New data suggests machine-generated articles have plateaued at around 50% of online content—offering unexpected hope for human writers
As reported here, just over half of new online articles now come from artificial intelligence—a figure that, surprisingly, offers reason for optimism.
When ChatGPT debuted in November 2022, many feared an unstoppable tsunami of machine-generated content would overwhelm the internet. While AI-produced material has indeed surged, recent evidence suggests we’ve avoided total submersion.
SEO company Graphite examined 65,000 randomly selected English articles published from January 2020 through May 2025. Using Surfer, an AI detection tool, researchers classified any piece with 50% or more automated content as AI-generated.
The findings confirm the expected pattern: AI article production skyrocketed following ChatGPT’s release, jumping from approximately 10% in late 2022 to over 40% throughout 2024, before the growth rate moderated.
The encouraging development? AI content appears to have plateaued. After peaking in November 2024, the ratio between machine and human writing has stabilized near an even split. As of May, AI accounts for 52% of new articles—barely edging out human content, which briefly held the majority just one month earlier.
The actual proportion of human-created work may be even higher than reported. Graphite’s analysis relied on Common Crawl, an open-source archive containing hundreds of billions of webpages that AI companies exploited for training data. Many paywalled publications have since blocked Common Crawl’s access, as Axios points out, meaning these predominantly human-authored articles wouldn’t appear in the dataset.
AI detectors themselves warrant skepticism given their questionable accuracy. Graphite’s internal testing revealed that Surfer mislabeled human writing as AI-generated 4.2% of the time—a frequent flaw in such tools—though it rarely made the reverse error, misidentifying AI content as human just 0.6% of the time.
Why has AI content growth stalled? The answer remains murky. Citing additional Graphite research, Axios notes that content farms may be discovering their low-quality output gets less traction with search engines and chatbots. Graphite found that 86% of Google Search results feature human-written articles versus only 14% AI-generated ones.
However, this masks an emerging reality: more writers now incorporate AI tools into their workflow, which can fool detectors and obscure the boundary between machine and human creation.
As UCLA computer science professor and Amazon Web Services vice president Stefano Soatto told Axios, “At this point, it’s a symbiosis more than a dichotomy.”
While the plateau in AI-generated content offers temporary relief, the situation remains precarious and could shift dramatically at any moment. AI tools are evolving at breakneck speed, becoming more sophisticated and accessible with each iteration. What required technical expertise yesterday can be accomplished with a simple prompt today, making automated content creation easier than ever for anyone with internet access.
More troubling still is the economic trap tightening around human content creators. AI systems are trained on—or in blunter terms, steal from—existing websites, extracting value without compensation. This undermines the advertising revenue model that has traditionally sustained online publishing, removing the financial incentive for producing original content. Why invest time and resources in creating new material when AI can replicate and repurpose it instantly, siphoning away traffic and ad revenue in the process?
This creates a vicious cycle: as content creation becomes economically unsustainable, fewer people will produce original work. The internet risks becoming an echo chamber where AI endlessly remixes existing material while genuine human insight and creativity atrophy from lack of support.
We urgently need solutions that reward original human content creation—whether through new compensation models, stronger intellectual property protections, or platforms that prioritize and authenticate human-generated work. Without intervention, the current plateau may not represent stability, but rather the calm before a more dramatic shift toward an internet dominated by derivative, machine-generated material. The question isn’t whether things will change, but whether we’ll act before that change becomes irreversible.
