Bots now account for roughly 57 to 58 percent of HTTP requests for HTML content, compared with about 42 to 43 percent from humans, according to Cloudflare's public Radar tracker. Meanwhile, on social platforms like LinkedIn, Medium, Twitter, and Reddit, approximately one in four long-form posts are fully AI-generated, with LinkedIn showing the highest saturation at over 40 percent AI-generated content. This fundamental shift in who actually uses the internet raises troubling questions about information quality, human connection, and the future of the web itself.

What Happens When AI Trains on AI-Generated Content?

One of the most dangerous trends emerging is what researchers call "model collapse." When artificial intelligence systems train on data that was itself generated by other AI systems, the quality degrades rapidly. Steven J. Vaughan-Nichols, a veteran technology journalist who wrote one of the first popular articles about the web in 1993, observed this firsthand while researching AI search engines. He noted that when diving into AI answers today, he finds them referring not to primary sources or reputable secondary sources, but to AI summaries stacked on top of other AI summaries. "When you pile garbage on top of garbage you do not get reliable information," Vaughan-Nichols explained.

This cascading problem is visible across major search platforms. Google's AI Overviews, which generate synthesized answers at the top of search results, frequently pull from AI-generated content rather than authoritative sources. The result is a web increasingly written by machines for machines, with humans left to consume whatever filtered-down information remains.

Where Is All This AI-Generated Content Coming From?

The sources of AI training data remain contentious. Major AI companies including OpenAI have faced legal challenges from publishers and news organizations over how their systems were trained. According to court filings by the New York Times and other outlets, OpenAI acknowledged in a deposition that the company had searched training datasets and output data despite earlier claims that it could not access that information. The same filings alleged that OpenAI deleted logs, which may have violated court preservation orders. This suggests that much of the AI content now flooding the web was trained on human-created material without explicit permission or compensation.

The irony is sharp: AI systems trained on human writing are now generating content that trains the next generation of AI systems, creating a closed loop that excludes the original human creators.

How to Evaluate AI-Generated Information Online

• Check the Sources: Use AI search engines like Perplexity that display their sources alongside answers, allowing you to verify whether information comes from primary sources or other AI summaries.
• Verify with Multiple Platforms: Cross-reference AI answers with traditional search results and expert websites, especially for health, legal, or financial questions where accuracy is critical.
• Assess Author Credibility: When reading long-form content on social platforms, look for author credentials and publication history rather than accepting AI-generated posts at face value.

Why Should You Care About This Trend?

The practical implications are significant. Millions of people now turn to AI for health advice, financial guidance, and other critical decisions, despite the systems' well-documented tendency to produce confident-sounding but inaccurate answers. Vaughan-Nichols noted that he has no way to verify medical information from AI systems and would never trust them for serious health problems, yet he knows millions of people do this every day. This represents a genuine public health and safety concern.

Beyond accuracy, there is a deeper human cost. As AI increasingly handles both the creation and consumption of web content, the space for authentic human connection and expertise shrinks. The web was supposed to democratize access to information and bring people together. Instead, it is becoming a machine-to-machine ecosystem where humans are peripheral observers.

The data tells the story: Imperva's Bad Bot Report, based on 2025 data, found that bots accounted for approximately 53 percent of measured web traffic for the second year in a row, with humans at 47 percent. On social platforms, the situation is even more extreme. X (formerly Twitter) shows the worst saturation, with almost half of articles either fully AI-generated at 23.9 percent or AI-assisted and mixed at 22.9 percent, leaving only 53.2 percent flagged as fully human-authored.

"The reason I use Perplexity as my search engine isn't that it's more accurate than other AI LLMs; it's that it shows me its sources. I can see if what it just turned up is the real thing or just BS. Guess what? It's often crap," said Steven J. Vaughan-Nichols.

Steven J. Vaughan-Nichols, Technology Journalist

The web Vaughan-Nichols helped pioneer in 1993 was built on the promise that human knowledge and connection would flourish online. Today, that vision has inverted. We are left with a web written by AI for AI, where humans are reduced to consuming the scraps and wondering how we got here.