A conservative-slanted AI summary only convinced people who were already conservative — why didn't it change anyone else's mind?
Why does conservative framing persuade only readers already conservative?
This explores why, in an experiment where AI wrote historical summaries with a deliberately conservative slant, the slant moved conservative readers but left everyone else unmoved, and what that says about how framing and persuasion work.
This explores why a conservative-framed AI summary moved only readers who already leaned conservative, while the AI's default summaries nudged nearly everyone to the left. The finding comes from a preregistered experiment with almost 2,000 people. Default GPT-4o history summaries moved readers further left than the matching Wikipedia text did. Summaries explicitly prompted to sound conservative persuaded only conservatives Do AI historical summaries shift reader opinions without explicit prompting?. The corpus doesn't hold a paper that directly tests why this happens. It does hold several pieces that, taken together, suggest an explanation.
The first piece is that the reader may matter more than the text. A study of debate data found that knowing voters' political and religious leanings predicted who won better than any feature of the arguments themselves. Many apparent 'this wording persuades' effects disappeared once researchers accounted for who was in the audience Does what readers believe matter more than what debaters say?. Related work on interpretation shows that people from different social positions read the same sentence in different ways. That disagreement is real information, not noise Why do readers interpret the same sentence so differently?. So a conservative frame may not even register as the same message for a liberal reader as it does for a conservative one.
The second piece is less obvious: framing may persuade best when nobody notices it. Presuppositions persuade more than direct assertions because they slip claims in as background that is already accepted, so readers don't stop to evaluate them Why are presuppositions more persuasive than direct assertions?. Apply that to the experiment. The AI's default liberal tilt worked like a presupposition, unannounced and taken as the neutral baseline, so it reached everyone. A prompted conservative version stands out against that baseline. Readers who don't share its slant can spot it and discount it, while readers who already agree simply find it confirms what they think. That reading is an inference from these notes, not something the study measured. Still, it would explain why the asymmetry runs only one way.
This matters more because readers don't yet approach AI text with the built-in skepticism they bring to advertising How do we learn to read AI-generated text critically?. Default framing therefore gets through without resistance. And since models' political leanings can be measured, and vary a lot from one model to another Can we measure how deeply models represent political ideology?, the 'neutral' baseline is really a design choice that differs by model. The takeaway you may not have expected: the most persuasive slant in AI text isn't the one someone asked for. It's the one already built in, which readers never notice.
Sources 6 notes
A preregistered experiment (N=1,912) found AI summaries moved readers leftward by default relative to Wikipedia, and explicit conservative framing only persuaded already-conservative readers. The asymmetry suggests AI models embed liberal framing by default, requiring explicit prompting to produce conservative versions.
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.
Interpretation Modeling research shows that disagreement on socially embedded sentences reflects valid differences in reader perspective, not annotation failure. Structured human disagreement in NLI benchmarks confirms that interpretation distributions carry meaningful information.
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.
Every established discourse source carries an interpretive posture that filters how publics receive it. AI-generated text arrived too recently and shifts too quickly to anchor such a posture, allowing it to spread without the protective skepticism we automatically apply to interested speech.
Show all 6 sources
SAE analysis shows models vary dramatically in political feature count (up to 7.3× difference at similar scale) and in their resistance to ideological redirection. Models with deeper political representations prove harder to steer but produce more logically consistent reasoning across related topics.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Persona-Assigned Large Language Models Exhibit Human-Like Motivated Reasoning
- How latent and prompting biases in AI-generated historical narratives influence opinions
- Exploring the Role of Prior Beliefs for Argument Persuasion
- Beyond the Surface: Probing the Ideological Depth of Large Language Models
- Measuring and Mitigating Persona Distortions from AI Writing Assistance
- Computational structuralism: Toward a formal theory of meaning in the age of digital intelligence
- Can LLMs Ground when they (Don't) Know: A Study on Direct and Loaded Political Questions
- Presuppositions are more persuasive than assertions if addressees accommodate them: Experimental evidence for philosophical reasoning