Does Voice Make AI Feel More Human? Interaction Modality, Anthropomorphization, and the Design of AI Tutors

Artificial intelligence systems reach users through two dominant interaction modes: voice interfaces such as Siri and Amazon Alexa, and text interfaces such as ChatGPT. A recurring claim in human-computer interaction research is that voice systems draw stronger anthropomorphic responses than text systems, even when the underlying model and the content of the exchange are equivalent. This paper examines the psychological and communicative mechanisms that would produce such an asymmetry, drawing on the Media Equation, the Computers as Social Actors paradigm, Social Presence Theory, and Epley et al.'s three-factor theory of anthropomorphism. It argues that the paralinguistic cues carried by speech, including prosody, pitch, and pacing, engage social processing that text engages weakly or not at all. The supporting evidence is real but indirect and partly conflicting. Few studies hold content, platform, and conversational structure constant while varying modality alone. One controlled experiment finds that voice shifts anthropomorphism without shifting trust, and one pre-registered comparison finds text rated as more human-like than voice. The paper therefore treats the modality gap as a well-motivated hypothesis rather than a settled effect, and works through what would follow for educational technology if it holds. Four consequences receive sustained attention: the calibration of learner trust, parasocial attachment and dependency, the fit between voice characteristics and pedagogical role, and transparency. It closes with an argument for modality-aware design and a research agenda organised around the controlled comparisons the field currently lacks. This is a conceptual synthesis and position paper, not a systematic review and not primary empirical research. It has not been peer reviewed.