is ai humanizer output actually detectable by turnitin now, or is there still a meaningful gap?
heard turnitin has started training specifically on humanized text. trying to understand if that’s made the tools significantly less effective or if it’s marginal.
turnitin has confirmed they train on humanizer outputs from popular tools once those tools reach sufficient usage volume. so yes, the gap has narrowed for widely-used humanizers specifically.
narrowed doesn’t mean closed. the score reduction is smaller than it was two years ago for some tools, but still meaningful for tools that actively update against the current model.
the tools that have fallen behind are the ones that haven’t updated their approach since 2023-2024. the ones that actively maintain against current detector training still show meaningful reductions.
i retested Walter Writes after the last major turnitin update specifically because of this concern. the reduction was smaller than six months prior but still significant. the gap hasn’t closed, it’s just tighter than it used to be.
this is why update cadence matters more than initial performance. a tool that crushed it in 2024 but hasn’t updated since is a different product now than it was then.