Have Bots Taken Over Online Writing, or Are We Still Reading Human Work Without Knowing It?

The online world is quietly shifting in ways most readers barely notice. A new analysis from Graphite, an SEO research company, shows that artificial intelligence now produces more than half of all the articles found on the web. The finding, based on 65,000 English-language pages published between early 2020 and mid-2025, marks one of the most dramatic changes in digital publishing since the early blogging era.

Graphite’s team traced^[1] the rise back to late 2022, the moment ChatGPT appeared. Within twelve months, the number of AI-written articles had climbed to nearly half of all online output, and by November 2024, automated content overtook human work entirely. What began as a quick way to fill gaps in websites or boost traffic has become a dominant form of digital writing.

The study shows that growth began to slow around May 2024, when the balance between human and AI-produced text leveled off. Some months since then have seen human output slightly ahead, but overall, the two remain close. Researchers did not pinpoint why the surge tapered, though one likely reason is that machine-written pages do not attract much attention from search engines or readers. Graphite’s related findings suggest that most AI-generated material is buried deep in Google results or rarely appears in chatbot summaries. That lack of visibility may have curbed enthusiasm among publishers who once relied on AI to boost their rankings.

To measure how much content came from machines, Graphite analyzed articles drawn from Common Crawl, a vast public web archive. Each text was divided into 500-word segments and assessed with Surfer, an AI-detection system that labels an article as machine-written if more than half of its content appears algorithmic. Before running the full dataset, the researchers tested the detector’s accuracy. They checked nearly sixteen thousand pieces published before ChatGPT’s release, assuming these were human-written, and found only about four percent misclassified. They then generated more than six thousand trial articles with OpenAI’s GPT-4o model, and the software correctly identified almost all of them as AI.

Even with those checks, Graphite acknowledged that detecting AI remains unreliable. Models are improving so quickly that the difference between human and machine expression is narrowing. Many writers also use hybrid methods, letting AI draft first and then revising manually. Those mixed pieces are difficult to categorize, but they likely make up a growing share of online text.

The dominance of synthetic writing doesn’t necessarily mean that quality has declined. In some studies, including one from MIT^[2], people who read AI and human work without knowing the source rated machine-written pieces as more polished and better organized. Yet those results raise another question about what readers actually value: fluency or originality. AI-generated text draws on huge databases that include earlier online material, so every time a new model writes, it is in part reusing fragments of the web that came before. Over time, that cycle could dilute the originality of the internet itself.

Still, automation offers a tempting advantage for publishers who need a constant stream of updates or product reviews. Producing hundreds of articles in a day costs little compared with hiring writers, and as long as the content passes basic checks, many sites are content to keep feeding it into the system.

The bigger concern lies in what happens next. If half the words on the internet already come from machines, and future models learn by reading that same content, the boundary between real and generated information may disappear altogether. At that point, the web would no longer reflect what people know or think… it would echo what algorithms predict they might say.

For now, the world’s online writing sits at an uneasy halfway point. Humans and machines produce roughly the same volume, but for very different reasons. One writes to communicate. The other writes to generate more of itself.

Notes: This post was edited/created using GenAI tools. Image: DIW-Aigen.

Read next:

• WhatsApp Blocks AI Chatbots to Protect Its Business Platform^[3]

• Wikipedia Faces Drop in Human Traffic as AI and Social Video Change Search Habits^[4]

• When AI Feels Like a Friend: Study Finds Attachment Anxiety Linked to Emotional Reliance on Chatbots^[5]