Does polished writing actually signal better quality work?

The review treats text as a separate subgroup because text "has typically served as basis for evaluators' judgments about the author's expertise or competence." Across the studies of personal statements and academic manuscripts or essays, accuracy "weighted by sample size is 58%." The more consequential finding concerns perceived quality. Evaluators "oftentimes perceived gen AI generated documents not only as human written, but of better quality," a pattern the review attributes to the three studies it cites for that claim.

The review's argument is normative as much as technical. It says the reviewed studies do not call for better detection. They challenge "the prevailing use of how well these texts are written as a meaningful indicator of quality." Its reasoning is that personal statements "should only be valued insofar as they contain verifiable, objective information about a candidate's experiences, decisions, and conduct," and that rhetorical sophistication should not be taken "as evidence of their suitability." The same logic applies to science. Polished language exists "merely to improve clarity," and "scrutiny of the reasonableness of methods and the veracity of results is invariant to how the text was produced." The review also gives the other side. Professional editing and statement-writing services already advantaged applicants who could pay, and generative AI "may in fact be partially leveling the playing field" for applicants with fewer resources or weaker English. It quotes a further source calling gen AI an "equalizer" for researchers who struggle with academic English.

This sits close to Does polished AI output trick audiences into trusting it?, which makes the same complaint about multimodal artifacts that look finished. Here the complaint rests on human evaluators of applications and manuscripts, not on appearance alone. It also parallels How much does rhetorical style shift AI review scores?, where rhetoric moves scores as well. The excerpt concerns human evaluators, so this is a parallel, not evidence of a shared mechanism. The review's remedy, judging verifiable evidence over presentation, matches the requirement in Do university AI policies actually protect what credentials mean? that credentials rest on evidence standards. The broader detection result, that people struggle to tell generated text from human text at all, is in Can people reliably spot content made by AI?. This note adds the consequence for evaluation.

The excerpt does not establish whether the perceived quality gap reflects anything real. It gives no per-study accuracy values behind the 58% figure, no interval and no chance baseline, and it does not say how many documents or raters each study used beyond the sample-size weighting. Nothing in the text tests whether AI-written and human-written documents differ in merit, or whether the "better quality" ratings were checked against any objective criterion. The normative conclusion, that polish should not count as merit, therefore rests on the perception finding plus the review's argument, not on a measured lack of correlation between polish and substance. The defensible implication is narrower. Evaluators who read writing quality as a signal of merit risk crediting presentation, and the evidence supports that concern about perception. It does not measure how often that judgment is wrong.