AI-Generated Websites Gain Traction Following ChatGPT’s Launch
By mid-2025, 35% of newly published websites are classified as AI-generated or AI-assisted. This marks a significant rise from nearly zero before ChatGPT was introduced in November 2022. This shift underscores the growing impact of AI on digital content creation in recent years.
The study titled “The Impact of AI-Generated Text on the Internet” investigated the prevalence of AI-generated content by utilizing extensive data and rigorous methods. It analyzed 33 months of website snapshots from the Internet Archive’s Wayback Machine. This comprehensive timeframe allowed researchers to assess developments before and after the launch of ChatGPT in November 2022. The classification of web pages as AI-generated or AI-assisted was executed using an AI text detector known as Pangram v3. This tool helped in accurately differentiating between human-written and AI-generated content across the analyzed websites.
The study “The Impact of AI-Generated Text on the Internet” unveiled several key findings regarding AI-generated website content. It revealed that AI-generated sites exhibit a 33% higher pairwise semantic similarity score compared to those authored by humans. Additionally, the content produced by AI demonstrated positive sentiment scores that are more than 107% higher than their human counterparts. Importantly, the study did not find statistically significant evidence suggesting that AI content diminishes the factual accuracy of information on the web.
Furthermore, 83% of survey respondents endorsed the ‘stylistic monoculture’ hypothesis, although this claim was not corroborated by the data. Notably, as AI penetrates 35% of the web, the risk of model collapse transitions from a theoretical possibility to a tangible, observable concern.
The study presents analytical, research-focused findings about AI-generated websites, reporting measured differences in semantic similarity, sentiment, and prevalence while noting a lack of statistical evidence for reduced factual accuracy. Taken together, the results underscore a rapid shift from an internet shaped primarily by human authors to one substantially influenced by AI within roughly three years. The tone of the paper remains analytical and cautious in its presentation of these empirical observations.
