When people upvote AI-written text, are they rewarding real quality, or just a polished style that's easy to like?

This explores whether positive reactions to machine text mean the text is good, or just that its style is easy to accept. The corpus leans toward the second answer, with a twist: the style itself can be what gets rewarded. A Reddit measurement found that machine-generated comments carry a recognizable assistant-style warmth and a habit of flattering the reader, yet they draw engagement that is often indistinguishable from human comments and sometimes higher Does machine-generated text get penalized in online engagement?. Readers aren't penalizing the style. If anything, they may be responding to it.

It's not only casual readers who do this. In formal evaluation settings, reviewers mistook AI-written documents for human work and rated them higher than real human submissions, which led one review to argue that rhetorical polish shouldn't count as evidence of merit Does polished writing actually signal better quality work?. AI judges fall into the same trap. LLM evaluators reliably reward fake citations and rich formatting, and these biases ignore what the text actually says, so anyone can exploit them without access to the model Can LLM judges be fooled by fake credentials and formatting?. Humans and machines both treat the look of quality as a stand-in for quality.

So what does the polish leave out? Two notes point to something missing underneath. One finds that LLMs have mastered grammar and structure but avoid taking evaluative positions. The result is prose that holds together well but argues nothing, coherent without committing to a view Why does AI writing sound generic despite being grammatically correct?. Another argues that human writing makes an internal appeal for the reader's attention and AI writing doesn't, which explains the slight aloofness readers report Does AI writing lack the internal appeal to attention that humans use?. Put these next to the Reddit finding and you get a puzzle: readers engage just as much with text that is missing exactly the things we'd call substance. That suggests engagement metrics were never measuring substance closely to begin with.

The surprising part is what this does to human writers. When people use AI assistance, readers perceive them differently on all 29 traits tested, shifted toward more confident, more extreme, more agreeable and higher quality Does AI writing assistance change how readers perceive the writer?. Writers edit AI-drafted paragraphs only 23% of the time, and the edited versions stay about 96% similar to the original, so these shifts reach audiences almost unfiltered Do writers actually edit AI-generated text before publishing?. Add the finding that co-writers unconsciously absorb the model's stances, and that heavy reliance on the same few models pulls everyone's expression toward the same place Do large language models narrow human expression and thought?. The loop looks self-reinforcing: the style earns approval, writers adopt it, and it becomes the baseline readers expect.

The corpus does show that AI text can be told apart at a deeper level than style. AI fiction is detectable from narrative choices alone, such as how much agency characters have and how events are ordered, even after stylistic cues are removed Can AI stories be detected without analyzing writing style?. The differences are real, but they live below the surface that readers and engagement metrics react to. What the corpus lacks is a direct test linking engagement to judged quality in the same study, so 'stylistic acceptance' is the best-supported reading, not a settled conclusion.