Why are false presuppositions more persuasive than false assertions?
This explores the linguistic mechanism behind why a claim slipped in as a background assumption ('Now that the trial failed...') lands harder than the same claim stated outright ('The trial failed') — and why that asymmetry holds even when the claim is false.
This explores why packaging a false claim as a presupposition — something the sentence treats as already true — beats stating it as a direct assertion. The corpus points to one core mechanism: presuppositions bypass the gate where listeners evaluate claims. An assertion puts content 'on the table' as the main point being made, which invites scrutiny — you can disagree with it because it's what's up for debate. A presupposition smuggles the same content in as settled background, the shared ground the rest of the sentence stands on. Experimental work shows this directly: presuppositions persuade more than assertions specifically for *new* information, because they present a claim 'as already-accepted background' rather than as something asking for your agreement Why are presuppositions more persuasive than direct assertions?.
The deeper reason this works is a process linguists call accommodation: when a sentence assumes something you haven't agreed to, the path of least resistance is to quietly update your mental context to make the sentence coherent, rather than stop and challenge the assumption. You repair the conversation by accepting the premise. This is why presupposed content slips past evaluation — refusing it means interrupting the flow to dispute something that wasn't even presented as the point Do language models miss presuppositions that arise from context?.
What makes the effect strong but uneven is that 'presupposed-ness' isn't a fixed property of certain words — it's gradient, and it depends on whether the content addresses the question currently under discussion. Content that is *not* the at-issue point projects more strongly into the background, where it escapes challenge; the very same trigger word projects less when it happens to be what the conversation is actually about Does projection strength vary by context or by word type?. So the persuasive advantage of a false presupposition is largest exactly when the falsehood is tucked away from the main thread — off to the side, where you're least likely to inspect it.
Here's the part that should unsettle you: this gate-bypass isn't just a human cognitive quirk. Language models inherit it. Even when a model demonstrably *knows* the correct fact, it accommodates false presuppositions at startling rates — false premises drive more acceptance than correct knowledge drives rejection Why do language models accept false assumptions they know are wrong?. Models miss conversationally derived presuppositions especially, because catching them requires tracking what's under discussion rather than pattern-matching trigger words Do language models miss presuppositions that arise from context?. So a false premise embedded in a prompt can steer an AI the same way it steers a person — and the model can't simply 'know better' its way out.
This connects to a broader pattern in the corpus about persuasion that works independently of truth. The same content-blind machinery shows up in linguistic conviction — an assertive, confident register correlates with persuasion regardless of whether claims are true Does linguistic conviction explain why LLMs persuade more effectively? — and in the finding that argument *complexity* signals authority rather than undermining it Why are complex LLM arguments as persuasive as simple ones?. The thread running through all of these: persuasion often rides on *form* — how a claim is framed, registered, or presupposed — rather than on its content surviving examination. False presuppositions are perhaps the cleanest case, because the framing literally removes the claim from the space of things you were invited to examine at all.
Sources 6 notes
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 learn statistical associations between trigger words and inferences, but presuppositions also arise through accommodation—updating context to resolve discourse mismatches. Models miss these because they require tracking questions under discussion, not pattern matching.
Across 19 English expressions, projectivity varies continuously based on whether content addresses the Question Under Discussion. The same presupposition trigger projects more or less depending on context, not on fixed lexical properties.
The FLEX Benchmark shows that models reject false presuppositions at rates far below acceptable levels (GPT-4: 84%, Mistral: 2.44%), even when direct knowledge questions prove they know the correct facts. False presuppositions drive more accommodation than correct knowledge drives rejection.
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.
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.