How AI is Erasing the Internet’s Collective Memory

Photo by Shubham Dhage on Unsplash

The Problem: AI is Consuming the Web

Something concerning is happening to the internet right now, and most people aren’t paying attention. As artificial intelligence systems become more powerful, they’re being trained on massive amounts of data scraped from the web. But unlike when humans browse the internet, AI systems consume content in a way that’s fundamentally different—and it’s making information harder to find and preserve.

The core issue is straightforward: when AI models are trained on billions of web pages, that training process extracts value from original creators while creating new problems for the entire information ecosystem. Original articles, research, tutorials, and discussions are being used to build AI systems, but the original sources are increasingly being bypassed. People start asking ChatGPT or Claude instead of searching Google and reading the original articles.

What’s Really Happening to Online Content

Think about how the web worked ten years ago. If you had a question, you’d search Google, find relevant articles from different websites, and read through several sources. Website owners benefited from this traffic. Content creators knew their work had an audience. The entire system created incentives for people to publish original, high-quality content.

Now imagine a different scenario. AI systems train on that same content, but instead of directing people to the original sources, they synthesize that information into a generated response. The user gets an answer without visiting any of the original websites. The creators don’t see traffic. The original articles gradually become less valuable—both economically and in terms of visibility.

This creates what researchers and observers are calling “collective memory loss.” When people stop visiting original sources and instead rely on AI summaries, those sources become less discoverable. Search ranking algorithms change. Ad revenue dries up. Eventually, creators stop maintaining those pages, and they disappear or become outdated. The web loses its depth of original knowledge.

The Scale of the Problem

This isn’t theoretical. We’re already seeing it happen. AI companies have been scraping the entire public web—including articles behind paywalls, copyrighted material, and personal blogs—to train their models. Some estimates suggest that AI systems have already consumed years’ worth of human-generated content.

Meanwhile, website traffic patterns are changing. Reddit and Stack Overflow, which host millions of community-driven Q&A threads and discussions, are seeing increased value as training data—but also facing requests to limit that access. Forums and independent blogs report declining traffic as users shift toward asking AI assistants instead.

The irony is sharp: the more information AI systems consume from the web, the less sustainable web publishing becomes for original creators. If there’s no audience, there’s no motivation to publish. If people don’t publish original content, what will future AI systems train on?

Why This Matters for Knowledge and History

Beyond the economic impact on content creators, this creates a serious problem for long-term knowledge preservation. The internet has been humanity’s largest collective archive—a messy, imperfect, but remarkably comprehensive record of human knowledge, culture, and events.

When that archive becomes less accessible because original sources disappear or stop being maintained, we lose something valuable. Historians won’t have the original blog posts and forum discussions that documented how people actually lived in the early 2020s. Researchers won’t have access to niche expert knowledge that only existed on specialized websites. Future AI systems won’t have quality training data because the sources have vanished.

There’s also the question of verification. When information comes from an AI model trained on unverified web content, it’s impossible to trace claims back to original sources. Users can’t assess the reliability of information the way they could when reading original articles. The “collective memory” isn’t just disappearing—it’s being filtered through a process that obscures its origins.

What Happens Next?

The internet is at a potential inflection point. If current trends continue, we might see a fundamental shift: a web optimized for AI training rather than human reading, with fewer original sources and more synthetic content generated by AI models.

Some organizations are already responding. Publishers are blocking AI scraping. Creators are exploring alternative models. There’s conversation about AI companies compensating creators for their work. But these are early responses to a problem that’s still unfolding.

The central concern remains clear: as AI systems eat the web for training data, the incentive structure that made the web a valuable archive of human knowledge breaks down. Without active steps to preserve and sustain original content creation, the internet’s collective memory—the accumulated knowledge that makes the web useful in the first place—is disappearing.