The fear everyone has, and why the data doesn't actually support it
A domain drops in rankings. Someone on the team used AI to help draft a few pages last quarter. The two facts get connected instantly, and the panic that follows sends people looking for AI detection tools instead of asking what actually changed. It's an understandable instinct. It's also not what actually happens.
AI-written content is not systematically penalized, and you don't need a proprietary dataset to see it. It sits everywhere in search results right now, including at the top of competitive keywords. Google has said plainly, repeatedly, and publicly that it does not detect or penalize AI authorship as a category. What it penalizes is content, regardless of who or what wrote it, published at scale specifically to manipulate rankings rather than help the person reading it.
What actually happened in March
Google rolled out a core update starting March 27, 2026, days after a separate spam update completed in under 20 hours. The core update introduced what's widely being called an Information Gain evaluation, a way of measuring whether a page tells a reader something genuinely new compared to what already ranks for the same question, or whether it just rephrases the same five answers everyone else already gave.
Google has been consistent, publicly and repeatedly, on the actual target: what it calls scaled content abuse, publishing large volumes of pages primarily to manipulate rankings rather than to genuinely help a reader, regardless of whether a human or a model produced them. AI didn't create that problem. AI just made it dramatically cheaper to do at a scale that used to require a content farm's entire staff.
The actual pattern, without borrowed precision
Independent SEO trackers and site owners have described a consistent shape to what happened, even without one authoritative dataset to point to for exact numbers. Sites that leaned on AI to mass-produce large volumes of unedited content saw real, often severe, traffic losses. Sites that used AI to accelerate drafting, then had a knowledgeable person shape, fact-check, and add something the model couldn't have known on its own, held their ground or gained it.
The split isn't about the tool. It's about whether a human being was actually accountable for what got published. The sites that gained ground used AI as a drafting accelerant. The sites that lost ground used it as a mass-production machine and skipped the part where someone checks the output actually says something.
The new signal: information gain
For years, publishing more was itself treated as a strategy. More pages, more keywords covered, more surface area for something to rank. That instinct is precisely what's now triggering the penalty. If your page doesn't tell a reader anything they couldn't already get from the top few results, it's no longer neutral filler sitting harmlessly on your site. It's now an active liability dragging down how Google evaluates everything else you publish.
Depth beats breadth, and Google is now checking at the domain level too
One of the quieter but more important shifts in this update: Google is increasingly evaluating quality signals at the domain level, not purely page by page. A site that consistently publishes real depth on one coherent topic area is now outperforming a site that spreads thin across dozens of unrelated ones, even if some individual pages on the broader site are decent in isolation.
That's not a new idea dressed up as an algorithm update. It's the same logic behind going deep on one real, well-researched topic instead of scattering content across everything that might conceivably get a click, the entire premise behind how we structure content for clients in the first place. Google didn't invent this principle. It just started actually enforcing it.
The boring stuff still matters, maybe more than ever
Author credentials are carrying more weight than they used to, particularly on money, health, legal, or education content, where Google's own quality rater guidelines have emphasized demonstrated expertise and experience for years. Technical performance matters too: pages that load fast and don't shift around while loading are holding ground against slower competitors in the same niche.
None of this is exotic. It's the same set of practices that made a site genuinely good five years ago: know who wrote it, make it fast, say something real. The difference is that volume alone used to be enough to paper over weak execution on all three. It isn't anymore, and AI didn't cause that shift, it just made it obvious faster.