The 900% Surge: What Deepfakes Are Actually Teaching Us

In late 2023, there were roughly 500,000 deepfake videos online. By 2025, that number had grown to over 8 million — a 900% annual growth rate. Most coverage of that statistic focuses on the volume. The more important story is what the volume is doing to us.

It isn't only that there are more synthetic images in circulation. It's that the existence of synthetic images — even images we never personally encounter — is quietly changing the epistemic ground we all stand on. When it becomes plausible that any image could be manufactured, the authenticity of every image becomes slightly suspect. And that suspicion, distributed across millions of daily encounters with visual information, produces something more corrosive than any individual fake: it produces a culture of plausible deniability.

Real things can now be dismissed as fake. Fake things can now pass as real. That double movement — in both directions at once — is the actual terrain we're navigating. 

The Detection Gap

The detection problem is real and worth understanding clearly. Research on deepfake detection describes the field as a fast-evolving arms race, with generation capabilities consistently outpacing detection tools and significant limitations in real-world conditions.

More troubling than the technology gap is the human one: studies suggest people can correctly identify high-quality synthetic content only about 25% of the time under realistic conditions.

That number deserves some attention. It means that visual confidence — the feeling of certainty that comes from seeing something — has become an unreliable guide. The mechanism we evolved to use as a ground truth has been partially decoupled from the thing it was supposed to track.

The instinctive response is either blanket skepticism — assume everything is fake — or resigned acceptance — you can't know anything, so why try. Both are wrong, and both are convenient for different reasons. Blanket skepticism is a way of opting out of the hard work of verification. Resigned acceptance is a way of abandoning the standard entirely.

What visual literacy looks like in this context is a third posture: skeptical, not paranoid.

The difference matters. Skepticism is a practice — it asks questions, follows evidence, and remains open to revision. Paranoia is a posture — it assumes the worst and closes inquiry. Skepticism is productive. Paranoia is just a different way of stopping thinking.

The Real Question: Provenance

The deeper issue that deepfakes reveal isn't about fakes at all. It's about what we were using as evidence before they existed.

If seeing was our primary ground for believing, and seeing has now become unreliable, then the question isn't just "how do I detect fakes?" The question is: what should have been doing the epistemic work that we mistakenly assigned to visual confirmation?

The answer is provenance — the documented history of an image's origin, chain of custody, and context. Provenance is what transforms a visual claim into an evidential one.

It's not enough to see; we need to know where the image came from, when it was made, by whom, and for what purpose. This is not a new standard invented for the deepfake era. It's the standard that serious documentary, journalism, and historical scholarship have always applied. What's changed is that we can no longer defer applying it.

Three Verification Moves

Three practical habits help build a provenance-first approach to visual claims.

The first is lateral reading: rather than scrutinizing the image itself for signs of manipulation, leave the page and search for independent verification. What is the earliest known instance of this image? What do credible outlets report about its origin?

Does the claim it's being used to support appear in sources with independent editorial standards?

The second is structural logic: does the 'how' of this image align with its claimed 'what'?

Manufactured authority often breaks down at the seams — context details that don't hold up, metadata that doesn't match the claim. You don't need to be a forensics expert to notice when something doesn't add up; you just need to slow down enough to look.

The third is provenance tracing: for high-stakes content, find the earliest upload, look for full-length versions, and search for multiple independent sources confirming origin. Treat 'I saw a clip' the way a good editor treats 'a source told me' — as a starting point for verification, not a conclusion.

None of this is foolproof. But none of it needs to be. The goal isn't perfect detection. The goal is to stop letting visual confidence do work it was never equipped to do.

Terrain Lens: Deepfake Posture

Before you accept, share, or react to a visual claim — try this:

1.  What is this asking me to feel before I think?

2.  What is the provenance — earliest upload, original source, chain of custody?

3.  Does the structural logic hold: does the 'how' align with the 'what'?

4.  What do independent sources report about this image or claim?

5.  Am I being skeptical — or paranoid? One asks questions; the other closes them.

6.  Is this inviting understanding — or offering permission to stop thinking?

7.  What would I need to see to change my mind — and is that possible?

As the synthetic landscape keeps expanding — what's one verification habit you could build that would change how you encounter visual claims?