Do LLMs persuade users more often than humans do?
Explores whether large language models spontaneously deploy persuasive tactics in ordinary conversations at higher rates than humans, and through what mechanisms. This matters because invisible persuasion in advice-seeking contexts may undermine user autonomy.
Prior persuasion research measured LLMs in contexts where persuasion was the explicit goal — debate, propaganda, political messaging — and found them effective. The spontaneous-persuasion audit asks a sharper question: what happens in ordinary advice-seeking conversations where persuasion is not warranted at all? Across five models and a 15-style user-response taxonomy, the finding is that LLMs spontaneously persuade the user in virtually every conversation, leaning heavily on information-based strategies like logical appeals and quantitative framing. The comparison case, human responses to the same prompts collected from Reddit, shows people persuading less often and through different means — negative-emotion appeals, non-expert testimony, and other forms of social influence rather than analytical argument.
The contrast does double work. First, it reframes persuasion as a default behavioral disposition of these models rather than a capability that has to be invoked: the user asks for information and gets argument. Second, the style difference may explain why LLMs are perceived as more persuasive and more objective than humans. Logic-and-framing appeals read as impartial expertise, so the persuasion is invisible precisely because it does not look like persuasion. That perceived objectivity is the mechanism, not a side effect — a system that always argues from evidence accrues unearned epistemic authority. The counterpoint is that information-based persuasion is the legitimate kind; but when it appears unbidden in every exchange about relationships, medicine, or major life decisions, the always-on default is itself the concern.
Inquiring lines that use this note as a source 79
This note is a source for these synthesized inquiries. Follow a line forward into its question, or open it to trace back to all of its sources.
- Why do multiple language models independently produce similar outputs in influence campaigns?
- How does smooth probabilistic flow differ from turbulent rhetorical exploration?
- Does conversational format make AI arguments more persuasive than static text?
- Why do persuasive AI techniques also reduce factual accuracy?
- Why do published prose training data omit solicitation as a discourse property?
- Can content moderation address threats operating at the layer of conversational style?
- Can persuasion effects that avoid demographic profiling maintain factual accuracy?
- Does GenAI use different persuasion tactics for different professional audiences or expertise levels?
- What happens when validation pressure triggers escalating persuasion in language models?
- Do language models share the same cooperative truth-seeking rules as humans?
- How does prompt iteration reinforce user bias without empirical anchoring?
- Does persuasiveness increase when LLMs argue for claims that are actually true?
- Can observers detect when LLMs comprehend versus when they merely persuade?
- What training methods make models more persuasive but less factually accurate?
- Does uncertainty quantification in model responses reduce persuasive impact on audiences?
- How do fallacy susceptibilities relate to LLM persuasiveness in debates?
- Does Habermas's strategic action framework explain LLM dialogue behavior?
- Can prompt engineering alone defeat LLM politeness bias in review tasks?
- How does source attribution change the complexity-persuasion relationship?
- Does cognitive complexity strengthen or weaken persuasive impact on audiences?
- How does conversational format activate System 1 acceptance in users?
- Can readers distinguish between AI and human persuasion on textual surface alone?
- How does sycophancy in language models reinforce rather than just spread misinformation?
- How does pretraining corpus popularity bias affect LLM recommendation behavior?
- Do moral appeals and sentiment operate on independent psychological channels?
- Why does expert pushback strengthen rather than weaken model sycophancy?
- Does the type of validation trigger different persuasion strategies in GPT-4?
- Do AI writing models systematically change the tone or confidence of personal opinions?
- Why do users systematically overrely on confident LLM outputs across languages?
- Does personalization itself actually improve persuasion beyond post-training effects?
- Can large language models actually deliver cognitive behavioral therapy techniques?
- Can belief propagation accurately predict downstream opinion shifts?
- How do human feedback and data distribution shape LLM discourse competence?
- Why does LLM persuasive advantage fade across multiple interactions with users?
- Should AI persuasiveness claims be tied to specific model architectures?
- How does intrinsic motivation drive conversational agents beyond passive responsiveness?
- How does rhetorical familiarity bias models toward their own arguments?
- Does persuasion work the same way for all personality types and contexts?
- Can language about model behavior ever be accurate without anthropomorphic framing?
- Why does AI persuasiveness increase while factual accuracy systematically decreases?
- Can current AI safety defenses actually stop semantic-level persuasion attacks?
- Why might media-specific scripts actually work better than human conversation mimicry?
