Can AI humanizer output still be detected, or does running it through one make it functionally indistinguishable?
I assumed humanizing would be the end of it, but I’ve seen mentions of detectors specifically training on humanized output now, not just raw AI text. Are detectors actually catching up, or do humanizers just vary a lot in quality with the weaker ones being easier to catch?
Yes, this is real. Several detector companies have explicitly stated they’re training on outputs from popular humanizer tools specifically because the volume of humanized-then-submitted content has grown enough to justify it.
From what I’ve seen as an evaluator: detection of humanized text is inconsistent rather than solved. Some detectors catch the more aggressive paraphrasing patterns certain tools use repeatedly. Lighter humanization that doesn’t lean on the same structural tricks is harder to flag.