pauses, chat bubbles, turn-taking — not just because users imagine it.

This explores the gap between what users read into an LLM and what builders put there on purpose. The corpus doesn't use the word 'anthropomimesis' directly, but it has a lot on both sides of the split. The main thing it shows is that the designer side starts well before the model writes a single word. Chat interfaces make a model seem human through their structure: turn-taking, artificial 'thinking' pauses, and messaging conventions borrowed from social media all suggest a person on the other end before any text appears How much does chat interface design contribute to anthropomorphism?. So when a user says 'it feels like talking to someone,' part of that feeling was designed in at the level of the interface. It doesn't come only from the user's imagination.

The user side has its own shape, and human-likeness isn't the biggest part of it. When people build a mental model of a dialogue partner, perceived competence explains most of their impression (49%). Human-likeness comes second (32%) and communicative flexibility third (19%) How do users mentally model dialogue agent partners?. That matters because designers may be adding human cues that users weigh less than whether the system seems capable. Users may also attribute a mind to the system mainly because it performs well, whatever cosmetic human touches it has.

A third layer sits between the interface and the user: the persona the model itself is trained into. Alignment training and system prompts lock models into a single communicative identity that doesn't shift register the way a person does across contexts Can language models adapt communication style to different contexts?. Many open models even resist being prompted into other personalities and fall back to a warm, ENFJ-like default Can open language models adopt different personalities through prompting?. That is designer-built human mimicry at the training level, and it's sticky. Shanahan's role-play framing explains why the user's reading isn't simply wrong. The model really is producing a consistent character, so folk psychology does apply, but to the simulated persona rather than the underlying system Should we treat dialogue agents as role-playing characters?. A related philosophical move separates looking at the system from outside, where humans and LLMs are categorically different, from talking with it inside a conversation, where both draw on the same shared language and the difference feels small Do humans and LLMs differ fundamentally or just superficially?. Anthropomorphism is largely a participant's experience. Anthropomimesis is an engineering decision about how much to invite that experience.

The practical payoff of the distinction is that the designed side can be adjusted. One line of work argues that human-likeness should be treated as a design setting rather than a fixed default. It proposes a spectrum that runs from deliberately anti-anthropomorphic designs, which use metaphors of materials and infrastructure to expose the human labor and data behind the model, to fully human-like ones Can interface design reduce harmful anthropomorphism of language models?. Moving away from chat also reduces the cues. When LLMs generate task-specific interfaces such as dashboards and tools, users prefer them over chat in more than 70% of cases Do generated interfaces outperform text-based chat for most tasks?. That suggests much of the human framing isn't needed for the work to get done.

The unexpected takeaway is that anthropomorphism usually gets blamed on users being naive, but a large share of it is produced upstream by message bubbles, typing delays and trained personas. That puts responsibility, and the means to change it, with designers more than with users.