- A Pew Research Center study found that roughly a third of new English-language webpages published since November 2022 are AI-authored, corroborated by separate work from Stanford University, Imperial College London and the Internet Archive.
- Commercial websites adopt AI-generated content at roughly 10 times the rate of educational or government domains, yet most of that output ranks poorly in search results, per a May 2026 Graphite study.
- As AI-generated text accumulates on the open web and re-enters training datasets, researchers warn of “model collapse”, a feedback loop that compresses originality across successive model generations.
A third of all new English-language webpages published since November 2022 are AI-authored, according to a new Pew Research Center study released this week. Stanford University, Imperial College London and the Internet Archive had independently classified a similar share of newly published web content as AI-generated or AI-assisted. The scale of that shift carries direct consequences for SEO strategy, content verification and the training data that future models will consume.
Search Rankings Under Pressure
Commercial websites, identified primarily by.com domains, are adopting AI content at roughly 10 times the rate of educational or government sites, according to the Pew report. The efficiency logic is straightforward: more content at lower cost, but volume is not translating into visibility. A May 2026 study by research firm Graphite found that most machine-written output published since late 2024 remains largely unseen in search results.
Researchers at Stanford and Imperial College London, in a paper examining AI-generated text’s impact on the internet, raised a related concern: mass AI content adoption may be degrading semantic diversity online, with reported increases in positive sentiment homogeneity across publishing categories. That homogeneity is the kind of quality deficit that search algorithms are increasingly designed to penalise.
Detection and the Verification Gap
AI content analysis firm Copyleaks detected a significant increase in AI-generated web content between November 2022 and March 2024 across a sample of one million analysed pages. The Stanford research, contrary to some expectations, found no measurable increase in verifiably false statements directly attributable to AI-generated content during that period. The clearer signal was semantic, not stylistic: AI-driven content growth correlated with reduced semantic diversity and a rise in uniformly positive sentiment across newly published material.
That homogenisation has a second-order consequence that trust-and-safety and content governance teams are beginning to flag. As AI-generated text accumulates on the open web, it enters the training datasets used to build future models. If left unmanaged, the result is a feedback loop where models train on their own prior output, compressing originality over successive generations, what researchers call “model collapse.” How quickly this degrades output quality in practice remains an open empirical question, but the structural risk is real.
Infrastructure Costs and Financial Exposure
The content volume driving these concerns is itself a function of infrastructure investment. Alphabet, Microsoft, Meta, Amazon and Oracle are all running multibillion-dollar data centre build-outs to sustain generative AI capacity, the demand side of which is, in part, the content production surge the Pew study documents. That spending has pushed equity valuations in the sector toward historical peaks. The European Central Bank, in a statement dated August 17, 2026, warned that AI-driven stock valuations face a high probability of correction.
The financial exposure runs deeper than equity prices. Hyperscalers are increasingly funding infrastructure expansion through debt, and returns remain unevenly distributed. For enterprises weighing their own AI content investment decisions against this backdrop, the data centre risk picture is worth tracking alongside content strategy. The Pew finding that a third of the web’s recent additions are already machine-authored did not arrive alongside a commensurate monetisation story. That gap is where the financial risk concentrates.
