So here’s something that should probably bother you more than it does: a good chunk of what you’ve been reading online lately wasn’t written by a person. And I don’t mean in that abstract “oh, AI is getting better” way. I mean right now, today, you’ve probably scrolled past dozens of bot-written articles without noticing. According to new research from Pew, we’re living in a version of the internet that’s increasingly… fake.

The Numbers Are Kind of Horrifying

Pew researchers did something pretty straightforward but illuminating – they scanned nearly half a million English webpages from the past five years using Common Crawl, which is basically a massive web archive. Then they ran those pages through an AI detection tool called Open Pangram to see what was human and what wasn’t.

The thing is, they didn’t just look at content from after ChatGPT dropped in November 2022. They went back five years. And when they examined a sample of 10,000 pages from last month specifically, looking for telltale AI phrases and patterns – the kind of language quirks that bots love but humans don’t really use – the results weren’t great.

Now, the researchers were careful to note that AI detection isn’t perfect. Sometimes it flags human writing as AI, sometimes it misses obvious bot content. But the trend here is pretty clear, and it lines up with something people have been whispering about for years.

The Dead Internet Theory Wasn’t Supposed to Be Real

You might’ve heard of the Dead Internet Theory. It’s been floating around since 2021, and the basic idea is… well, exactly what it sounds like. The theory claims the internet essentially died back in 2016, and most of what we see now is AI-generated content, bot activity, and algorithm-driven noise. It started as a fringe conspiracy theory, the kind of thing you’d find in weird corners of Reddit.

Except it’s not looking so fringe anymore.

What Does “Dead Internet” Even Mean?

Look, I’m not saying there aren’t real people online anymore. Obviously we’re all still here, scrolling and posting and arguing in comment sections. But here’s what’s changed: the ratio.

Think about your typical day online. You read news articles, blog posts, product reviews, social media captions, email newsletters. How many of those were written by an actual human being who sat down and thought about what they wanted to say? And how many were cranked out by ChatGPT or Claude or whatever tool some content farm is using this week to meet their SEO quota?

The honest answer is you probably don’t know. And that’s the problem.

“The internet died in 2016 and much of what we see online now is AI-generated.”

Why This Happened (And Why It’ll Get Worse)

Here’s the thing – this wasn’t some evil master plan. It’s just economics. Creating content is expensive. Humans need salaries, benefits, time to research and write. AI needs… I don’t know, electricity? A subscription fee? That math might get harder now that a court just closed a major AI training loophole companies relied on to build these models cheaply.

So publishers, brands, and basically everyone who needs to maintain a “content presence” online did the math. And the math said: why pay a writer $75 an hour when you can get 50 articles from GPT-5 for the cost of lunch?

I get it, I really do. But the result is an internet that’s increasingly generic, repetitive, and weirdly soulless. You’ve felt it, right? That sense that everything kind of sounds the same now? That articles about completely different topics somehow use the exact same phrases? Gaming communities are already tearing themselves apart over this same slop problem as AI-generated content floods their spaces too.

That’s not a coincidence. That’s what happens when large language models train on each other’s output in an increasingly closed loop. It’s called model collapse, and it’s basically the internet eating itself. Meta's new AI agent, Muse, is set to pour even more of this self-referential content into that same closed loop.

The Trust Problem Nobody’s Talking About

What really gets me is how this erodes trust in ways we haven’t fully processed yet. When you read something online, you’re making an implicit assumption that there’s a human being behind it – someone who researched, fact-checked, and put their name on it. Someone you could, theoretically, hold accountable if they got it wrong.

But if it’s a bot? Who do you blame? The company that deployed it? The model that generated it? The training data?

And more importantly – if you can’t tell the difference between human and AI writing anymore, why would you trust anything you read?

What This Actually Means

I don’t have a neat solution here, and I’m not gonna pretend I do. The genie’s out of the bottle. AI writing tools exist, they’re cheap, they’re getting better, and people are going to keep using them. That’s just reality.

But maybe we need to start being a lot more skeptical about what we consume online. Maybe we need better disclosure requirements – like, actual requirements, not optional little notes at the bottom that nobody reads. Maybe we need to get comfortable asking: “Wait, who actually wrote this?”

Because if the Dead Internet Theory is even halfway true, we’re not just losing authenticity. We’re losing something more fundamental – the sense that when we go online, we’re connecting with other people. And once that’s gone… I mean, what’s the point of any of this?