Marty Pesis, founder and CEO of Troveo, is a data specialist focused on providing real-world information for AI model training. In recent commentary, he argued that AI-generated web pages now comprise roughly 75% of newly created web content, creating an urgent shortage of authentic human-generated data that AI companies need to improve their systems.
Key Takeaways
- Approximately 74-75% of new web pages contain AI-generated content, according to multiple 2025 studies.
- AI content flooded the internet rapidly after ChatGPT’s launch in November 2022, surpassing human-written content by late 2024.
- Search engines penalize AI-generated content; 86% of Google’s top results remain human-written.
- Companies are desperately seeking authentic video data and human-generated content to train higher-quality AI models.
- McKinsey projects AI-powered search will drive $750 billion in US revenue by 2028.
The AI Content Explosion: What the Data Shows
Pesis’s 75% claim aligns closely with independent research. An Ahrefs analysis of 900,000 newly created English-language web pages found that 74.2% contained AI-generated content. The breakdown reveals a mixed landscape: 2.5% pure AI, 25.8% pure human, and 71.7% mixed content combining both sources. A separate Graphite study tracking 65,000 articles from 2020 to 2025 documented an even sharper rise, with AI-generated articles surpassing human-written ones around November 2024. By May 2025, approximately 52% of newly published articles contained substantial AI content.
The timeline matters. Before ChatGPT’s public release in November 2022, AI-generated content represented less than 10% of web pages. By late 2023, that figure had climbed to roughly 39%, and by late 2024, the split reached nearly 50-50 human versus AI. This acceleration happened in just two years—a dramatic shift in the composition of the internet itself.
Why Search Engines Are Rejecting AI-Generated Content
Despite AI-generated content dominating the raw numbers, search engines have begun actively filtering it out. Google’s top search results remain 86% human-written, according to Graphite analysis. This disparity reveals a critical problem: quantity does not equal quality. AI-generated content—particularly news articles, how-to guides, lifestyle pieces, and product reviews—ranks poorly in search results and gains minimal visibility on AI-powered search engines like ChatGPT and Perplexity.
The gap between AI-generated web pages and AI-generated web pages that actually rank suggests search algorithms now penalize low-quality AI slop. Google has made explicit moves to downrank unhelpful AI content, and the market is responding. McKinsey research indicates that 50% of Google searches currently display AI-powered summaries, with projections exceeding 75% by 2028. This shift toward AI-mediated search will drive an estimated $750 billion in US revenue by 2028.
The Desperate Hunt for Real Human Data
Pesis’s argument centers on a paradox: AI companies trained their models on vast amounts of internet data, but now that the internet is flooded with AI-generated content, those same companies face a data quality crisis. New AI models trained primarily on AI-generated content tend to produce lower-quality outputs, introducing errors and degradation with each generation. The solution, according to Pesis and other data specialists, is to source authentic human-generated data—particularly real-world video footage and written content created by actual people.
This demand has created a market opportunity. Companies like Troveo focus specifically on curating and providing high-quality human-generated data for AI training. The shift reflects a fundamental realization: garbage in, garbage out. As AI-generated web pages now comprise the majority of new online content, the scarcity of authentic human data has become a competitive advantage. Firms willing to invest in collecting and labeling real human data can train models that perform noticeably better than competitors relying on contaminated datasets.
How Real People Experience AI-Generated Content Online
Pew Research surveyed 900 respondents tracking their web browsing across 2.5 million page visits in March 2025. The findings show that 58% of users encountered AI-generated search summaries during their browsing. Interestingly, only 7% of the pages users visited contained explicit AI-related terminology, and 49% visited at least one AI-focused page, though typically only a few times. The data suggests that while AI-generated content is ubiquitous, most users do not consciously recognize it as such.
The Reliability Problem: Can We Even Detect AI Content?
Multiple AI detection tools exist, but none are perfectly reliable. The Surfer AI detector, used by Graphite in their research, has a false positive rate of 4.2%—meaning it incorrectly classified 4.2% of human-written articles as primarily AI-generated. Ahrefs warns against using AI detection tools for high-stakes decisions, acknowledging that no detector catches every instance of AI content. This uncertainty complicates efforts to clean datasets and verify authenticity, adding another layer to the data quality crisis.
Frequently Asked Questions
Why do AI companies need human-generated data if they already have billions of web pages to train on?
Training AI models on AI-generated content creates a degradation loop where errors and biases compound across generations. Authentic human data produces higher-quality, more reliable models. As AI-generated web pages now comprise 74-75% of new content, human-generated data has become scarce and valuable.
Is AI-generated content actually ranking poorly on Google, or is that just a myth?
It is real. Graphite’s analysis found that 86% of Google’s top search results remain human-written, despite AI-generated content dominating the broader web. Search engines actively penalize low-quality AI content, which is why most AI-generated articles never gain visibility.
Will the internet eventually be 90% AI-generated?
A 2022 Europol projection suggested AI could account for 90% of online content by 2026, but more recent data from 2024-2025 shows growth plateauing around 50-75%. Search engine penalties and quality concerns may slow further expansion of AI-generated content.
The AI-generated web pages crisis is not a distant threat—it is happening now. As Pesis and independent researchers have documented, the internet has fundamentally shifted in just two years. The scarcity of authentic human data is reshaping AI development, creating new business opportunities for those who can provide it and pushing search engines to increasingly favor human-created content. For AI companies betting on next-generation models, the path forward runs directly through the challenge of finding and securing genuine human-generated data.
Edited by the All Things Geek team.
Source: Tom's Guide
