What I’ve learned trying to tell a real answer from an AI-generated one

Been deep in the problem of spotting AI-generated or copied answers, and a few things genuinely surprised me. Sharing in case it’s useful, and because I’d like to hear how others are handling it.

First, asking nicely does nothing. Telling a model, or a person, not to use outside help just gets ignored, the instruction has no teeth.

Second, a single answer tells you almost nothing. One polished response could be real or pasted, with no reliable way to know in isolation. The signal only shows up across a whole session, consistency, timing, how close the answer sits to what a tool would generate.

Third, follow-ups are the real test. Pasting one good answer is easy, but holding it up through two or three “why”s is where the rehearsed stuff falls apart.

Curious how others dealing with AI-generated input are approaching it, and not just in hiring, support tickets, reviews, applications, anywhere someone can now paste a perfect-looking answer. Are you detecting it, designing around it, or deciding it doesn’t matter?