Do writers edit AI drafts enough to fool detectors, or do most publish them nearly unchanged?

This explores whether writers rework AI-assisted text enough to slip past AI detectors. The short answer from the collection is that most writers barely edit at all, so the question of fooling filters usually never comes up. One large study found writers changed AI-suggested paragraphs only 23% of the time, and the edits they did make left the text about 96% identical to the original Do writers actually edit AI-generated text before publishing?. In practice, the human writer is a very weak filter: whatever the model produced is mostly what gets published.

Why so little editing? One answer is that writers like what the AI gives them. Researchers trained reward models to remove the ways AI assistance distorts a writer's voice, and writers then accepted the output less often Can AI writing assistance remove distortion without losing appeal?. The polish, confidence and clarity people want come from the same tendencies that make the text sound less like them. Those tendencies are measurable. Across 29 traits, AI-assisted writers came across as more confident, more agreeable and more extreme Does AI writing assistance change how readers perceive the writer?. Readers also judged them far more likely to be educated, well-paid native English speakers Does AI writing make authors seem more privileged than they are?. Writers have little reason to edit away a voice that flatters them.

When people do rewrite heavily, the evidence on detector evasion is thinner than it sounds. One paper claims heavily rewritten messages hide both who wrote them and the fact that AI was involved. But it only tested authorship attribution and never ran an AI detector, so the 'fools detectors' half of the claim is unproven Do rewrites that hide authorship also fool AI detectors?. Detection may also be harder to escape than people think. A fiction detector reached 93% accuracy using only story-level choices, such as how much control characters have over events and how the timeline is ordered, with no word-level style clues at all Can AI stories be detected without analyzing writing style?. Swapping words and smoothing sentences won't touch those signals. Only rethinking the story itself would.

The strongest detector may be ordinary readers. In an experiment with AI commenting tools, readers rated AI-assisted discussion as more generic and less authentic. That drop in perceived quality spread even to conversations among people who never used the tools Do AI writing tools improve online discussion or degrade it?. And the penalty for simply admitting AI help turns out to be small: disclosure lowered article ratings by less than 0.15 points on a 7-point scale, from both human and LLM raters Does disclosing AI assistance make readers trust articles less?. Some writers even prefer shared editors that show collaborators their AI prompts Do writers want to see each other's AI prompts in shared editors?.

The collection doesn't contain a direct test of how much editing it takes to beat a commercial AI detector. What it does suggest is that the 'fooling filters' framing may miss the point. Most AI-assisted text passes through almost unedited. The bigger effect isn't detectors being fooled. It's the AI's voice spreading quietly, changing how writers come across to the people reading them.