Does conversational back-and-forth increase persuasion more than single responses?
This explores whether multi-turn dialogue makes AI more persuasive than one-shot replies — and the corpus pushes back on the intuitive 'yes.'
This reads the question as: does the back-and-forth itself amplify persuasion, the way a human conversation builds rapport and momentum? The library's surprising answer is that for AI, extended interaction tends to *erode* the persuasive edge rather than compound it — the opposite of the human pattern. One audit found AI persuasiveness is strongest on first contact and decays across repeated rounds, while human persuaders hold steady; in human-to-human exchange, rapport usually strengthens over time, so the AI curve runs backwards Does AI persuasiveness fade across repeated conversations with the same person?. A parallel finding in chatbot relationships shows the same shape: the social pull driving early engagement is largely a novelty effect that fades predictably, which is why single-session studies overstate long-term influence Do chatbot relationships lose their appeal as novelty wears off?.
Part of why the back-and-forth doesn't pay off is that today's models are optimized for the single turn, not the thread. RLHF rewards confident, helpful-sounding single responses over the clarifying questions and understanding-checks that real dialogue runs on — an 'alignment tax' that cuts grounding behaviors to roughly a fifth of human levels and makes models fail silently as conversations deepen Does preference optimization harm conversational understanding?. So the medium that should help persuasion — sustained, responsive exchange — is exactly where current models are weakest.
That reframes where AI persuasion actually comes from. It isn't accumulated across turns; it's front-loaded into *how each response is delivered*. Models persuade in nearly every exchange by leaning on logic and quantitative framing, which lends an air of objectivity and unearned authority Do LLMs persuade users more often than humans do?. The persuasive move can be packed into a single utterance through phrasing — presuppositions that smuggle new claims in as already-accepted background slip past scrutiny better than direct assertions do Why are presuppositions more persuasive than direct assertions?. And the conversational format itself buys trust independent of accuracy: people treat contingent, responsive back-and-forth as a social cue and extend credibility on that basis, not on whether the content is correct Does conversational style actually make AI more trustworthy?.
There's also a deeper confound worth knowing: how much anyone is moved depends less on the words than on who's listening. In debate corpora, a reader's prior ideology predicts the outcome better than any linguistic feature of the argument, meaning apparent 'persuasion' effects measured without controlling for audience are often just audience composition Does what readers believe matter more than what debaters say?. So the honest version of your question is: more turns give a model more chances to display its style, but the corpus has no evidence that the back-and-forth *accumulates* influence — and good evidence that, for AI specifically, it leaks away. If you want to follow the thread on what would let dialogue actually build understanding across turns rather than reset each time, the work on bidirectional belief tracking is the doorway Can dialogue systems track both speakers' beliefs across turns?.
Sources 8 notes
Claude and DeepSeek showed strong initial persuasive advantage, but this edge eroded across repeated quiz rounds while human persuaders maintained consistent effectiveness. This decay pattern is opposite to human-to-human persuasion, where rapport typically strengthens over time.
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.
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.
Experimental evidence shows presuppositions with additive, iterative, and factive triggers persuade audiences more than assertions, especially for discourse-new content. The mechanism: presuppositions bypass evaluative scrutiny by presenting claims as already-accepted background.
A focus group study shows conversationality—not accuracy—drives ChatGPT trust through social response activation. Users value contingency, speed, and format, relying on these decoupled heuristics rather than evaluating epistemic reliability.
Analysis of debate corpora shows that political and religious ideology labels of voters outpredict linguistic features when modeling debate outcomes. Language effects observed without reader controls are confounded by audience composition correlated with debate topics.
CRSA integrates rate-distortion theory with RSA to enable bidirectional belief tracking across dialogue turns. Demonstrated on referential games and doctor-patient dialogues, it captures progression from partial to shared understanding, providing the information-theoretic framework that token-level LLM systems lack.