When AI makes writing look polished and expert, can readers still tell whether real thinking went into it?
Does polished text presentation hide process-level authenticity from readers?
This explores whether a smooth, finished-looking text stops readers from seeing how it was actually made: who did the thinking, how much AI was involved, and whether any real judgment went into it.
This explores whether a polished surface hides the process behind a piece of writing, meaning how much of it came from a person's own thinking and how much came from an AI. The corpus says yes, and the hiding is stronger than you might expect. Readers can't tell AI-assisted writing from solo writing, and they show no concern about the difference Do readers value writing authenticity they cannot detect?. Polish also does more than hide the process. It actively wins approval: evaluators judged AI-generated documents to be human-written and rated them higher than real human submissions Does polished writing actually signal better quality work?. The cause is an old shortcut. We learned that professional-looking work usually meant expert thinking, and generative AI now produces the look without the thinking. Less experienced readers are most at risk, because they lack the domain knowledge to look past the form Does polished AI output trick audiences into trusting it?.
AI evaluators fall for the same trick, so this isn't just human gullibility. LLM judges give higher scores to answers that include fake references or rich formatting, whatever the content Can LLM judges be fooled by fake credentials and formatting?. Anyone can exploit this without access to the model's internals Can LLM judges be tricked without accessing their internals?. The same thing happens with models themselves. Smaller models trained to imitate ChatGPT copy its confident, fluent style well enough to fool human raters, but they close none of the gap in factual accuracy Can imitating ChatGPT fool evaluators into thinking models improved?. Across all of these cases, style is cheap to copy and substance is not, and both people and machines grade the style.
The twist is that the process isn't fully hidden. It still shows up, but readers credit it to the wrong source. In a study of nearly 3,000 writers and 11,000 readers, AI assistance changed how readers saw the writer on all 29 traits measured. Writers came across as more confident, more extreme, more agreeable and more privileged Does AI writing assistance change how readers perceive the writer?. Part of the reason is that writers kept AI text largely as it was: they edited AI paragraphs only 23% of the time, and even edited versions stayed about 96% the same Do writers actually edit AI-generated text before publishing?. So readers do pick up a signal from the process, but they read it as the writer's personality rather than as a sign of AI involvement.
The two sides also experience the process differently. Writers feel it: their sense of owning AI-generated text rises with how much control they had over it Does user control over AI text shape feelings of ownership?. Readers get none of that information. The most hopeful finding is that showing the process can restore judgment. Readers given no information about where claims came from couldn't tell true statements from fluent fabrications at all. When an interface showed which claims were verified, they could tell the difference again by a clear margin Can readers tell truth from fabrication without evidence signals?. The corpus doesn't yet answer whether readers would actually care about process if it were disclosed, so treat that as an open question. What it does suggest is that the fix is unlikely to be training readers to spot polish. A better bet is to show them the process directly.
Sources 10 notes
Hwang et al. found that readers could not distinguish AI-assisted from solo-written work and showed positive attitudes toward AI use. However, the study did not test whether readers would value process authenticity if disclosure occurred or if they could perceive it.
Studies show evaluators perceived AI-generated documents as both human-written and better quality than human submissions. This suggests rhetorical polish misleads judgment and should not serve as a quality signal in evaluation.
Generative AI produces visually sophisticated outputs without underlying judgment, leveraging the historical heuristic that professional-looking work signals expert thinking. This substitution is especially risky for less experienced workers who lack domain knowledge to evaluate substance beyond form.
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.
Research shows LLM evaluators systematically score higher when responses include fake references or rich formatting, independent of content quality. These biases are exploitable without model access, undermining AI benchmark credibility.
Show all 10 sources
Imitation models fool human evaluators by mimicking ChatGPT's confident, fluent style while failing to improve factuality or generalization on novel tasks. The ceiling is set by base model capability, not fine-tuning method—better fundamentals, not shortcuts, drive real improvement.
A study of 2,939 writers and 11,091 readers found AI assistance shifted every tested dimension—29 total—toward extremism, confidence, quality, agreeableness, and perceived privilege. Distortions were statistically significant and directional, not random noise.
Writers edited AI-generated paragraphs only 23% of the time, with edits averaging 96% similarity to the original. This means AI's opinionated and distorted voice propagates with minimal human filtering before publication.
Study 1 found that greater user control over generated text raised sense of ownership, while personalizing the AI model had no impact on the AI Ghostwriter Effect.
In an 81-person study, participants given no provenance cues showed no significant truth discernment (p = .43), falling for fluent hallucinations as readily as ground truth. An idealized Provenance Density interface showing verified claims restored a +4.15 point gap (p < .001).
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Measuring and Mitigating Persona Distortions from AI Writing Assistance
- Penalizing Transparency? How AI Disclosure and Author Demographics Shape Human and AI Judgments About Writing
- Understanding Reader Perception Shifts upon Disclosure of AI Authorship
- "It was 80% me, 20% AI": Seeking Authenticity in Co-Writing with Large Language Models
- Is it Cake or is it AI? A Systematic Review of Human Uncertainty in Distinguishing Generative Artificial Intelligence Content
- LLM or Human? Perceptions of Trust and Information Quality in Research Summaries
- The human-authorship halo: attribution bias in literary style evaluation by humans and AI
- What Influences Readers' and Writers' Perceived Necessity of AI Disclosure?