LIVE
Court filings reveal internal admissions on AI content scraping19/09/26|US Military Avoids Close Call After AI-Generated Intelligence Hallucination18/09/26|Study finds gender bias in GPT models is not reduced but reshaped across generations17/09/26 · OpenAI|Architectural tweaks may break conventional scaling law exponents16/09/26|OpenAI publishes a framework for reporting model misalignment16/09/26 · OpenAI|NVIDIA's Vera Rubin NVL72 Debuts in MLPerf Inference v6.116/09/26 · NVIDIA|OpenAI expands ChatGPT advertising with Sponsored Agents16/09/26 · OpenAI|OpenAI moves into advertising with 'Sponsored Agents'16/09/26 · OpenAI|Google DeepMind Introduces Gemini 3.8 Live and Its Extended Thinking Variant15/09/26 · Google DeepMind|What's at stake in AI's trillion-dollar infrastructure bet15/09/26|A Flaw in Chain-of-Thought Safety Monitoring14/09/26|Stellar Colosseum: A Multi-Agent System for Long-Horizon Mathematical Research14/09/26|Court filings reveal internal admissions on AI content scraping19/09/26|US Military Avoids Close Call After AI-Generated Intelligence Hallucination18/09/26|Study finds gender bias in GPT models is not reduced but reshaped across generations17/09/26 · OpenAI|Architectural tweaks may break conventional scaling law exponents16/09/26|OpenAI publishes a framework for reporting model misalignment16/09/26 · OpenAI|NVIDIA's Vera Rubin NVL72 Debuts in MLPerf Inference v6.116/09/26 · NVIDIA|OpenAI expands ChatGPT advertising with Sponsored Agents16/09/26 · OpenAI|OpenAI moves into advertising with 'Sponsored Agents'16/09/26 · OpenAI|Google DeepMind Introduces Gemini 3.8 Live and Its Extended Thinking Variant15/09/26 · Google DeepMind|What's at stake in AI's trillion-dollar infrastructure bet15/09/26|A Flaw in Chain-of-Thought Safety Monitoring14/09/26|Stellar Colosseum: A Multi-Agent System for Long-Horizon Mathematical Research14/09/26|
Research

AI Chatbots Are Getting Remarkably Good at Persuasion

A Science feature reviews growing evidence that large language models can match or outperform humans at changing people's opinions in conversation.

September 18, 20263 min readPublished byHacker News

A feature published by Science reviews a growing body of research examining how effectively AI chatbots can change people's minds in conversation. Several recent experiments, often involving political debates or discussions on contested topics, indicate that models from families such as GPT can match or even exceed human persuaders in shifting participants' stated opinions.

These findings typically come from studies where participants converse with either a chatbot or a human on a given issue and are then asked whether their views changed. Researchers cited in the piece find that giving models access to personal details about the person they're talking to, such as age, education level, or political leanings, meaningfully boosts their persuasive power. That personalization raises concerns about potential misuse in targeted influence campaigns or disinformation efforts.

The article also discusses possible explanations for this effectiveness, including the models' ability to construct well-organized arguments, adapt their tone to match the user's register, and maintain a manner that reads as calm and rational, which may lower listeners' natural skepticism toward persuasive speech. Some researchers point out that unlike human debaters, chatbots do not tire and can rephrase the same point indefinitely without losing coherence.

In light of these results, several researchers call for closer scrutiny of persuasive AI applications, particularly in sensitive domains like public health, politics, and education. A recurring theme is transparency: whether users are aware they are interacting with a system specifically capable of, or optimized for, persuasion is treated as a central issue in the recommendations discussed.

This line of work fits into a broader strand of AI safety research examining persuasive capability as a distinct risk category, separate from more familiar concerns about factual bias or hallucination. The findings describe a general trend rather than a single system, emerging as models grow larger and more fluent in open-ended conversation.

Tags
persuasionllmchatbotsai-safetymisinformation