You can't tell what's real anymore

A survey out this week put a number on something we've all felt coming. 85% of people now say they can't reliably tell real content from AI-generated content — up from 66% just a year ago. Half had run into an AI-driven scam in the past twelve months: personalized scam messages, faked product reviews, AI images, voice-cloned calls that sounded like someone they knew. And the line that should stop you: 84% said "convincing video evidence" no longer feels like proof.

Sit with that last one. For the entire history of the internet, the default assumption was "what I'm seeing is basically real, and I'll get suspicious only if something's off." That assumption just inverted for almost everyone. The new default is doubt. And once you see that, you realize it's not only a scam story — it's a quiet change in what every product that touches users now has to do.

Trust stopped being free

The thing AI broke isn't any single piece of content. It's the baseline. Trust used to be the default state of digital life — you assumed the email was from your bank, the review was from a customer, the voice on the phone was your colleague, the video was of a real event. AI made all four cheap to fake, so the baseline flipped from "trusted unless suspicious" to "suspect unless verified." Authenticity, which was ambient and free, is now scarce and expensive.

That has a direct cost for anyone building products, on two sides.

The defensive side: your users are now targets

If your product has any flow that trusts "it looks like them" or "it sounds like them," that flow is now exploitable. The survey is full of the gap between fear and action: 67% worry about voice cloning but only 19% have done anything about it, 81% fear someone faking their family's likeness but only 13% have set up a codeword. Your users won't defend themselves — they don't know how. So a support agent approving a request because the caller's voice matches, a password reset that trusts a face, a review system that assumes a human wrote it — these are no longer edge cases. Design as if every "is this really them" signal can be synthesized, because now it can.

The opportunity side: provenance becomes a feature

Here's the more interesting half. When authenticity is scarce, being verifiably real becomes valuable — and that's something you can build. The same week, regulation and industry are converging on it: the EU AI Act will require machine-readable marking of AI content from August, and standards like C2PA — cryptographically signed "content credentials" that record who made something, with what tools, and whether AI was involved — are showing up in OpenAI's outputs and even Samsung's camera.

The product lesson generalizes past media. In a world that assumes everything might be fake, the things that earn trust are the ones that show their work: a claim with a checkable source, an answer grounded in something verifiable rather than asserted, a provenance trail a user can inspect. "Trust me" is worth nothing now. "Here's how you can verify this yourself" is worth a lot. Building that verifiability in — where your content came from, why an answer is what it is, what's human and what's machine — is shifting from a nice-to-have to table stakes.

Be honest about the limits

One caution, because the easy version of this overpromises. Provenance is not a truth machine. A valid signature proves a file wasn't tampered with after signing — not that it's fair, accurate, or in context, and a missing signature doesn't prove something's fake. And the deeper problem, the "liar's dividend," doesn't fully go away: once everyone knows fakes exist, bad actors dismiss real evidence as "probably AI." Provenance raises the bar — signed content can't be waved away without explaining the forgery — but it doesn't end the war. Sell verification as evidence, not as certainty, or you'll erode the very trust you're trying to build.

The bottom line

The headline is that people can't tell real from fake anymore. The deeper story is that the internet's free, ambient trust — the assumption underneath email, reviews, calls, and video — is gone, and it's not coming back. AI made fakery cheap, and in doing so made the opposite, verifiable authenticity, the scarce and valuable thing.

So build for the world that actually exists now, not the one your auth flows were designed for. Assume any "it's really them" signal can be faked, and stop relying on it. And on the other side, make trust something your product earns and lets users check — provenance, sources, grounding, showing your work — because in a sea of convincing fakes, the rare thing worth paying for is something you can actually verify. Trust used to be the default. Now it's a feature, and the products that build it will be the ones people still believe.