When people publish AI-written text, they usually barely edit it, and often prefer the AI's version to their own.

This explores how much people actually revise AI-written text before it goes out into the world, and what follows from that. The short answer from the corpus: not much. In one study, writers edited AI-generated paragraphs only 23% of the time, and even when they did, the edited version stayed about 96% similar to what the AI produced Do writers actually edit AI-generated text before publishing?. Job applications show the same pattern. On Freelancer.com, most workers sent AI-drafted cover letters with very little revision Does editing time on AI drafts predict hiring success?.

The more surprising finding is why the editing doesn't happen. Writers aren't just skipping a chore. They often like the AI's version better. Across 4,503 cases, writers picked the AI rewrite over their own paragraph 63% of the time, and about half said it reflected their views better. They said this even though the AI versions consistently shifted the stance of the original Do writers actually prefer AI-edited versions of their own text?. So the light editing is partly a matter of trust: people read the AI's version, recognize themselves in it, and leave it alone, even when it has quietly changed what they meant.

That has effects downstream. When AI heavily rewrites someone's text, the stylistic fingerprints that identify the author mostly disappear. Attribution accuracy dropped 66.5 points on personal blogs but only 10 points on news writing, because structured, topic-driven prose holds on to authorship cues that personal writing doesn't How much does AI rewriting erase distinctive author voice?. There's also a mismatch in who the text was written for. AI writes for the person typing the prompt, not for the public who will eventually read it Does AI writing collapse the author-to-public relationship?. Light editing means that mismatch goes straight to publication.

Editing still seems to matter when people do it. Within the same freelancer, more time spent revising an AI draft was associated with higher hiring success Does editing time on AI drafts predict hiring success?. Readers also care more about AI involvement than writers assume. They rate disclosure as more necessary than writers do, especially when AI text is used directly and couldn't easily be replaced Do readers and writers differ on AI disclosure necessity?. The actual penalty for disclosing is small, though: less than 0.15 points on a 7-point scale, from human and LLM raters alike Does disclosing AI assistance make readers trust articles less?.

The takeaway you might not expect: the risk isn't lazy writers. Writers sincerely believe the AI's version says what they meant, so the real question is whether a writer can still notice when it doesn't. The corpus says less about what does produce real revision. It hints that setting up the AI's role in advance changes how writers engage Can writers benefit from configuring AI writing partners in advance?, and that seeing collaborators' prompts helps people check AI text Do writers want to see each other's AI prompts in shared editors?. Neither study measures editing rates directly, so this part is still an open question.