How much does chat interface design contribute to anthropomorphism?

The paper argues that anthropomorphism in LLM interfaces is reinforced through "two main pathways: language-as-interface, and non-language interface elements," and that prior research and interventions have concentrated almost entirely on the first while neglecting the second. It states that "beyond language, LLMs are crucially anthropomorphized through chat-based interfaces, which mimic messaging and social media platforms, that include turn taking and artificial thinking delays" — so that "we chat with LLMs the same way we text our friends." As an example, it points to ChatGPT's greeting, "Where should we begin?," as the interface framing itself as "equal conversation partners" from the first turn, with differences from a human interlocutor disclosed, if at all, only "through disclaimers in small font."

The argument draws on the media-theory claim that "the medium is the message," which the paper extends to say that in LLMs "the metaphor has become the message: the metaphor is the medium, and LLM interfaces instantiate the metaphor." Because chat interfaces replicate the surface conventions of texting a person, the interface carries anthropomorphic meaning independent of what the model's language says. The paper compounds this with "Jakob's Law," noting that a user's expectations are shaped by every other system they have used — so the near-universality of chat-style LLM interfaces reinforces the anthropomorphic metaphor across "the ecosystem of LLMs," not just within any single product.

This complicates mitigation strategies that focus only on hedging or disclaiming in a model's text output, since the claim here is that the anthropomorphic cue is structural — built into turn-taking and pacing — and persists regardless of what the language itself says. It sits alongside How do chatbots enable distributed delusion differently than passive tools?, which locates LLMs' seductive power in their intersubjective conversational stance; this paper identifies a parallel, interface-level cause that operates even before any stance-taking occurs. The chat-interface conventions named here are also a concrete instance of the designer-built anthropomimesis described in Who bears responsibility when AI seems human-like?. It is the mechanism-level counterpart to Can interface design reduce harmful anthropomorphism of language models?, which proposes disrupting exactly these conventions through materiality.

The claim is presented as design analysis within a provocation paper, not as an empirical measurement; the excerpt reports no study isolating interface-level anthropomorphism from language-level anthropomorphism, nor any quantification of how much each contributes to user beliefs about an LLM. The paper does note, citing separate prior work, that providing multiple responses can increase friction and "encourage critical engagement and reduce overreliance," but that is reported as a finding of the cited study, not of the present paper. Whether removing chat-style turn-taking and pacing would measurably change user trust or overreliance remains untested.