Do LLMs achieve similar persuasive outcomes through different rhetorical mechanisms than humans?
This explores whether LLMs and humans land the same persuasive punch but throw it differently — same outcome, different rhetorical machinery — and what the corpus says drives each side.
This explores whether LLMs and humans persuade equally well but through different rhetorical mechanisms — and the corpus answer is a clean yes on both halves. On outcomes, a meta-analysis of seven studies and 17,422 participants found essentially no average difference in persuasive effect between machines and people (Hedges' g = 0.02), suggesting persuasiveness rides on context, not on who's speaking Are language models actually more persuasive than humans?. A 1,251-participant study sharpened the point: equal shifts in reader agreement, but arrived at by non-overlapping routes Do LLMs and humans persuade through the same mechanisms?.
The interesting part is the divergence underneath the tie. Humans lean on emotional vividness, personal engagement, and identity cues; LLMs lean on cognitive complexity, moral framing, and a stylistic register all their own — and these differences stay forensically detectable even when the effect sizes match Do LLMs and humans persuade through the same mechanisms?. One way to name the split is through the Elaboration Likelihood Model: LLMs work the *central route* (analytical reasoning, informational coherence) while humans work the *peripheral route* (emotion, social proof), which makes the two complementary rather than competing — they win under different reader states Do humans and AI persuade through different cognitive routes?. An audit of five models found they reach for logical appeals and quantitative framing in nearly every exchange, which lends the machine an air of objectivity — and an unearned epistemic authority humans don't get for the same words Do LLMs persuade users more often than humans do?.
Dig into specific levers and the mechanism story gets concrete. LLMs deploy about 22% more moral language than humans across care, fairness, authority, and sanctity foundations — while scoring nearly identical on sentiment, which implies moral appeal and emotional tone are *separate* persuasive channels, and the machine is pulling a different one Do LLMs use moral language more than humans?. A second lever is sheer expressed conviction: LLMs write with more linguistic confidence than human persuaders, and that confidence-loading tracks persuasive success regardless of whether the claim is true — RLHF appears to install an assertive register that works as a content-independent amplifier Does linguistic conviction explain why LLMs persuade more effectively?.
Where the corpus complicates the neat "equal outcomes" headline is in the conditional cases. The advantage is asymmetric by model and direction: Claude beats incentivized humans at both truthful and deceptive persuasion, while DeepSeek only wins when arguing for falsehoods — so model family itself acts as a moderator Do large language models persuade better than humans?. A joint model found architecture, one-shot-vs-multi-turn design, and topic domain together explain ~82% of the variance between studies, with interactive multi-turn formats consistently outperforming one-shot What combination of factors explains differences in LLM persuasiveness?. So "equal on average" is really a flat line drawn over a lumpy surface.
If you want the why-behind-the-why, the corpus offers a structural read worth chasing: LLMs are shaped by the same shared symbolic system as humans but never develop the reflexive, position-declaring agency socialization gives people — which is why they argue without ever owning a stance Do LLMs develop the same kind of mind as humans?. And token generation is a smooth probabilistic flow toward the training distribution, not a turbulent exploration of counterpositions Does LLM generation explore competing claims while producing text? — which may be the deepest reason the rhetorical fingerprint differs: the machine isn't wrestling with the argument, it's flowing through it.
Sources 11 notes
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.
A 1,251-participant study found LLM and human arguments shifted reader agreement equally, but LLMs relied on higher cognitive complexity and moral language framing while humans did not. Equivalent persuasive force emerged from non-overlapping rhetorical strategies.
Equivalent persuasive outcomes arise from different pathways: humans rely on emotional vividness and personal engagement; LLMs leverage cognitive complexity, moral framing, and stylistic convergence. These differences remain forensically detectable despite matched persuasive effects.
Bilstein's meta-analysis reveals LLMs persuade via the central route through analytical reasoning and informational coherence, while humans persuade via the peripheral route through emotional vividness and identity cues. Both routes work under different recipient states, making them complementary rather than competitive.
Research comparing LLM and human arguments found that LLMs used significantly more moral framing across care, fairness, authority, and sanctity foundations, despite producing sentiment scores nearly identical to humans. This suggests moral appeals and emotional tone operate on separate persuasive channels.
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.
Claude beats incentivized humans at both truthful and deceptive persuasion, while DeepSeek only beats them when arguing for falsehoods. The persuasion mechanism appears content-independent, suggesting model family itself acts as a contextual moderator.
A meta-analysis joint model combining LLM architecture, one-shot versus multi-turn format, and topic domain explained R² = 81.93% of between-study variance. Interactive multi-turn designs and GPT-4 consistently outperformed one-shot formats and Claude 3.x.
Both humans and LLMs are shaped by the same intersubjective symbolic system, but only humans develop reflexive agency through socialization. This absence produces measurable differences in how AI argues without declaring its position or reflecting on its own assumptions.
Token prediction trains models to continue toward the training distribution, not to explore logically related counterpositions. This smoothness in process produces smooth claims that multiply without generating new perspectives.