Spotting AI-written text

It is tempting to want a trick for telling AI-written text from human writing. Here is the honest answer: you usually cannot tell reliably, and text detectors are even less trustworthy than image ones. They frequently flag human writing as AI — a problem that has unfairly harmed students and non-native English writers — and they are easily fooled. So do not build your defense around detecting the author. Build it around evaluating the content.

That said, some soft signals can make you suspect a mass-produced or AI-generated piece — never as proof, only as a prompt to check further:

• Fluent but empty. Text that is grammatically smooth yet says little, repeats itself, and hedges without committing to specifics.
• Generic and sourceless. Confident claims with no named sources, dates, or specific people, and details that stay vague where a knowledgeable human would be precise.
• Off in small ways. Odd repetition, a strangely even tone, or facts that sound plausible but turn out wrong on checking.
• Volume and timing. Floods of similar posts or reviews appearing at once, or brand-new accounts pushing the same message, suggest coordinated generation regardless of any single texts wording.

But here is the reframe that actually protects you. Whether a human or a machine wrote something is far less important than whether the claims are true. A true statement is worth the same no matter who typed it; a false one is dangerous whoever wrote it. AI has made misinformation cheaper and faster to produce, but the defense is not authorship-detection — it is the same claim-checking you would apply to any assertion: who is the source, what is the evidence, and does independent reporting back it up.

So when a piece of writing matters — it asks you to believe something, act, vote, buy, or share — skip the guessing game about the author and go straight to the claims. Trace the key assertion to a named, credible source. See whether independent outlets report the same thing. That habit, covered fully in the next module, works whether the text was written by a person, a model, or a person using a model.

Recap. You cannot reliably tell AI-written text from human writing, and text detectors are unreliable enough to wrongly flag real people, so do not center your defense on catching the author. Soft signals like fluent-but-empty, generic-and-sourceless writing, or floods of similar posts can prompt a closer look, but never prove anything. The real move is to judge the claims, not the authorship — trace assertions to credible sources and check for independent corroboration — because a claim truth is what matters, whoever or whatever wrote it.