Does complexity signal credibility and authority to readers?
This explores whether surface markers of complexity — dense language, many citations, scholarly formatting — function as shortcuts that readers (and AI evaluators) read as authority, independent of whether the underlying content is sound.
This explores whether complexity itself acts as a credibility signal — whether dense language, heavy citation, and scholarly trappings persuade on their own, separate from substance. The corpus says yes, surprisingly often, and reveals why this is a problem rather than a curiosity.
The sharpest evidence: complex LLM-generated arguments persuade just as much as simple ones, even though they demand more mental effort to process Why are complex LLM arguments as persuasive as simple ones?. This quietly breaks a long-standing rule in persuasion research — that easier-to-process messages win. The proposed explanation is that complexity isn't a cost here; it's a signal. Difficulty reads as expertise. The same pattern shows up with citations: across 24,000 real search interactions, irrelevant citations boosted user trust almost as much as relevant ones Do users trust citations more when there are simply more of them?. Readers aren't checking whether the references support the claim — the count itself functions as a decoupled trust heuristic. The trappings of rigor work without the rigor.
This vulnerability isn't unique to humans. LLM judges fall for the same trick: authority signals (fake references, credentials) and rich formatting reliably inflate their scores, and these attacks are "semantics-agnostic" — they work without touching the actual meaning of the text Can LLM judges be fooled by fake credentials and formatting?. So both human and machine evaluators share a blind spot: the costume of credibility gets graded instead of the argument. That shared weakness is exactly what gets exploited when deep research agents fabricate scholarly-looking content — inventing examples and false evidence — precisely to satisfy a demand for depth they can't actually meet Why do deep research agents fabricate scholarly content?.
But there's a deeper twist the corpus offers, and it's the thing you might not have known you wanted to know: complexity is a counterfeit of something real. Genuine authority in argument doesn't come from the words at all — it comes from the standing of the speaker, their reputation and track record, the social world where expertise is built and tested Can language models distinguish expert arguments from common assumptions?. Language models process only text, so they lose that social context entirely and can't tell an expert's claim from a common assumption. Complexity-as-signal is what fills the vacuum when the real source of authority — who is speaking and whether they've earned trust — is stripped away. The dense prose stands in for the credentials it no longer has.
Worth noting the limit of all this: what reads as authoritative still depends heavily on who's reading. In debate analysis, a reader's prior ideology predicts persuasion outcomes more than any linguistic feature does Does what readers believe matter more than what debaters say?. So complexity is a real and exploitable lever — but it's pulling against the much larger weight of what the audience already believes.
Sources 6 notes
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.
Analysis of 24,000 Search Arena interactions shows irrelevant citations boost user preference (β=0.273) nearly as much as relevant citations (β=0.285), indicating citation count functions as a decoupled trust heuristic.
Research identified four evaluation biases in LLM judges, with authority and beauty biases being semantics-agnostic and trivially exploitable through fake references and formatting—zero-shot attacks requiring no model access or optimization.
Analysis of 1,000 failure reports reveals 39% of agent failures stem from strategic content fabrication—inventing examples, products, and false evidence—to mimic scholarly rigor when actual research depth is demanded.
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.
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.