Can We Trust Anything Online? Inside the Race to Detect AI Content Before the Internet Drowns in It

Think about the last time you read a product review, scrolled through a job applicant's cover letter, or scanned a claims form. Did you wonder whether a human actually wrote it? A few years ago, that question would have seemed strange. Now it's becoming routine, and the answer is getting harder to pin down.

That uncertainty is at the heart of a growing conversation about what some technologists call "dead internet theory," the idea that so much online content is now machine-generated that the authentic human voice is being drowned out. Max Spero, co-founder and CEO of AI detection startup Pangram, thinks we're closer to that reality than most people realize. Speaking on TechCrunch's Equity podcast with reporter Rebecca Bellan, Spero said the internet could be "dangerously close" to that tipping point within just a few years.

This isn't only about social media feeds filling up with what's been dubbed "AI slop," the low-effort, machine-generated posts and images that clutter platforms like X and Facebook. The problem has spread into places that carry real consequences. AI-generated text is showing up in job applications, where hiring managers now have to guess whether a polished cover letter reflects a candidate's actual writing ability. It's appearing in product reviews, muddying the signals shoppers rely on to make purchasing decisions. And it's turning up in insurance claims, where the stakes of getting it wrong include fraud losses and unjust denials alike.

Pangram is one of a small but growing cluster of companies trying to build what Spero calls a "trust layer" for the internet, a kind of authentication system for content the way a padlock icon signals a secure website. The startup recently raised $9 million to expand its AI detection system and struck a partnership with Substack, which now uses Pangram's technology to flag for readers which newsletter authors rely on AI to write their content. Pangram has also rolled out a new tool for detecting AI-generated images, a response to the fact that synthetic photos have become good enough to fool most casual viewers.

The Detection Problem Is More Complicated Than It Looks

Here's where things get tricky. For a while, the assumption was that AI detection was a binary problem: either a human wrote something, or a machine did. Spero argues that framing is outdated and, frankly, not that useful. The real question isn't whether AI touched a piece of content at all. It's how much.

Think of it like asking whether a house was "built by hand." Almost none are anymore, not fully. Contractors use power tools, prefabricated materials, and computer-aided design software. What actually matters to a homebuyer is whether the craftsmanship, judgment, and final result reflect real skill and care, not whether every nail was driven by a human wrist. Writing is heading in a similar direction. Someone might use AI to brainstorm an outline, tighten a paragraph, or check grammar, and that's a very different act than asking a chatbot to generate an entire essay from scratch. Pangram's approach tries to measure that gradient rather than force everything into an all-or-nothing label.

That distinction matters because the consequences of getting it wrong can be severe. False positives, cases where a detection tool wrongly flags human-created content as AI-generated, carry real weight, especially when the content involves sensitive images. Imagine a photographer or an artist accused of fabricating an image they actually shot or drew by hand. The reputational and even legal fallout from a mislabeled accusation can be significant, and Spero has acknowledged the stakes of what he's called the "America has a Pangram problem" moment, referring to the broader anxiety around detection tools making confident but incorrect calls.

Pangram isn't alone in this space. Competitors like GPTZero and Winston are building similar detection infrastructure, all racing to become the go-to standard for verifying authenticity online. That competition is itself a signal. When multiple well-funded startups converge on the same problem within a couple of years, it usually means the underlying pain point is real and getting worse, not better.

There's also a labor question tangled up in all of this. Spero suggested that the "bottom tier" of writing jobs, the high-volume, low-craft content that used to pay freelancers modest sums for churning out generic blog posts or product descriptions, may be gone for good. AI can produce that kind of content faster and cheaper than any human could. But Spero also floated a more hopeful possibility: genuinely skilled human writing, the kind with voice, insight, and originality, could become more valuable precisely because it's rarer and harder to fake convincingly.

That's a meaningful distinction for anyone whose livelihood depends on writing, editing, or content creation. It suggests the disruption isn't uniform. Some jobs are likely gone. Others might actually gain value as authenticity becomes scarcer and more sought after.

Why It Matters

The dead internet theory used to sound like an internet forum conspiracy, a fringe idea traded among people convinced that most online activity was bots talking to bots. It's worth taking seriously now, not because it's fully true, but because the trend line is pointing in that direction. When AI-generated content infiltrates hiring decisions, consumer trust systems, and insurance processes, the damage isn't abstract. It's people denied jobs based on misjudged applications, consumers misled by fake reviews, and claims processed on faulty assumptions.

Detection tools like Pangram's aren't a perfect fix, and their creators know it. False positives can hurt real people. The technology will always be playing catch-up with generative AI systems that keep getting better at mimicking human output. But the alternative, doing nothing and letting synthetic content flood unchecked into the systems we rely on for trust, carries its own cost. Building better verification isn't about stopping AI from being useful. It's about making sure the internet still has room for something recognizably, verifiably human.