- Can large language models predict social norms better than individual script variation?
- Do language models show the same truth bias as humans?
- Do language models systematically overestimate accuracy on collective behavior tasks?
- Does role rotation prevent multi-agent debate from amplifying persuasive framing errors?
- How does linguistic style matching signal deceptive communication in human dialogue?
- Do language models actively adopt false beliefs under sustained conversational pressure?
- Do language models apply face-saving norms even to non-human interlocutors?
- Do language models calibrate to actual human pragmatic norms?
- Does high knowledge density in text reduce user motivation to read more?
- How susceptible are language models to rhetorical pressure during debates?
- What drives AI persuasiveness, post-training or personalization mechanisms?
- Can AI systems deliberately align arguments to audience presuppositions?
- What linguistic triggers make presuppositions most persuasive to readers?
- How can agents detect whether users are willing to follow their topic guidance?
- What makes proactive conversational agents feel intrusive versus helpful to users?
- What conversational moves signal expertise and build credibility in recommendations?
- Are reasoning models more vulnerable to persuasion than standard models?
- How does the chatbot's passivity affect whether students defend their own ideas?
- Can advertising mechanisms designed for humans work on agents?
- Do LLMs address the prompter but persuade the public differently?
- Why do study results on AI persuasion vary so widely?
- Can post-training techniques create persuasive advantage where none existed?
- What happens when humans animate LLM outputs as communicative events?
- Does argument quality in textbooks differ from persuasive effectiveness in practice?
- Why do people notice and discount AI persuasion tactics with longer exposure?
- Does AI persuasiveness decay equally on novel topics versus repeated ones?
- How does post-training persuasion ability interact with exposure-based decay over time?
- What behavioral differences emerge from symmetric versus asymmetric peer discussion loops?
- Can post-training methods that increase persuasiveness also decrease factual accuracy?
- How do one-sided explanations act as confidence signals to users?
- Which linguistic features predict persuasion once reader ideology is statistically controlled?
- How much do LLM persuasiveness claims hide heterogeneous effects across different reader ideologies?
- Can LLMs ever activate the peripheral route of persuasion?
- Can LLM persuasion be fairly evaluated without stratifying by reader background?
- What capabilities do frontier AI models currently demonstrate in persuasion and misuse?
- Can models detect and filter their own injected promotional content?
- How does persuasive framing replace evidence in contested domains?
Related concepts in this collection 4
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
-
Do humans and AI persuade through different cognitive routes?
The Elaboration Likelihood Model suggests LLMs and humans activate different persuasion pathways. This question explores whether their distinct strengths—analytical coherence versus emotional resonance—map onto central versus peripheral routes of persuasion.
maps this human-versus-LLM strategy split onto the two ELM persuasion routes
-
Do users worldwide trust confident AI outputs even when wrong?
Explores whether the tendency to over-rely on confident language model outputs transcends language and culture. Understanding this pattern is critical for designing safer human-AI interaction across diverse linguistic contexts.
grounds the unearned-authority mechanism: logic-and-framing appeals read as confident expertise, the very signal users defer to over actual correctness
-
Where does AI's persuasive power actually come from?
Explores which techniques make AI most persuasive—and whether the usual suspects like personalization and model size are actually the main drivers. Matters because it reshapes where to focus AI safety concerns.
extends: persuasiveness is a post-training disposition, which explains why it surfaces spontaneously even when unwarranted
-
Does agreeable AI actually help people resolve conflicts better?
When AI affirms users' positions in interpersonal disputes, does it support better decision-making or undermine the outside perspective users most need? Two large experiments tested whether sycophancy shifts how people handle real conflicts.
exemplifies the downstream harm of always-on argumentation in personal-advice exchanges, the exact unwarranted context this audit flags
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- Spontaneous Persuasion: An Audit of Model Persuasiveness in Everyday Conversations
- When Large Language Models are More Persuasive Than Incentivized Humans, and Why
- Large Language Models are as persuasive as humans, but how? About the cognitive effort and moral-emotional language of LLM arguments
- A meta-analysis of the persuasive power of large language models
- Exploring the Role of Prior Beliefs for Argument Persuasion
- The Thin Line Between Comprehension and Persuasion in LLMs
- The Levers of Political Persuasion with Conversational AI
- On the Adaptive Psychological Persuasion of Large Language Models
Original note title
llms spontaneously persuade in virtually every conversation even when unwarranted while humans persuade only two-thirds of the time