How Much of the Internet Is AI-Generated?

Dead internet theory used to get you laughed off a forum. The idea: most of what you see online, the posts, the comments, even the accounts posting them, is bots and AI, and real people are quietly becoming the minority.

It sounded like a conspiracy theory, and for a while it basically was one, forum lore nobody serious bothered to check. Then researchers actually measured it.

In 2025, researchers at Stanford, Imperial College London and the Internet Archive combed through 33 months of archived web pages using the Wayback Machine. As of May 2025, 17.6% of all newly published websites were entirely AI-generated, and another 17.7% were AI-assisted. That's more than a third of new sites at least partly machine-written. Before the chatbot boom of late 2022, that share was close to zero.

That's why "AI slop" stopped being just a vibe. For a couple of years it was only a feeling: your Facebook feed got weirder, your Google searches got less reliable. Now there's a paper trail.

AI Slop by the Numbers

• Researchers at Stanford, Imperial College London and the Internet Archive put the AI-generated share of newly published websites at 17.6% as of May 2025, with AI-generated or AI-assisted content combined reaching 35.3%.
• Graphite, the SEO firm whose research was reported by Axios, tracked newly published articles classified as AI-generated rising from around 10% in late 2022 to roughly 50% by 2026, a share that has plateaued rather than kept climbing.
• Still, 86% of articles ranking in Google Search results and 82% of those cited by AI chat tools are human-written, per the same Graphite research.
• By June 2026, AI-generated tracks made up more than 50% of daily uploads to the streaming service Deezer, up from just 10% in January 2025. They still account for only 1% to 3% of actual listening, Deezer said.
• Spotify removed more than 75 million "spammy" tracks, many AI-generated, in the 12 months before September 2025, the company said.
• 76% of Americans say it's important to know whether content was made by AI. Only 53% are confident they could actually tell, according to Pew Research Center.

How Much New Web Content Is AI-Generated

The growth wasn't gradual. It was closer to a light switch.

Before November 2022, the AI share of new web content sat near zero. It climbed to 39% within a year, briefly overtook human-written content altogether in November 2024, and has held close to 50% since, according to the Axios report on the research. That trend line comes from Graphite, the SEO firm that sampled 43,000 URLs from Common Crawl, an open archive of web pages, and classified each one as human- or AI-written.

Merriam-Webster named "slop" its 2025 word of the year, defining it as "digital content of low quality that is produced usually in quantity by means of artificial intelligence." The same Stanford researchers found most doomsday predictions about a slop-filled web, like a sitewide collapse in factual accuracy, didn't hold up. Only two of six commonly tested fears did.

AI Slop in Search Results and Social Feeds

Google's AI Overviews, the summaries that sit above search results, appeared in about 18% of searches in March 2025, according to a Pew Research Center study that tracked the browsing activity of 900 U.S. adults. Nearly six in ten of them ran into at least one that month.

The behavior change was immediate. Searchers who saw an AI summary clicked a traditional result link 8% of the time, compared with 15% for searches without one, Pew found. If you've noticed yourself skipping the blue links entirely some days, the data backs up the instinct.

Facebook has its own version. One post from this world cracked the platform's 20 most-viewed pieces of content in the third quarter of 2023, pulling in 40 million views. Stanford Internet Observatory researchers Renée DiResta and Josh Goldstein found it while studying 120 Facebook Pages that had each posted at least 50 AI-generated images. Facebook's own creator bonus program, which pays out on engagement regardless of whether a photo is real, has helped fuel the flood, according to 404 Media's reporting on accounts churning out AI disaster and animal-rescue images for profit.

How Platforms Are Responding to AI Slop

Every major platform is trying some mix of labeling, automatic detection and outright deletion, and so far none of them is winning.

• Label it. Meta began rolling out "Made with AI" tags across Facebook, Instagram and Threads in 2024, tied to the C2PA Content Credentials metadata standard. It was later renamed "AI info" so ordinary photo retouching wouldn't get swept in.
• Detect it automatically. YouTube said in May 2026 that disclosure labels would move below the video player instead of staying buried in the description, and that it would apply labels itself when creators don't disclose.
• Remove and filter it. Spotify rolled out a new spam filter targeting mass uploads and duplicate tracks, on top of a broader takedown effort.

"In the past 12 months alone, a period marked by the explosion of generative AI tools, we've removed over 75 million spammy tracks from Spotify," the company said in a statement announcing the policy.

Deezer went further, becoming the first major streaming platform to tag AI-generated tracks for listeners and pull ones that get no plays within six months or are tied to fraudulent streaming. The company said 85% of streams on its AI-flagged tracks are already caught and blocked as fraud.

Measuring the AI in Your Own Feed

You don't have to guess anymore. The same kind of detection model that flags one image can be pointed at a whole feed, which is roughly how AI or Not approaches it: run the images, count what comes back positive, and you have a real percentage instead of a hunch.

A few things we check before trusting an unfamiliar photo enough to share it:

1. Reverse image search unfamiliar photos, especially disaster or animal-rescue shots.
2. Check the account's posting volume. The Facebook pages in the Stanford study posted dozens of AI images a week.
3. Notice mismatched comments, like people congratulating a clearly synthetic photo as if it were real.
4. Run a suspicious image through a detector before sharing it further.

Pinterest has its own version of this, mostly recipe photos and animal pictures that look slightly too perfect, and there are ways to filter it out board by board.

For a browser-level option, AI Blocker is a free Chrome extension that scans images as they load and blurs the ones its detection model, powered by AI or Not, flags as AI-generated, so you decide whether to look closer instead of scrolling past something fake. The definitional rundown of what these tools do, and don't do, is worth a read before installing one.

AI-Generated Internet FAQ

How Much of the Internet Is AI-Generated?

It depends what's being measured. As of May 2025, 17.6% of newly published websites were entirely AI-generated, according to researchers at Stanford, Imperial College London and the Internet Archive. Separately, roughly half of new online articles were AI-generated by 2026, per Graphite research reported by Axios, though AI-written pages are still a small share of what ranks in Google Search.

What Is AI Slop?

AI slop is low-quality content, usually images, video or articles, mass-produced with generative AI and posted for clicks or platform bonus payouts rather than any real purpose. Merriam-Webster named "slop" its 2025 word of the year for exactly this reason. Facebook's creator bonus program and Deezer's flood of AI music uploads are two documented examples of the incentive behind it.

Is the Dead Internet Theory Real?

Not in its original form. There's no evidence humans are a minority online or that bots and government agents secretly run the internet, the theory's more extreme claim. But the underlying concern, that a growing share of new content is machine-made, is now something researchers can measure, and the numbers aren't small.

How Can I Avoid AI-Generated Content?

Reverse image search anything that seems staged, check posting volume on unfamiliar accounts, and run suspicious images or videos through a detector before trusting or sharing them. Browser extensions like AI Blocker can also blur flagged AI images automatically as you scroll, cutting down on how much slop you see without checking each post by hand.

Dead internet theory started as a joke about a web secretly run by bots. The number now sitting next to it in a research paper is 17.6 percent.