Pew: a third of new web pages show AI authorship

Pew Research Center published a data essay today measuring how much of the web is written by machines. In a random sample of 10,000 English-language pages collected in July 2026, about one in ten showed significant signs of AI authorship. Restrict the sample to pages published after ChatGPT shipped in November 2022 and the number goes past one-third. If you sell anything online, your content is now competing in a pool that is mostly synthetic.

What actually happened

Pew's analysis ran nearly 490,000 English-language pages from the Common Crawl archive, January 2021 through July 2026, through Open Pangram — a classifier that scores text on aggregate linguistic patterns rather than any single tell.

The split by top-level domain is the useful part:

• .com — roughly 10% of pages show AI authorship
• .org — 4.6%
• .edu and .gov — roughly 1% each

Commercial domains run about ten times the rate of educational and government ones. Nobody should be surprised. Commercial pages are the ones with a volume incentive.

Pew also tracked stylistic drift from 2023 to 2026: em dashes doubled, Oxford commas rose 63%, AI-favored vocabulary more than doubled, and "negative parallelism" — the it is not X, it is Y construction — nearly tripled. Pew is explicit that no single marker proves anything. Detection works statistically across large samples, not on one page. TechCrunch's write-up adds the obvious caveat: classifiers misfire, and this is directional, not a census.

Why AI authorship at this scale matters for your business

We use AI to draft. So does everyone reading this. The finding is not "stop using the tools." It is that the cost of producing average text went to zero, so average text stopped being an asset.

Volume is no longer a moat. If your content plan is thirty posts a month, you built a plan any competitor can copy this afternoon for the price of tokens. The scarce inputs are the things a model cannot generate: your actual numbers, your actual customers, the specific thing that broke in production last Tuesday.

Watch the stylistic tells, but do not chase them. Stripping every em dash from your site is cargo cult. Detectors score aggregate patterns and they get individual pages wrong in both directions. Write in your own register and let it be uneven.

Assume search evolves around this. When a third of new commercial pages read as machine-written, ranking systems have every incentive to weight signals that are expensive to fake — original data, real bylines, links from people who know you.

Sources: Pew Research Center — How Much of the Internet Is Written With AI?, TechCrunch.