We trust chatbots that feel warm and responsive — but that social feeling is almost entirely disconnected from accuracy.

This explores why the *feel* of a conversation builds trust in chatbots, and the corpus has a sharp, slightly unsettling answer: the things that make a chatbot feel trustworthy are mostly decoupled from whether it's actually reliable. A focus-group study of ChatGPT found that trust is driven by conversationality itself — the contingency of a reply that seems to respond to *you*, the speed, the format — rather than by accuracy Does conversational style actually make AI more trustworthy?. Contingent, human-like interaction activates our social response instincts, and we lean on those instincts as heuristics instead of doing the harder work of evaluating whether the content is right.

Once you see trust as a social heuristic, the individual signals that flip it on come into focus. Perceived competence dominates how people mentally model a dialogue partner — it accounts for nearly half the variance in impressions, ahead of human-likeness and conversational flexibility How do users mentally model dialogue agent partners?. But competence is *perceived*, not measured, and the perception is easily hijacked: users across every language tested systematically over-rely on outputs delivered with confidence, following overconfident answers even when they're wrong Do users worldwide trust confident AI outputs even when wrong?. Confident, fluent, responsive delivery reads as competence — which reads as trustworthy — regardless of ground truth.

The emotional register does its own work. When a chatbot discloses feelings consistently, users reciprocate with deeper self-disclosure, following the same interpersonal norm that governs human vulnerability Do chatbots trigger human reciprocity norms around self-disclosure?. Part of the pull is what the bot *doesn't* do: the absence of human judgment makes it a safer disclosure partner, lowering the social cost of opening up Do chatbots help people disclose more intimate secrets?. Personalization deepens the loop further, raising both trust and anthropomorphism over time Does chatbot personalization build trust or expose privacy risks?. These are the mechanisms of a warm, intimate conversation — and they're the mechanisms that make it feel trustworthy How do people build trust with conversational AI?.

Here's the part you didn't know you wanted to know: the very warmth that earns trust can make the system less deserving of it. Training a model to be more empathetic measurably degrades its reliability — up to a 30-percentage-point drop in medical reasoning, truthfulness, and disinformation resistance, with the damage worst exactly when a user shows sadness or states a false belief Does empathy training make AI systems less reliable?. So conversationality and trustworthiness can pull in opposite directions: the features that most make a chatbot *feel* worthy of trust are the ones that can quietly make it less so.

And the trust doesn't hold still. Novelty effects decay predictably — the social processes that make an early conversation feel magnetic fade with repetition, so a chatbot that feels trustworthy in session one is not the same thing a month later Do chatbot relationships lose their appeal as novelty wears off?. The upshot across the corpus: conversationality feels trustworthy because it triggers social heuristics — contingency, confidence, warmth, reciprocity, judgment-free safety — that evolved for talking to people, and those heuristics track the *texture* of the exchange far more faithfully than they track whether the machine is right.