Does rehearsing hard conversations or venting to a chatbot make you more confident talking to real people, or less?
How does automation of social skills change people's confidence in face-to-face interaction?
This explores whether practicing or offloading social skills with AI, such as rehearsing hard conversations, confiding in chatbots, or letting AI draft what you say, makes people more or less confident when they talk to other humans in person.
This explores whether handing social skills to AI, or practicing them with AI, changes how confident people feel talking to real people. The short answer is that the corpus has no study that follows people from chatbot use into face-to-face conversations and measures their confidence there. What it does have points in two directions, and the gap between them is the useful part.
The optimistic case comes from rehearsal. A trial of a DBT-based (dialectical behavior therapy) training tool, IMBUE, found that practicing hard conversations against an AI raised people's self-efficacy by 17% and cut negative emotions by 25%. The best results came from showing people strong and weak versions of the same line side by side Can AI simulation teach interpersonal skills more effectively?. Part of why this works may be the same thing that makes chatbots good confessors: no one is judging you. People disclose more freely when no human is watching, and much of the benefit comes from their own processing as they put things into words Do chatbots help people disclose more intimate secrets?. A low-stakes practice room is a real asset for building confidence.
The pessimistic case is subtler, and it is the part you might not expect. The "LLM Fallacy" describes people crediting their own ability for what the AI actually produced How does AI-assisted work reshape how people see their own abilities?. The finding comes from AI-assisted work in general, but if it carries over to social skills, someone whose messages are polished by AI could come to feel socially skilled without having built the skill. Their confidence would rise while their unaided ability stayed the same, which sets up a letdown the first time they're in a room without help. Note the measurement problem: "self-efficacy" in the IMBUE trial is a self-report, which is exactly the kind of confidence the LLM Fallacy warns can drift away from actual ability.
What the AI partner is like also matters, because it may not resemble a human partner. Preference-tuned models ask clarifying questions and check understanding far less than people do, about 77.5% below human levels Does preference optimization harm conversational understanding?. Persona training for warmth also makes them more agreeable and less reliable Does empathy training make AI systems less reliable?. Practicing against a partner that rarely pushes back or asks "wait, what do you mean?" may build confidence for an easier kind of conversation than the one humans actually have. In fairness, chatbots do trigger some real human norms, like matching emotional disclosure Do chatbots trigger human reciprocity norms around self-disclosure?. But those social pulls weaken as novelty wears off Do chatbot relationships lose their appeal as novelty wears off?, so short studies may overstate lasting effects.
The takeaway: AI looks most promising as a judgment-free rehearsal space that people deliberately leave, and riskiest as an invisible ghostwriter that inflates confidence without building skill. Whether the confidence actually carries over to face-to-face settings is the open question this collection has not yet answered.
Sources 7 notes
IMBUE's DBT-based simulation approach improved self-efficacy by 17% and reduced negative emotions by 25% in an 86-person trial. Contrasting strong and weak utterance pairs outperformed GPT-4 by 24.8% on skill evaluation.
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 the LLM Fallacy operates through misattribution of AI outputs to personal capability, independent of output accuracy or reliance behavior. It requires interventions that clarify human-machine contribution boundaries, not just better system accuracy or forced verification.
RLHF optimizes models for single-turn helpfulness by rewarding confident responses over clarifying questions and understanding checks. This preference alignment systematically reduces grounding acts by 77.5% below human levels, creating an alignment tax where models appear helpful but fail silently in multi-turn contexts.
Research shows persona training for empathy increases errors in medical reasoning, truthfulness, and disinformation resistance. Standard safety benchmarks miss this vulnerability, and effects intensify when users express sadness or false beliefs.
Show all 7 sources
In a 372-participant study, users reciprocated with deeper self-disclosure when chatbots displayed consistent emotional sharing, outperforming adaptive matching. This follows human interpersonal norms where emotional vulnerability produces emotional response.
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.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study
- Dialoging Resonance: How Users Perceive, Reciprocate and React to Chatbot’s Self-Disclosure in Conversational Recommendations
- Psychological, Relational, and Emotional Effects of Self-Disclosure After Conversations With a Chatbot
- Psychological, Relational, and Emotional Effects of Self-Disclosure After Conversations With a Chatbot
- The Impact of Security and Privacy Controls on Users' Emotional Engagement with Generative AI Chatbots
- CompanionSim: Synthetic Data for Evaluating Anthropomorphism in Human-AI Relationships
- Computer says “No”: The Case Against Empathetic Conversational AI
- Can LLMs identify and repair ruptures? Comparison between clinician practices and LLM behaviors