How often does LinkedIn wrongly flag legitimate posts?

The 94 percent figure is the only measurement in this excerpt of how well LinkedIn's filter works, and it does not answer the question that matters most to a legitimate writer. The excerpt reports that "in initial tests, the company says it correctly tagged generic content 94 percent of the time." It adds that the company "hasn't shared any data that can be independently verified," and that how often legitimate posts get wrongly flagged as AI slop "is anyone's guess." On its face, a rate of correct tags on generic content does not report the rate at which distinctive posts are wrongly tagged as generic.

The consequence of a flag is reduced reach, with content staying "within the author's own network." The excerpt does not say whether an author is told they were flagged or can contest the decision. A false positive would therefore be a quiet demotion. The mechanism also makes the error hard to separate from provenance. The target is text that "appears to be generated by AI and lacks clear perspective," so a classifier has to infer both machine origin and an absent voice from the text alone.

That inference is where the neighboring notes bear. Do writers actually edit AI-generated text before publishing? describes machine prose that passes to readers with minimal revision, which is the easy case: flagged text is often unedited. The harder case is the one Does AI writing assistance change how readers perceive the writer? describes, where AI assistance shifts a writer's apparent stance, emotion and competence. A machine-assisted post that carries a real perspective could look generic to a classifier. Conversely, a comprehensive, authoritative-sounding post of the kind Why do AI posts get likes without inviting conversation? describes is the register a generic-content filter most needs to separate from substance. These are inferences from the neighbors. The excerpt tests neither case.

The excerpt does not establish a false-positive rate, the test sample, the scoring rule, or whether the 94 percent applies to comments as well as posts. Until LinkedIn or an outside party publishes those, the effect of the reach cut on legitimate writers is a question, not a finding. The honest reading is that the platform has described a filter and its rationale but has not shown its error rate on the people it most affects. This note adds the question. It does not claim the filter is wrong.