Do embodied agents outperform chatbots because of physical presence alone?
This explores whether the edge embodied agents (like therapy robots) have over text chatbots comes from physical embodiment itself, or from something else the corpus identifies underneath the body.
This reads the question as: when a robot beats a chatbot, is the body doing the work? The corpus's answer is a fairly clean no — physical presence is a carrier for the active ingredient, not the ingredient itself. The sharpest evidence is a study where robots and even paper worksheets reduced students' psychological distress while a chatbot running the *identical* language model did not Why do robots outperform chatbots in therapy despite identical language models?. Same words, different medium — so the difference can't be language capability. But notice the worksheets also worked. A worksheet has no physical 'presence' in the social sense, which already tells you the lever is structure and social framing, not embodiment as such.
What that lever actually is gets named elsewhere as *conversational presence* and judgment-free listening. One note argues ELIZA — a 1960s script with no body and no real understanding — matches modern chatbots on symptom reduction, that RLHF training can actually *degrade* emotional attunement, and that embodied robots beat text bots with the same model Is conversational presence more therapeutic than clinical technique?. Bundle those three together and 'physical presence alone' collapses: if a disembodied script can hold its own, the body is amplifying an effect, not creating it.
The cleanest disproof of 'embodiment wins' comes from the opposite direction. Chatbots — maximally disembodied — turn out to be *superior* partners for intimate disclosure precisely because they carry no social judgment, and the therapeutic benefit comes from the user's own processing while disclosing, not from the agent at all Do chatbots help people disclose more intimate secrets?. So embodiment helps in some contexts (structured therapeutic ritual) and *hurts* in others (vulnerable self-disclosure, where a felt social presence reintroduces the fear of being judged). That's the tell that presence is contextual, not a flat advantage.
Lateral support comes from work on non-embodied media that still beat plain chat: dynamically generated task-specific interfaces outperform text chat in over 70% of cases by reducing cognitive load through structure Do generated interfaces outperform text-based chat for most tasks?. Again — structure and format win, no body required. And how users judge any agent is dominated by perceived competence (about half the variance), with human-likeness second How do users mentally model dialogue agent partners?, so even the 'presence' that does matter is filtered through whether the thing seems capable.
The thing you didn't know you wanted to know: a lot of measured 'embodiment advantage' may be borrowed from novelty. Relationship effects with chatbots decay predictably as novelty wears off, and single-session studies don't extrapolate to the long run Do chatbot relationships lose their appeal as novelty wears off?. The 15-day robot study is short — so part of what looks like a body advantage could be the freshness of a robot in the room, which is exactly the variable that fades. If you want to go deeper, the honest design question isn't 'body or no body' but 'which ingredient — structure, judgment-free framing, perceived competence — am I actually buying, and does the medium deliver it past week one?'
Sources 6 notes
A 15-day study with 38 students found that robots and worksheets significantly reduced psychological distress while a chatbot using the same LLM did not. The active ingredient was the medium—social presence and structured format—not language capability.
ELIZA matches modern chatbots on symptom reduction, RLHF training degrades emotional attunement, and embodied robots outperform text-based ones with identical language models. The active ingredient is judgment-free listening, not therapeutic framework.
The absence of social judgment in chatbot interactions removes barriers to self-disclosure that normally constrain conversation with humans. The therapeutic benefit derives from the user's own cognitive processing during disclosure, not from the chatbot's understanding.
Research shows users strongly prefer LLM-generated interactive interfaces—dashboards, tools, animations—over text blocks, especially for structured and information-dense tasks. Structured representation and iterative refinement reduce cognitive load.
The Partner Modelling Questionnaire reveals that perceived competence dominates user impressions (49% of variance), followed by human-likeness (32%) and communicative flexibility (19%). This three-factor structure reflects how people evaluate dialogue partners against both functional and social standards.
Longitudinal studies with Mitsuku show that social processes driving relationship formation decline as novelty wears off. Single-session study findings cannot be reliably extrapolated to medium- or long-term chatbot design.