If the internet has started to feel strangely repetitive lately — the same phrasing, the same listicle structure, the same oddly confident tone on topics nobody seems to have firsthand experience with — that instinct isn’t wrong. A growing body of independent research puts a real number on it, and it’s higher than most people would guess.
The Headline Number: 74%
Ahrefs analyzed nearly a million new web pages published in a single month and found that 74.2% contained detectable AI-generated content. That doesn’t mean three-quarters of the internet was written entirely by a machine with zero human involvement — most of that content is a human-AI blend, with only about 2.5% of new pages classified as “pure AI” with no editing at all. Still, a three-quarters figure for any level of AI involvement marks a genuine shift in how the web gets written, and it happened faster than most industries anticipated.
Why the Number Jumped So Fast
The cost of producing content with AI assistance has fallen sharply — one industry analysis put the average cost of a 2,000-word article at roughly $268 with AI tools involved, down from around $480 without them. That’s a 44% drop in production cost, and it’s exactly the kind of economic pressure that explains rapid, industry-wide adoption faster than any single company’s marketing could. If you’re evaluating how ChatGPT and similar tools can actually be used in a business, content production is one of the clearest, most immediate use cases — and one of the first places most companies actually saw a return.
This Isn’t Just Text
The shift extends well beyond written articles. On Instagram, TikTok, and Pinterest specifically, an estimated 79% of newly posted images now show detectable signs of AI generation. TikTok has started auto-labeling AI-generated video using industry-standard content credentials, though its own detection currently catches only 35-45% of AI content, meaning the true share circulating on the platform is likely higher than what’s officially labeled.
The Part That Should Worry Readers More Than Marketers
A separate and more concerning trend sits underneath the broader AI-content statistic: dedicated “AI content farm” websites that exist purely to republish AI-rewritten news and product content at scale for ad revenue, with little to no editorial oversight or fact-checking. NewsGuard’s tracker had identified 3,006 of these sites operating across 16 languages as of 2026 — up sharply from just a few hundred when the tracker first launched in 2023. These sites are functionally different from a business blog or news outlet using AI as a writing aid; they’re built specifically to exploit search and social algorithms rather than inform anyone.
Does More AI Content Actually Mean Worse Search Results?
Not necessarily, and this is the part most alarmist headlines skip. Ahrefs also found a near-zero statistical correlation between a page containing AI content and that page facing a ranking penalty, across a dataset of 600,000 pages. What matters far more than whether AI was involved in drafting is whether the final content is actually accurate, genuinely useful, and edited by someone who knows the subject — the same standard that applied to human-written content before AI tools existed at all. Businesses trying to dominate search rankings as a small business are learning this distinction quickly: publishing more content faster only helps if the content clears the same quality bar readers and search engines always expected.
Why Only a Fraction of AI Content Actually Ranks Well
Despite AI-written material now making up roughly half of everything newly published online by some measures, only about 14% of the articles that actually rank on Google’s first page are AI-generated — and at the very top positions, that share is smaller still. The gap between how much AI content exists and how much of it succeeds is really the whole story: AI dramatically lowered the cost of producing content, but it didn’t lower the bar readers and search algorithms use to decide what’s worth surfacing.
Why Detection Tools Struggle to Keep Up
AI writing models are updated constantly, and detection tools trained on older model outputs often lag behind the newest generation of AI writing, which tends to read more naturally and triggers fewer detection flags. This creates an ongoing arms race rather than a solved problem — treat any AI-detection score as a rough signal, not a definitive verdict, especially for content produced with the most recent tools.
What This Means If You Run a Business Website
The practical lesson isn’t “avoid AI tools” — it’s “don’t mistake AI assistance for a finished product.” Companies using AI purely for research, outlines, and data analysis while keeping actual drafting and editorial judgment human-led report the highest satisfaction scores with these tools, alongside the strongest content performance. This is also exactly where [CLIENT LINK PLACEHOLDER] fits for businesses trying to scale content production without falling into the content-farm trap — the tools that succeed are the ones built around editorial review, not around removing it entirely.
