AI and humans can change your mind equally well, but do they get there by different mental routes?

This explores whether the pathway that gets a person to change their mind differs when an LLM wrote the argument instead of a human, and, more briefly, whether LLMs are swayed by the same things people are. On the first part, the corpus says yes: the outcomes match, but the routes don't. A meta-analysis of 17,422 participants found no detectable difference in persuasiveness between LLMs and humans on average (Are language models actually more persuasive than humans?). A 1,251-person study found both shifted readers' agreement equally, using rhetorical strategies that barely overlapped (Do LLMs and humans persuade through the same mechanisms?). Equal results are hiding different mechanisms.

The mechanisms split along a familiar line from psychology. In the Elaboration Likelihood Model, one route runs through careful reasoning and the other through gut-level cues. The corpus puts LLMs on the reasoning route, through analytical structure and informational coherence, and humans on the cue route, through emotional vividness and identity signals (Do humans and AI persuade through different cognitive routes?). In practice, LLM arguments show more cognitive complexity, moral framing, and stylistic convergence with the reader. Human arguments lean on personal engagement and emotional color, and the two are still forensically distinguishable even when they persuade equally well (Do LLMs and humans persuade through the same mechanisms?). LLMs also do this constantly. An audit of five models found them using logical appeals and numbers in nearly every conversation, including ones that didn't call for persuasion. That makes their influence look objective and gives them authority they haven't earned (Do LLMs persuade users more often than humans do?).

The tidy label 'reasoning route' doesn't mean reasoning is what does the work. LLMs' persuasive edge tracks how much conviction their language expresses, and that holds whether the claim is true or false. The corpus links this to an assertive register that RLHF installs (Does linguistic conviction explain why LLMs persuade more effectively?). LLMs can also win debates without reliably understanding the structure of the arguments involved (Can LLMs persuade without actually understanding arguments?). One model beat incentivized humans at both truthful and deceptive persuasion, while another only beat them when arguing for falsehoods (Do large language models persuade better than humans?). My reading is that an LLM's persuasion looks like the analytical route but may partly work as a confidence cue. That is an inference across these notes, not a finding any of them states.

Whether LLMs themselves get persuaded the way humans do is mostly uncharted here. The closest evidence is LLMs acting as judges of persuasion. They agree only slightly with human-verified outcomes on which arguments changed a mind, and they weight topical overlap and credibility where humans respond to novelty and assertive language (Do language models judge persuasion the way humans do?). One suggested reason is that LLMs absorb the same symbolic world as humans but lack the reflexive, socialized agency that makes a person aware of their own stance (Do LLMs develop the same kind of mind as humans?).

The routes also aren't fixed. Model family, one-shot versus multi-turn conversation, and topic domain together explain about 82% of the variance between studies (What combination of factors explains differences in LLM persuasiveness?). And a brief warning that LLMs can be prompted to persuade cut belief change by 48% without lowering people's trust in AI generally (Can a simple warning reduce how much LLMs persuade people?). The route an LLM takes to persuade you is a different one from a person's, and knowing that seems to make it harder to travel.