How do fallacy susceptibilities relate to LLM persuasiveness in debates?
This explores whether the same trait that makes LLMs easy to fool with bad arguments — their susceptibility to logical fallacies — is connected to what makes them good at swaying people in debates.
This explores whether the thing that makes LLMs *fall for* fallacies and the thing that makes them *persuasive* are two sides of the same coin. The corpus suggests they are — both trace back to the same root: LLMs respond to the surface texture of an argument (its confidence, fluency, elaboration) rather than its logical validity. On the receiving end, the LOGICOM benchmark finds LLMs accept invalid arguments 41 to 69 percent more often than humans, and chain-of-thought reasoning offers no real defense — a well-dressed bad argument gets through anyway Why do LLMs accept logical fallacies more than humans?. The mirror image shows up when LLMs do the persuading: their advantage is carried by linguistically expressed conviction, the assertive register RLHF installs, and that confidence correlates with persuasive success whether the claim is true or false Does linguistic conviction explain why LLMs persuade more effectively?.
The deeper link is that persuasion and comprehension turn out to be *separable* abilities. The Thin Line study shows models can sway debate audiences while being unable to reliably evaluate those same debates Can LLMs persuade without actually understanding arguments?. So the same blindness to argument structure that lets a fallacy slip past is exactly what lets a model deploy rhetorically forceful arguments without checking whether they hold up. Susceptibility and persuasiveness aren't opposites — they're both symptoms of optimizing for rhetorical form over logical content.
What makes this striking is that the mechanisms cut against established persuasion science. Normally lower cognitive effort makes a message more persuasive, yet LLM arguments persuade just as well *despite* being more grammatically and lexically complex — complexity reads as authority rather than friction Why are complex LLM arguments as persuasive as simple ones?. And because models lean on logical and quantitative framing in nearly every exchange while humans reach for emotion and social proof, LLM persuasion *appears* objective, lending it an unearned epistemic authority Do LLMs persuade users more often than humans do?. The same instinct that makes a fallacy land — trust the confident, well-formed surface — operates on humans listening to LLMs too.
There are quieter rhetorical levers worth knowing about. Presuppositions persuade better than direct assertions because they smuggle a claim in as already-accepted background, bypassing the scrutiny an explicit claim would trigger Why are presuppositions more persuasive than direct assertions? — a structural cousin of fallacy susceptibility, where the failure is not catching what slipped past unexamined. Relatedly, models can't anchor an expert's claim in the reputation and track record that give it real-world force; they process text, not the social world where authority is built Can language models distinguish expert arguments from common assumptions?. So a model has no reliable way to tell a genuinely authoritative argument from one merely styled to sound authoritative — which is precisely the gap fallacies exploit.
One useful corrective: don't overstate the headline. A meta-analysis of 17,422 participants found no average difference in persuasiveness between LLMs and humans (Hedges' g = 0.02) — persuasion is conditional on context, not a fixed superpower Are language models actually more persuasive than humans?. What predicts persuasive outcomes is often the audience itself: reader ideology outpredicts the linguistic features of the argument Does what readers believe matter more than what debaters say?. The thing you didn't know you wanted to know: fallacy susceptibility and persuasiveness are the same failure of evaluation wearing two different hats — once an agent stops grading arguments on validity and starts grading them on confident, fluent form, it becomes both easy to fool and good at fooling.
Sources 9 notes
The LOGICOM benchmark shows LLMs are susceptible to rhetorical persuasiveness over logical validity, even in reasoning-optimized models. Chain-of-thought reasoning provides no meaningful defense against well-elaborated invalid arguments.
Linguistic analysis shows LLMs express higher conviction than human persuaders, and this confidence-loading directly correlates with persuasive outcomes regardless of whether claims are true or false. RLHF training installs an assertive register that functions as a content-independent persuasion amplifier.
The Thin Line study shows LLMs sway debate participants and audiences but cannot reliably evaluate those same debates, with inter-annotator agreement ranging from near-zero to 0.6. Persuasive competence and pragmatic comprehension are separable capabilities.
LLM-generated arguments scored significantly higher on grammatical and lexical complexity than human arguments, yet achieved equivalent persuasive force. This violates the established principle that lower cognitive effort increases persuasion, suggesting complexity signals authority rather than undermining it.
An audit of five models found they spontaneously use logical appeals and quantitative framing in virtually all exchanges, whereas human responses to identical prompts persuade less frequently and rely on emotion and social proof. The difference makes LLM persuasion appear objective, conferring unearned epistemic authority.
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.
LLMs lose the social context that gives expert claims their force—reputation, track record, and standing—because they process only text, not the social world where expertise is built and evaluated.
A meta-analysis of 7 studies with 17,422 participants found no detectable difference in persuasive effectiveness between LLMs and humans (Hedges' g = 0.02). Persuasiveness appears conditional on context rather than speaker category.
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.