AI Content Now Exceeds 50% of the Internet
▼ Summary
– A new report found that just over half (52%) of new internet articles are AI-generated, based on analysis of 65,000 English articles from 2020-2025.
– The share of AI-generated articles grew rapidly after ChatGPT’s 2022 launch but has recently plateaued around 50-50 with human-written content.
– The actual proportion of human-written content may be higher because paywalled sites blocking data collection were excluded from the analysis.
– AI detectors like Surfer used in the study have reliability issues, sometimes misidentifying human articles as AI-generated.
– The plateau may occur because AI content performs poorly in search results, and human-AI collaboration is blurring the distinction between content types.
A new report reveals that artificial intelligence now generates just over half of all new online articles, marking a significant milestone in the digital content landscape. While this figure might seem alarming, it actually signals a potential stabilization after years of rapid AI content expansion.
The analysis comes from SEO company Graphite, which examined 65,000 randomly selected English-language articles published between January 2020 and May 2025. Using Surfer AI detection technology, researchers classified any article with at least 50% AI-generated content as machine-created. The data shows a dramatic surge following ChatGPT’s public debut in late 2022, with AI content jumping from approximately 10% to over 40% within two years before growth began moderating.
What’s particularly noteworthy is that AI content appears to have reached a plateau after peaking in November 2024. Since then, the balance between human and machine-generated material has remained relatively stable, with AI content currently accounting for 52% of new publications. This represents a slight shift from the previous month when human-written articles briefly regained majority status.
The actual proportion of human content might be even higher than reported. The study relied on Common Crawl, an open-source dataset containing hundreds of billions of webpages. Many subscription-based websites now block Common Crawl from accessing their content, primarily because AI companies extensively used this data to train their language models. These excluded publications likely contain predominantly human-written articles that weren’t included in Graphite’s analysis.
AI detection tools themselves present reliability concerns. During accuracy testing, Graphite discovered that Surfer incorrectly identified human-written articles as AI-generated 4.2% of the time, though it rarely mistook AI content for human writing (just 0.6% errors). This tendency toward false positives highlights the challenges in precisely measuring AI’s footprint.
The reasons behind this apparent leveling off remain unclear, though search engine performance might play a role. A separate Graphite report indicates that human-written content dominates Google Search results, comprising 86% of articles compared to just 14% AI-generated material. This suggests content farms producing low-quality AI material may be receiving less visibility in search rankings.
Another factor complicating the measurement is the growing integration of AI tools into human writing workflows. Many writers now use language models for research, drafting, or editing assistance, creating hybrid content that blurs the distinction between human and machine creation. As one computer science expert observed, the relationship has evolved into more of a symbiosis than a simple competition between human and artificial intelligence.
(Source: Futurism)
