Skip to content
AI-Assisted Content5 min read

A Third of the Web Is Now AI-Written. Google Just Built a Trust Layer That Rewards Publishers Who Aren’t Noise

The Number Pew Dropped

August 20, 2026. Pew Research Center published a study that ran 490,000 English-language web pages through an AI detection model. The headline figure is 10% — one in ten pages across the full sample shows significant signs of AI authorship. But that figure spans the entire modern web, including content created before ChatGPT existed.

Filter to pages published after November 2022 and the number changes fast. Among web pages created in the post-ChatGPT era, more than one in three — 35% — show those signals. A separate study from Imperial College London, Internet Archive, and Stanford landed at the same figure independently.

Commercial domains carry most of it. Around one in ten .com pages register AI authorship signs — double the rate on .org domains and ten times the rate on .edu or .gov. Pew also mapped the linguistic tells: em-dash usage climbed from 5.79 to 11.19 per 10,000 words between early 2023 and early 2026. Words like "delve," "interplay," and "testament" more than doubled in frequency. Negative parallelism — the "it's not just X, it's Y" structure — nearly tripled.

That's a map of where undifferentiated AI output concentrates. And it explains why Google moved when it did.

What Google Launched the Same Week

Also August 20: Google announced a major upgrade to its Preferred Sources feature. Publishers can now embed an interactive button directly on their WordPress sites and article pages. A reader who clicks it marks that site as a trusted source across Google Search, AI Overviews, AI Mode, Discover, and Google News. Future searches on relevant topics from that reader will surface the preferred publisher's content more prominently — including inside Google's AI-generated answers.

The feature launched in May 2026 as a within-search selection tool. The August upgrade made it embeddable anywhere. By the time the update went live, more than 600,000 unique sources had been selected by readers, up from 345,000 at the May launch. Google's own data puts click-through likelihood at twice as high for preferred sources compared to unselected ones.

For a WordPress publisher running scheduled news content, that button is not a cosmetic addition. It is a direct mechanism for readers to tell Google's AI systems which sites to prioritize when assembling answers.

The Top Stories Shift Inside AI Overviews

The Preferred Sources button connects to a structural change in how Google renders news in AI search that John Shehata at NewzDash began documenting in July. Google has started embedding full Top Stories carousels inside AI Overviews — not below them, inside them, above the fold.

NewzDash's tracking across trending news queries: 15.5% of US searches and 17.46% of UK searches now show the Top Stories carousel embedded within the AI Overview rather than as a standalone module underneath it. Entertainment queries lead at over 35% in the US. World News sits near 32%.

The two placements are mutually exclusive. When Top Stories appears inside the AI Overview, Google does not also run a standalone carousel lower on the same results page. That distinction matters more than it sounds: publisher links inside the AI Overview sit higher on the page than they have in any traditional layout. It is the most prominent organic news placement Google has offered publishers in years.

There is a catch worth understanding. Most publishers who want out of AI Overviews reach for robots.txt and block Google-Extended. That does not remove them from AI Overviews. That control only governs model training and Vertex AI grounding. The relevant control is the Search Console generative AI exclusion — and choosing that exclusion likely removes the publisher from the embedded carousel, with no guarantee Google will show the standalone version on the same results page for the same query.

Opt out of AI features to prevent your content from being summarized, and you may exit the best organic news surface Google has built. That trade-off is worth making deliberately, not accidentally.

What the Preferred Sources Button Actually Requires

The button only produces a return if readers have a reason to click it. Google cannot mark your site as preferred for a reader who has never arrived there, and no reader will proactively trust a publisher they cannot name from memory.

This is where publishing cadence becomes strategic rather than operational. The Preferred Sources button is a loyalty mechanism. It rewards consistent, recognizable publishing across a specific topic area — the kind of presence that makes a reader think "this site covers this" before they consciously decide to click. One strong article per week, on a predictable schedule, in a defined niche builds that recognition. Irregular publishing does not.

Google's Top Stories eligibility already rewards sites that cover subject areas consistently. The Highly Cited label, freshness signals, and developing-topic gates that route content into the embedded carousel are downstream of sustained output — not traffic spikes or one-off viral moments.

The Case the Pew Data Actually Makes

The 35% figure is not an argument against using AI to assist content creation. Plenty of those pages are legitimate, well-edited, useful content that happens to carry AI authorship signals. The argument the Pew data makes is narrower: at this volume, the signal that distinguishes useful from noise is the human editorial layer — the brand voice, the factual verification, the original angle, the specific expertise a team adds before publishing.

Google's Preferred Sources button is the infrastructure that lets readers vote on which side of that line a given publisher stands on. The Top Stories carousel embedded in AI Overviews is the payoff for the publishers readers choose.

For content teams using AI-assisted workflows, the operational question is not whether to publish frequently. The web has already answered that — 35% of new pages are AI-written and the number climbs every crawl. The question is whether the human review step is adding enough to cross the threshold from noise to signal. That threshold is now encoded in a button your readers can click and a carousel placement your readers will see.

Sources: