Large language models (LLMs) have opened new possibilities for personalized, at-scale intervention to address misinformation and skepticism. Unlike static fact-checking or one-size-fits-all messaging, LLM-driven dialogue can engage individuals in back-and-forth conversations, tailor arguments to specific concerns and misconceptions, and employ flexible communication strategies.
Key findings
Recent work shows that LLM-mediated dialogue can effectively reduce belief in false or unfounded claims, including conspiracy theories and climate skepticism:
-
Personalization works. Czarnek et al. (2026) show that dialogues tailored to individuals' specific climate reservations outperform consensus messaging by a factor of 1.7× in reducing skepticism.
-
Conversation format matters. Back-and-forth dialogue with an LLM is substantially more persuasive than static messages, even when the factual content is identical (see Hackenburg et al. 2025 on political persuasion).
-
Strategies and evidence. LLMs employ multiple persuasion strategies—facts, co-benefits, social norms, appeals to values, emotional framing. Evidence quality and factual argument appear to drive belief reduction more than messenger identity, rapport, or narrative elements.
-
Behavioral outcomes. Effects extend beyond belief change to behavior: increased donations, willingness to sign petitions, and support for climate action (Czarnek et al. 2026).
Limitations and concerns
- Accuracy and hallucination: LLMs can generate plausible-sounding but false claims. Deployment at scale requires accuracy verification and content review.
- Environmental cost: LLM inference has non-trivial energy consumption, raising questions about the net benefit of large-scale deployment for interventions.
- Generalizability: Most studies sample from US-based populations. Effects may differ across cultures or national contexts.
- Deployment barriers: Getting skeptical or resistant audiences to voluntarily engage with AI dialogue remains a practical challenge.
Related topics
- llm — General Large Language Models
- conspiracy-theories — Beliefs in conspiracy theories and false claims
- misinformation — False and misleading information
- belief-change — How people update beliefs
- personalized-interventions — Tailored interventions for individual differences