35% of New Websites Are Now AI-Generated, Study Finds
A study by Stanford and Imperial College reveals that by mid-2025, 35% of new websites were AI-generated. Effects include reduced semantic diversity and heightened artificial positivity, with no impact on factual accuracy yet.
Quick Take
AI-generated content now comprises 35% of new websites, up from zero pre-ChatGPT.
Semantic similarity among AI pages is 33% higher, narrowing topic diversity.
Positive sentiment score 107% higher due to LLMs' sycophantic tendencies.
No statistical evidence of increased misinformation or stylistic homogeneity found.
Market Impact Analysis
NeutralThe article discusses AI content prevalence with no direct crypto market connection.
Speculation Analysis
Key Takeaways
- AI now produces 35% of new websites, up from near zero before ChatGPT's late 2022 debut.
- Semantic similarity among AI pages is 33% higher, narrowing the range of online ideas.
- Positive sentiment in AI text is 107% higher, creating an artificially upbeat web.
- No measurable increase in factual errors or stylistic uniformity, despite public belief.
- At 35% saturation, model collapse risk becomes an empirical concern for future AI training.
What Happened
A study from Stanford, Imperial College London, and the Internet Archive found that by mid-2025, 35% of newly published websites were AI-generated or AI-assisted. That figure was essentially zero before ChatGPT launched in November 2022. The research, drawing on 33 months of Wayback Machine snapshots, used an AI text detector to classify each page. The speed of AI's takeover has stunned researchers, with a significant portion of the internet shaped by machines in just three years. This rapid shift marks a fundamental change in who—or what—creates the web's content.
The Numbers
AI-generated sites showed 33% higher pairwise semantic similarity scores compared to human-written ones, indicating less diversity of expression. Positive sentiment scores were more than 107% higher in AI content, attributed to the overly agreeable nature of large language models. Despite widespread concerns, the study found no statistically significant correlation between AI prevalence and factual error rates. Similarly, the fear of a stylistic monoculture—where AI flattens individual voices—was not supported by character-level analysis. These findings challenge common assumptions about AI's impact on the web.
Why It Happened
The explosion of AI content traces back to ChatGPT's release in late 2022, which made large language models accessible to millions. These models are trained to produce outputs close to their training data, inadvertently reducing semantic diversity. Their tendency toward sycophantic, positive text comes from optimization for human approval signals. As adoption accelerated, the web's composition shifted from predominantly human-created to a growing AI share, with profound effects on the tone and range of online discourse. The speed of this transition has outpaced oversight, leaving the web's character transformed in under three years.
Broader Impact
At 35% AI prevalence, the risk of model collapse—where AI models degrade when trained on AI-generated data—moves from theory to reality. A web flooded with homogenous, cheerful content could narrow the Overton window, marginalizing dissent without explicit censorship. This has implications for everything from public opinion to the training of future AI systems. Researchers are now building a live monitoring tool to continuously track AI's share of the web, aiming to provide early warnings as the digital landscape evolves.
What to Watch Next
- The launch of a live monitoring tool by researchers to track AI-generated content share in real time.
- Signs of model collapse in next-generation AI models as they increasingly train on synthetic data.
- Regulatory or platform responses to the growing AI content saturation, especially around transparency and labeling.
This article is for informational purposes only and does not constitute financial advice.
Always late to trends?
Join for the latest news, insights & more.
Disclaimer: Bytewit is an independent media outlet that delivers news, research, and data.
© 2026 Bytewit. All Rights Reserved. This article is for informational purposes only.
