The web has passed a milestone: more than half of the pages we see on the internet are now generated by algorithms. In a new analysis from the SEO boutique Graphite, which analyzed 65,000 URLs pulled from Common Crawl web data, AI-generated text is now tied in the race with human-written text and holding its own.
The finding won’t shock anyone who has been watching the post-chatbot boom, but the turning point is important. It hints that the default format for a lot of online writing is no longer human prose assisted by computation — not even text written by a machine with human help.
What the New Report Reveals About AI vs Human Writing
Graphite’s team ran passages from the articles through Surfer, a public AI-content classifier, and flagged an entire piece as HUMAN-NONHUMAN when it was determined that 50% or more of the text would be machine-written.
The firm also reports a leveling in the growth of AI articles after an initial spike, and some months with almost as many human- and AI-authored articles.
This plateau is a head-scratcher considering the growing user base for mainstream chatbots — OpenAI says it has hundreds of millions of weekly active users — and the existence of tools that can spit out articles in minutes. One theory: Platforms and publishers are quietly fiddling with workflows to stamp out the obvious “AI slop,” while search engines are getting better at squelching low-value stable of pages that no satisfied user searches for.
The Detection Caveat: Why AI Classifiers Can Mislead
AI identification is an imperfect science. Graphite recognizes the potential for false positives and stress-tested Surfer by creating 6,009 synthetic articles with a best-in-class model; the tool correctly called out 99.4% of those fakes. In clean lab conditions, that’s an impressive rate, but real-world content is messier, mixing human edits with templates and often sprinkled with boilerplate text that can trip up detectors.
Dozens of academic researchers and assessment vendors have warned that AI detectors will misclassify well-structured human writing or non-native English, and models change too fast for classifiers to keep up. So consider the 50% number to be directional, not absolute. The overall trajectory — more machine-written text — is hard to argue with.
Who Is Actually Reading AI-Generated Articles Today
One separate study referenced by Graphite said AI-penned articles are far less likely to rank high in Google search results or chatbot answers than those written by a human. That squares with what a lot of SEOs are seeing: there are huge numbers of low-value, AI-generated pages out there, but only a fraction are actually eaten up by readers at scale.
Meanwhile, watchdogs like NewsGuard has recorded thousands of AI-generated news and information sites churning out high volumes of shallow, error-filled content. The tactic is simple: inundate the index, pull in long-tail queries, and hope ad tech pays no notice. The danger is just as clear: If search and social recommendation systems swallow that stuff up alongside everything else, it’s possible that the same AI tools will re-amplify the very disinformation from which they tried to wean themselves, circulating a feedback loop of junk.
Why the Web Is Awash in Machine-Generated Text
It all comes down to speed and cost. Generative models can churn out passable drafts in an instant, and newer systems, such as OpenAI’s Sora 2, have expanded that automation to video and multimodal content. For content farms, affiliate sites and programmatic publishers, the marginal cost of another article is nearly zero.
Respectable newsrooms are experimenting, too, but there are guardrails for them. The Reuters Institute has recorded cautious uptake among publishers: AI will assist with summaries, headlines and translations, and in extracting data, while human editors will have control of reporting and accountability. And public sentiment is skeptical — surveys by the Pew Research Center have found many readers uncomfortable with news generated using AI, particularly on sensitive subjects.
Search and Platforms Are Scrambling to Curb AI Spam
Search engines say they rank based on usefulness, not authorship, but the explosion of AI-written pages has certainly put some strain on those algorithms to continually roll out spam and quality updates. Google’s AI Overviews and chatbot responses from large models can accidentally spew low-quality text if checks fail. More and more, platforms are relying on signals of originality, engagement quality and publisher identity to sift through the deluge.
There is more reliance on provenance standards like C2PA content credentials, as well as on-source signals (e.g., bylines, disclosures and detailed author pages).
Those markers aren’t proof of humanity, but they can differentiate accountable publishers from anonymous mills.
How to Operate Effectively in the AI Slop Era
For publishers and brands, the playbook is moving toward depth and verifiability. And give as much priority as possible to reporting and analysis that generic models can’t replicate. Demonstrate your work with citations, original charts and transparent methodology. Audit any AI-facilitated workflows, prohibit fully automated publication and disclose machine support where employed.
On the technical stuff: curb index bloat; merge similar pages, set canonicals and fix orphans — measure what truly earns engagement. Clean up sitemap hygiene, and pay attention to crawl budgets so low-value programmatic pages do not obscure your great work.
The takeaway is brutal but actionable. AI text exceeds human output while attention is remarkably scarce and reputation becomes stronger. The winners will be the companies and teams that leverage AI without sacrificing editorial judgment — and can prove it to both algorithms and audiences.
