By mid-2025, artificial intelligence had become one of the noticeable factors reshaping the structure of the internet: around 35% of new websites were created fully or partially with AI. This conclusion was reached by researchers from Stanford University.
Before the public launch of OpenAI’s ChatGPT in November 2022, the share of such websites was close to zero. However, within just a few years, AI tools have evolved from an experimental technology into a mass-market way to generate content for web pages, blogs, media platforms, and commercial websites.
As part of the study, researchers analyzed 33 months of archived website copies from the Wayback Machine. The Pangram v3 detector was used to identify AI-generated content. The main goal of the study was to understand how the mass spread of texts created by neural networks affects the development of the World Wide Web.
Main Changes on the Internet
One of the key findings was a decline in semantic diversity. According to the researchers, pages created with neural networks are 33% more similar to each other than texts written by humans.
In other words, different websites are increasingly publishing materials that retell the same ideas using similar wording. For Russian-speaking audiences, this is especially noticeable in news aggregators, SEO articles, product reviews, financial forecasts, and marketplace descriptions, where phrases increasingly look standard and interchangeable.
The study’s authors emphasize that the problem is not only the mass use of AI for copywriting. More importantly, the diversity of ideas, tones, and ways of explaining information is gradually narrowing. Large language models naturally tend to choose the most averaged and probable answers, which is why they often reproduce a template-like style.
AI Content Has Become Noticeably More Positive
The researchers also recorded a change in the emotional tone of publications. According to their assessment, AI-generated content turned out to be 107% more positive than texts written by humans.
Stanford researchers link this to a known feature of large language models — their tendency to provide more pleasant, cautious, and socially acceptable responses. During training and fine-tuning, such systems are often optimized for safe, polite, and non-confrontational wording.
As a result, a significant share of new websites is creating a more “sterile-friendly” information environment. It contains fewer sharp opinions, disputes, and confrontational statements, but at the same time, it also has less of a real human position, authorial style, and genuine debate.
What Was Not Confirmed
At the same time, several popular concerns were not statistically confirmed. The researchers did not find a significant link between the growing share of AI-generated content and a decline in factual accuracy, an increase in obvious errors, or the complete stylistic leveling of all texts into a single template.
This means that the impact of AI on the internet appears to be more complex than the simple idea that “neural networks make texts worse.” The problem may lie not so much in the number of errors, but in the gradual decline of content diversity and the strengthening of uniform information presentation.
The Risk of “Model Collapse”
The researchers paid special attention to an effect that had previously been discussed mostly in theory — so-called model collapse. It can occur if new neural networks are trained on datasets that already contain a large amount of AI-generated content.
In this case, the system begins learning not from original human texts, but from the averaged responses of previous models. This can reduce variability, worsen the quality of future models, and lead to new AI systems increasingly reproducing not human experience, but a “synthetic echo” of their predecessors.
For media, SEO, education, and online commerce, this could become a serious challenge. The more the internet is filled with similar AI-generated texts, the harder it becomes for users to find original opinions, expert materials, and truly unique content.
What Comes Next
The researchers, together with the Internet Archive, plan to turn the project into a permanent monitoring system for tracking the share of AI-generated content on the internet. Such a tool would make it possible to monitor how quickly the volume of synthetic materials is growing and how it affects the quality, diversity, and structure of the web.
For now, the main conclusion of the study can be formulated as follows: AI has already become an integral part of the modern internet, but its mass use is changing not only the speed of content publishing, but also the very nature of online communication.
What Can You Order from Us?
At LA Marketing, you can order website development and improvement, SEO promotion, GEO optimization for AI search, SMM, Google Ads, design, redesign, technical support, corporate websites, landing pages, online stores, mobile applications, and web applications.
The LA Marketing team works with businesses in the US and Europe, creating modern digital solutions for specific company goals. We use AI where it truly helps speed up processes, but the foundation of every result is always strategy, specialist experience, and an individual approach.
If you need not a generic “AI-made” website like everyone else, but a professional project with a well-thought-out structure, SEO, design, and a focus on leads, you can contact LA Marketing and order a solution tailored to your business.
