When AI rewriting hides who wrote a text, does it also hide the clues that a machine was involved?

This explores whether AI rewriting that wipes out a writer's fingerprint also wipes out the clues that a machine touched the text. The short answer is that the corpus doesn't know yet, and the evidence it does have points the other way. The "double erasure" claim, that heavily rewritten messages fool both authorship tools and AI detectors, was stated without any detector experiments. Its only support is attribution results showing that rewritten texts converge stylistically Do rewrites that hide authorship also fool AI detectors?. That convergence is real and large. Heavy AI rewriting cuts author-attribution accuracy by 66.5 points on blogs but only about 10 points on news, because topic-driven writing keeps cues that personal writing loses How much does AI rewriting erase distinctive author voice?.

But "converging" raises the question of what the texts converge toward. The most likely answer is the model's own voice. One large study found that AI assistance shifted how readers saw writers on all 29 measured dimensions. Writers came across as more confident, more agreeable, more extreme and more polished, and the shifts ran in consistent directions rather than at random Does AI writing assistance change how readers perceive the writer?. A separate experiment found the same pull across cultures: autocomplete nudged Indian writers toward Western phrasing and references Do AI writing assistants push non-Western writers toward Western styles?. If rewriting erases human variety by replacing it with a shared machine house style, it may make AI markers stronger rather than weaker. The text loses its author but gains a signature.

People also rarely clean up afterward. Writers edited AI-generated paragraphs only 23% of the time, and their edits left the text about 96% the same as before Do writers actually edit AI-generated text before publishing?. So whatever markers the rewrite introduces mostly reach readers unchanged.

There is a deeper twist. Some AI markers aren't stylistic at all. In fiction, a detector reached 93.2% accuracy using only story-level choices such as how characters act and how time is ordered. It kept 97% of that performance after all style cues were removed, because changing those choices requires real rewriting, not surface polish Can AI stories be detected without analyzing writing style?. A related argument holds that AI text lacks things human writing has by nature, such as a real author with lived experience. AI hotel reviews, for example, are detectable at over 80% partly because they make false claims about experiences no one had Does AI-generated text lose core properties of human writing?. Rewriting can sand down word choice, but it can't add a missing author.

The practical takeaway is that "who wrote this?" and "did a machine write this?" are different questions, and making the first harder to answer may make the second easier. Detection by humans is a separate and shakier matter. Comments that people accuse of being AI-written often lack any features that actually set AI text apart, so those accusations act more as gatekeeping than detection Do unfounded AI accusations harm human writers instead?. To settle the question, someone would need to run AI detectors on the same rewritten texts used in the attribution studies, and the corpus doesn't contain that experiment yet.