A Third of the Post-ChatGPT Web Is AI-Written, Pew Finds

We all knew the wave was coming, but the latest data from Pew Research suggests it has already crashed onto the shore. According to their recent study, nearly one-third of the content currently living on the public web shows signs of being generated by artificial intelligence. This is not just a trend for the tech-savvy; it is a fundamental shift in how the internet is being constructed.

For those of us building in the crypto and AI space, this shouldn't be surprising. We have seen the deluge of automated X threads, the synthetic LinkedIn thought leadership, and the endless stream of SEO-bait blogs. But seeing the numbers laid out—based on a scan of nearly half a million pages—provides a sobering reality check on the state of digital information.

The .com Problem

One of the most telling aspects of the Pew report is where this AI content is clustering. It is not hidden away in dark corners of the web; it is front and center on .com domains. The commercial internet is being hollowed out by tools designed to fill space rather than provide value. For founders, this creates a massive noise-to-signal problem.

When everyone has a button that generates a 1,000-word article in ten seconds, the value of a 1,000-word article drops to zero. We are witnessing the inflation of content, and just like monetary inflation, it devalues the holdings of everyone in the ecosystem. If you are building a brand today, relying on AI to handle your communication is essentially betting against your own uniqueness.

The LLM Feedback Loop

There is a technical debt being accrued here that many builders are ignoring. As the web becomes saturated with AI-generated text, future models are inevitably going to be trained on that same data. We are entering a cycle of model collapse, where AI learns from AI, slowly degrading the nuance, factual accuracy, and creative spark of the output.

If you are a developer building RAG (Retrieval-Augmented Generation) systems or data-driven products, this is a nightmare. The "truth" of the internet is being diluted. You can no longer trust a random scrape of the web to provide high-fidelity training data. The premium on human-verified, proprietary data sets has never been higher.

Why Builders Should Be Skeptical of Automation

It is tempting to automate your entire marketing stack. I get it. Founders are busy, and AI is cheap. But if a third of the web is already AI-written, the one thing that will stand out to a sophisticated user is something that feels distinctly human. The markers of AI writing—the overly structured lists, the lack of strong opinions, the repetitive transitional phrases—are becoming easier for the average person to spot, even without a detector.

In the crypto world, we talk a lot about decentralization and censorship resistance. But we also need to talk about cognitive resistance. If we allow our platforms to be flooded with synthetic filler, we are just recreating the same low-quality ad-tech hellscape that we claimed we wanted to replace.

The Search Engine Arms Race

Google and other search providers are in a bind. They need to index the web, but the web is increasingly becoming a mirror of their own AI tools. The Pew study suggests that the fingerprints of these tools are multiplying fast. This will lead to a more aggressive filtering process. If your startup's landing page or documentation looks like it was spat out by a basic GPT-4 prompt, don't be surprised when your organic traffic vanishes.

The takeaway for builders isn't to stop using AI—that would be foolish. The takeaway is to use it as a scaffold, not the finished building. AI is a great researcher and a decent proofreader, but it is a terrible visionary. If you let it speak for you, you are telling the world you have nothing original to say.

Strategic Realignment

As we move forward, the definition of "quality" is going to change. It won't be about who can produce the most content, but who can produce the most original insights. We need to focus on deep-dive analysis, primary source reporting, and actual lived experience. These are the things an LLM cannot fake, no matter how many tokens you give it.

For those building tools to detect AI or manage content, this Pew data is a clear signal of market demand. There is a growing hunger for authenticity. Whether that is solved through cryptographic proof of personhood or better heuristic analysis, the builders who figure out how to verify human intent will be the ones who win in the next phase of the web.

The internet is becoming a library where a third of the books were written by a machine that doesn't understand the words it is using.

Don't be another page in that machine-written library. If you are building in AI, build things that help humans think better, not things that think for them. If you are building in crypto, use the tech to prove what is real. The web is getting noisier; your job is to be the signal.

---

Read the original at Decrypt →