Pew Study: AI-Written Text Surges Across Web Since ChatGPT

What happened

Pew Research Center has published an analysis showing that AI-generated text has become widespread across the English-language web since the launch of ChatGPT in late 2022. Examining close to half a million web pages, researchers found that more than a third of those published after ChatGPT's debut carry detectable signs of machine-generated content.

The study also found a marked skew by domain type: commercial ".com" sites are roughly ten times more likely to contain AI-written text than educational or government sites, pointing to differing adoption patterns between commercial publishers and public institutions.

Why it matters

This is one of the first large-scale empirical measurements confirming what many in content, marketing and search have anecdotally observed: generative AI has moved from novelty to default production method for large volumes of public web content in under three years. For organisations building on AI, it signals that the training and reference data available on the open web is itself increasingly AI-derived — a feedback loop with implications for model quality, information reliability and search/answer-engine outputs.

For leaders in digital transformation and experience design, the finding is a prompt to reassess how content strategy, brand voice and trust signals are managed. If a large share of the open web is now machine-authored, differentiation increasingly depends on demonstrable expertise, sourcing and human oversight — not just publishing volume.

By the numbers

• Nearly half a million English-language web pages were analysed by Pew Research Center.
• More than a third of pages published since ChatGPT's late-2022 launch show signs of AI-generated text.
• Ten times — the rate at which .com domains are more likely to contain AI-written content compared with .edu or .gov sites.

The Renascence take

The headline number is striking, but the more useful signal is the domain split: commercial publishers are adopting generative text far faster than public institutions, which suggests a widening gap in how organisations balance speed against accountability.

Most commentary on this study will focus on the volume of AI text online; the sharper question is what it does to trust. When readers and answer engines can no longer easily tell human expertise from machine output, provenance becomes the new differentiator — not polish. Customer-obsessed organisations should treat visible authorship, sourcing and human review as a service-design decision, not a compliance afterthought, because in a web full of synthetic text, verifiable human judgement is what earns attention and belief.