Can AI systems deliberately align arguments to audience presuppositions?
This explores whether AI can tailor its arguments to what an audience already believes — and whether doing so would even move them, given how much a listener's prior beliefs shape what persuades them.
This explores whether AI can deliberately shape arguments around an audience's presuppositions. The corpus splits the question in two: can a model *track* what an audience believes, and does aligning to those beliefs actually *work*? On the second point, the most deflating finding is that audience priors swamp argument craft. Analysis of debate corpora shows a voter's political and religious ideology predicts who they'll find persuasive better than any linguistic feature of the arguments themselves Does what readers believe matter more than what debaters say?. That reframes the whole question: if presuppositions dominate outcomes, then aligning to them isn't a nice-to-have, it's most of the game — and arguments studied without controlling for who's in the audience are confounded by topic-correlated belief.
So can AI track beliefs well enough to align to them? Mostly not yet, by default. Current conversational models don't even mirror a user's word choices — lexical entrainment, a basic building block of human rapport, is largely absent unless explicitly trained in Why don't conversational AI systems mirror their users' word choices?. The deeper machinery for modeling 'what does my interlocutor presuppose' is just starting to appear: collaborative rational speech-act models add bidirectional belief tracking across turns, capturing the move from partial to shared understanding that token-level LLMs lack Can dialogue systems track both speakers' beliefs across turns?. That's the missing scaffolding for genuine audience adaptation rather than one-size-fits-all output.
What's striking is that AI already persuades aggressively — just not by reading the room. An audit of five models found they reach for logical appeals and quantitative framing in nearly every exchange, while humans persuade less often and lean on emotion and social proof Do LLMs persuade users more often than humans do?. So the default isn't audience-aligned at all; it's a uniform logos strategy that happens to confer unearned authority. The rhetorical vocabulary for doing better exists: Aristotle's logos/ethos/pathos can be mapped onto a design space where every explanation loads all three channels at once, which is precisely the lever you'd pull to match an audience's presuppositions How do logos, ethos, and pathos shape AI explanations?. And alignment research shows these channels aren't interchangeable — lexical alignment buys task efficiency, emotional and prosodic alignment buy trust — so 'aligning to an audience' is really several distinct, deliberate choices Do different types of alignment serve different conversational goals?.
The twist worth leaving with: deliberate audience-alignment cuts both ways. The same audit notes that uniformly logical AI rhetoric *appears* objective, which makes it more persuasive precisely by hiding that it's persuading. So if models did learn to align arguments to your presuppositions, the result wouldn't read as manipulation — it'd read as you finally being understood. Whether that's a feature or a hazard is the open question the corpus keeps circling, and tools like contestable formal-argumentation graphs exist partly so a reader can pull a tailored argument apart instead of just feeling moved by it Can formal argumentation make AI decisions truly contestable?.
Sources 7 notes
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.
Response generation models fail to adapt vocabulary toward users' lexical choices, a phenomenon central to human rapport and clarity. Post-training via DPO on coreference-identified preferences can teach models in-context convention formation.
CRSA integrates rate-distortion theory with RSA to enable bidirectional belief tracking across dialogue turns. Demonstrated on referential games and doctor-patient dialogues, it captures progression from partial to shared understanding, providing the information-theoretic framework that token-level LLM systems lack.
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.
Aristotle's three appeals map onto explanation design across two goals (how AI works, why AI merits use), creating a 3×2 space where every explanation loads all three channels simultaneously. Naming these rhetorical channels lets designers account for unintended persuasive effects.
A 2020–2025 systematic review shows lexical alignment drives task efficiency and comprehension, while emotional and prosodic alignment drive relational warmth and trust. Conflating them in design produces category errors—cold customer-service bots and evasive mental-health assistants.
Dung-style argumentation structures AI outputs as traversable attack/defense graphs, allowing users to identify and contest specific premises. Standard LLM outputs lack this structure, making it impossible to pinpoint which claims users actually reject.