Spot AI content

You can't reliably tell AI writing from human writing just by reading it — and neither can the experts. The honest skill is knowing which tells are real hints, and when to stop trusting your gut.

Tells are hints, not proof

There are real patterns: vague hedging, relentless balance, “in conclusion” scaffolding, a smoothness with no rough edges. But every one of them is only a hint. A careful human editor writes clean, balanced prose; a prompted AI can fake messy, casual, typo-ridden “human” voice on demand. The tells point a direction — they never settle it. Where a file actually came from is its provenance, and that's the thing worth checking.

Why your gut isn't enough

Detectors don't rescue you either: AI-text detectors are unreliable, flag human writing as fake, and are easily fooled. As models improve, the surface tells fade. So the reliable move isn't reading harder — it's checking the source: who published it, does the account have a history, is there a signed record of how the file was made? That signed record is the strongest single check available.

A source-first habit

Turn the question around. Instead of “does this read like AI?”, ask “where did this come from, and can I trace it?” For a shocking image, look for the original post and the account behind it. For a quote, find the primary source. For a file, check for Content Credentials or a watermark. Treat the writing style as a weak clue and the provenance as the evidence — that's the habit that survives better and better AI.

• —Stylistic tells are weak hints — they cut both ways and fade as AI improves.
• —AI-text detectors are unreliable; don't trust a verdict from them.
• —Check provenance — source, account history, Content Credentials — not the vibe.