When AI helps write something, do readers rate it higher because it looks polished, or because they judge it differently?

This explores whether AI-written text wins ratings on its surface polish, or whether readers and judges apply different standards to it. The corpus says both forces are real, they pull in different directions, and which one wins depends on whether the evaluator knows where the text came from.

Start with presentation. When readers don't know AI was involved, AI-assisted writing comes across as more polished. A study of nearly 3,000 writers and 11,000 readers found AI help moved every one of 29 perceived traits in the same direction: writers seemed more confident, more agreeable, more privileged, and more skilled, but also more extreme in their views Does AI writing assistance change how readers perceive the writer?. That polish reaches readers largely untouched, because writers edited AI paragraphs only 23% of the time, and their edits left the text about 96% the same Do writers actually edit AI-generated text before publishing?. Under the surface, the polish is partly empty. AI prose holds together but avoids taking evaluative positions, which makes it smooth and argumentatively inert Why does AI writing sound generic despite being grammatically correct?. Some readers sense this as a kind of aloofness: the writing never actually reaches for their attention Does AI writing lack the internal appeal to attention that humans use?.

Now judgment. Readers can't reliably tell AI text from human text. Detection hovers around chance across 30 studies Can people reliably spot content made by AI?, even though machines can measure clear differences in vocabulary Can human judges detect measurable differences in AI text?. So most judgment shifts come from labels, not from what's on the page. Give judges identical passages and they rate the one labeled 'human' about 14 points higher. AI judges show the same bias about 2.5 times as strongly, at 34 points Do authorship labels bias how we judge literary quality?. The label can even flip how rules get applied. AI judges forgave a lipogram (a text written without a certain letter) that broke its own rule when told a human wrote it, while human judges became stricter Do authorship labels change how AI judges evaluate rule violations?.

The result is a split. Hide the source and AI text tends to score higher on polish. Reveal it and judgment usually turns against the AI version, most sharply when the judge is itself an AI. One thing you might not expect: using LLMs as judges doesn't remove the bias. It makes the bias bigger. Here the corpus runs out. No study in the collection rates the same AI text with and without a label to measure how much of the score comes from polish and how much from the label. The closest thing to a judge that labels can't sway is structural analysis, which spots AI fiction from storytelling choices like character agency and timeline, not style Can AI stories be detected without analyzing writing style?.