I tested this myself when using Clever Humanizer AI Writer cleverhumanizer. ai/ai-writer for editing one of my essays. The text became way smoother and didnt feel “AI generated” at all lol. Feels like detectors are already falling behind while AI tools keep improving faster every month.
Did they forget to scan for em dashes? AIs can't help but use those long ones instead of short hyphens.
The em dash observation is the only reliable signal at scale right now. GPT-class models default to long dashes mid-sentence in a way native English writers basically never do. Beyond that: hedge stacks ('it's worth noting that...'), the closing engagement-bait questions everyone in this thread can already spot, and an over-fondness for tricolons (three-part lists where two would do). The real problem isn't whether detectors work, it's that Google never said they had to. Their guidance is content quality, not provenance. A site full of AI prose that ranks and converts is fine. A site with shallow research, no expertise signals, and no editorial spine gets buried whether you can detect the AI or not. The detector debate is downstream of the wrong question.
I asked Grok the other day whether gathering information from hundreds of sources to generate an answer should be considered "stealing" or an ethical dilemma. It said yes, because AI infringes on creators' property rights every time it produces an answer. Even the most "honest" AI knows it can act in shady ways while gathering information.