AI Generates Over a Third of New Online Content, Study Finds

A comprehensive analysis by the American research institute Pew reveals that artificial intelligence is rapidly reshaping the internet's content landscape. The study, which examined approximately half a million English-language web pages from the Common Crawl database, found that at least 10% of all active web pages exhibit clear signs of AI writing or editing. This figure surges to over one-third when considering only content published since the advent of generative AI.

The findings indicate an accelerated process where human-generated text is being displaced by AI-produced content, impacting search engines, social media, and news sites. Commercial websites, marketing blogs, and content farms are increasingly using AI models to rapidly generate high volumes of content for increased traffic at lower costs. In contrast, academic and government sites, along with non-profit organizations, show significantly lower adoption rates, with only a single-digit presence in some cases, likely due to stricter regulations and quality control.

This trend is not confined to the United States. Similar tests in Europe, coinciding with the implementation of the EU's AI Act, show comparable patterns. Regulators in the UK and France have expressed concerns about "model collapse," a scenario where future AI systems might be trained on flawed AI-generated text, compromising information reliability. China has implemented strict oversight, mandating digital watermarking for all algorithmically generated content, creating a more controlled environment.

In Israel, many content and commerce websites have adopted AI tools for search engine optimization, leading to a perceived decline in the quality of Hebrew text and the creation of uniform language patterns. The study also identified linguistic markers characteristic of AI writing, such as increased use of long dashes, serial commas, repetitive vocabulary, and formulaic sentence structures with overly structured comparative clauses. Tools like Open Pangram, GPTZero, and Copyleaks are used to detect AI-generated text, but developers acknowledge an ongoing arms race as AI models become more sophisticated at mimicking human writing styles.

The implications extend beyond copyright and factual accuracy, raising concerns about the loss of originality, the perpetuation of cultural biases, and a potential erosion of public trust in online information. The internet is increasingly becoming a space where machines are not just tools but the primary creators of digital discourse.