A photo of something shocking. A video of a leader saying something outrageous. A voice on the phone that sounds exactly like your daughter. AI can now make all of these convincingly, cheaply, in seconds. The instinct is to look harder for the flaw that gives it away. That instinct is now the trap — because the flaws are vanishing, and chasing them gives you false confidence either way. The skill that still works is a different one.
Why your eyes can't do this job anymore
The "tells" are real, unreliable, and disappearing
Extra fingers, garbled text, waxy skin, odd reflections — these used to give AI images away. Some still appear, but they're getting rarer with every model, they were never consistent, and scammers crop or fix the obvious ones. Worse, hunting for tells fails you both ways: you'll call a clumsy real photo "fake," and a clean fake "real." A tell you don't see proves nothing.
It's not just images
Voice can be cloned from seconds of audio. Video and live video calls can be faked. Text — fake reviews, fake "news," fake quotes — is effectively free to mass-produce. There is no single sense you can fall back on; "I saw it / I heard their voice" is no longer proof.
Detector tools won't save you either
"AI detector" apps are unreliable — they miss real fakes and falsely flag real things, and they fall behind each new model. Treat any single detector's verdict as a weak hint, never a ruling.
What actually works
Shift from inspecting the content to checking its origin and corroboration. None of this requires special tools.
- Consider the source and its track record. Did this come from a named, accountable outlet or person with a reputation to lose — or from an anonymous account, a forward, or a brand-new page? Provenance is the single strongest signal you have.
- Look for independent corroboration. If something real and important happened, more than one credible, independent source will have it. Search for it. A shocking clip that only exists on one anonymous account is a red flag by itself.
- Trace it to the original. Reverse-image-search a photo; find the full video, not a clipped 10 seconds; look for the original post. Fakes and misleading edits fall apart at the source.
- Check provenance signals — but don't trust their absence. Some real media now carries "content credentials" (C2PA) showing where it came from. Their presence is reassuring; their absence proves nothing, because most genuine content has none.
- Weigh the incentive and the timing. Who benefits if you believe this, and is it suspiciously well-timed or exactly what one side wants you to feel? Engineered content is built to land at the right moment.
- For a voice or face you "recognize," verify on a channel you chose. A panicked call in a loved one's voice, a video of your boss asking for money — hang up and call back on a known number, or use a family code word. The familiarity is the weapon.
Before you believe or share it
Tick what's true of the photo, video, clip, or message in front of you. Nothing is saved or sent.
The opposite trap
There's a mirror-image danger to falling for fakes: dismissing real things as fake because you can. As fakery spreads, every inconvenient truth can be waved away as "probably AI" — that's called the liar's dividend, and it's how the flood of fakes erodes trust in everything. So the goal isn't to disbelieve everything; it's to calibrate. Hold the unverified loosely, confirm what matters through its source, and let strength of evidence — not your gut reaction to the pixels — decide what you believe.
The close
The age where seeing was believing is over, and no amount of squinting brings it back. But you haven't lost the ability to know what's true — you've just had to move it from your eyes to your judgment: source, corroboration, provenance, incentive. Slower, yes. But it's the one method the fakes can't keep up with, because it never depended on how the thing looked in the first place.