I’ve read an interesting article that was published by Graphite which is a internet company specializing in SEO (search engine optimization), content strategy, and AI-driven marketing. One of the researches was to find out the percentage of content generated by AI. For that the analyzed 65,000 English-language URLs from the archives of Common Crawl (a nonprofit organization that crawls the web and collects snapshots of web pages), covering a 5 year period to May 2025.
How did the researchers know which content was generated by AI?
Each article was assessed using an AI content detector, which classifies content as AI generated, if it matches 50% of its detection criteria. Graphite conducted internal testing to determine the false positive (content marked falsefully as AI generated) and false negative (AI content is marked as human written), which were 4.2% and 0.6% respectivly.
Findings
In November 2024, AI generated articles accounted for 50.3% of new web content, surpassing human-written articles for the first time. This trend coincided with the release of ChatGPT in late 2022, which significantly increased the volume of AI generated content. However, since mid 2023, the growth of AI generated content has stabilized, possibly due to declining search performance and the prioritization of human written content by search engines.
So, is the fact that majority of content is generated by AI, a good or a bad trend?
A research by Zhang & Gosline (2023) had interesting results. They compared four types of content:
- Human made content
- AI only (ChatGPT-4)
- Augmented human – human decides (on subject for ex.), AI assists
- Augmented AI – AI decides, human assists
Findings
Care to guess? content generated by AI only and augmented AI (AI decides, human assist) is perceived as of higher quality than that produced by human experts and augmented human experts.
Second, revealing the source of content production reduces, but does not reverse, the perceived quality gap between human and AI-generated content. This bias in evaluation is predominantly driven by human favoritism rather than AI aversion: When people know a human created the content, they tend to rate it higher as their perception of quality goes up. When people know AI was involved, it doesn’t make them rate it lower as their perception doesn’t change much.
Bibliography: Zhang, Yunhao and Gosline, Renee, Human Favoritism, Not AI Aversion: People’s Perceptions (and Bias) Toward Generative AI, Human Experts, and Human-GAI Collaboration in Persuasive Content Generation (May 20, 2023). Available at SSRN: https://ssrn.com/abstract=4453958 or http://dx.doi.org/10.2139/ssrn.4453958
My personal views
One important remark about the Zhang & Gosline (2023) research – it’s important to acknowledge that although people perceived the AI content as better, doesn’t necessarily mean it really is. “Perceived as better” does not measure or involves fact accuracy, depth, originality, or long-term impact. It may lack nuances, originality and more, and in worst case senarios, AI can halucinate. I’m not sure I have a firm opinion on this, but one thing I’m sure of, and that I would like to see a disclosure added in AI generated content.