How Google's AI Search Is Reshaping Online Publishing

For about twenty years, the web ran on a simple bargain. Publishers let Google crawl their pages, Google sent readers back, and those readers saw ads or bought subscriptions. It was never a perfect deal, but it was understood. Now AI-generated answers sit at the top of the results page, and that bargain is wobbling.

I'm going to lay out what the available evidence says about the change: how big the traffic losses are, who seems to be protected, what Google says in response, and what it all means for those of us who read, and pay for, online information. One caveat up front: much of the data below dates from 2024 and 2025, so treat it as a snapshot of a shift that is still moving.

The old bargain and what changed

According to a roundup by Search Engine Journal, search, including Google Discover, typically accounts for 20% to 40% of referral traffic to major publishers. That makes it their largest external source of readers. When one channel supplies that much of your audience, its design choices become your business risk.

Google's AI Overviews grew out of the Search Generative Experience announced in 2023. Following a May 2025 expansion, the same article says the feature appears in over 200 countries and 40 languages. It calls AI Overviews the biggest change to search since featured snippets were introduced in 2014.

The difference from a featured snippet is scale and ambition. A snippet quoted a line from one page. An AI Overview synthesizes an answer from several sources and presents it as a finished product. For many questions, the user simply stops there.

What the numbers say

Several independent studies point in the same direction, even if they disagree on size. Pew Research Center tracked 68,000 real search queries and found that users clicked a result 8% of the time when an AI summary appeared, versus 15% when it did not. That is a relative drop of roughly 46.7%. Ahrefs, looking at informational keywords, estimated that AI Overviews reduce clicks by 34.5%. Both figures come from the Search Engine Journal compilation.

Similarweb data cited there says zero-click searches, where the user never leaves the results page, rose from 56% to 69% between May 2024 and May 2025. That figure captures more than AI Overviews, but the timing lines up with the rollout.

The worst case comes from DMG Media, owner of MailOnline and Metro. In a submission to the UK's Competition and Markets Authority, it reported click-through rates falling by as much as 89% when an AI Overview appeared. The article notes that the Daily Mail's desktop click rate fell from 25.23% to 2.79% in that scenario. DMG's number is a worst case, not an average, and it's important to say so. But a worst case of nearly 90% is the kind of figure that makes a finance director lose sleep.

Not everyone saw such steep falls. Digital Content Next found a 10% decline in overall search traffic among its member publishers between May and June of 2025. That is far milder, but for a company running on thin margins, a 10% loss in its biggest channel is still a serious hit.

Why the damage is uneven

The pain is not spread evenly across the web. Weather and stock prices are classic zero-click queries, and honestly, nobody needs a 1,200-word article to learn the temperature. The problem starts when AI answers creep into areas where publishers built real depth, the in-depth content that justified a click in the first place.

Educational sites are an example. Chegg reported a 49% drop in non-subscriber traffic between January 2024 and January 2025, and sued Google in February 2025, alleging that content from educational publishers was used to train AI systems that now compete with them. Google has its own view of such claims, and I'm only reporting what the lawsuit alleges, not how it will end.

There is also a visibility issue that has nothing to do with clicking. Advanced Web Ranking examined 8,000 keywords and found that an expanded AI Overview averages about 169 words and roughly seven links. Once it's open, the first traditional result can sit around 1,674 pixels down the page. A page that ranks first can end up below the fold.

Then there's volatility. An Authoritas report found that about 70% of the pages cited in AI Overviews changed over two to three months, and those changes were not tied to traditional rankings. So even being cited is a moving target. A site can feel stable in classic rankings and still lose traffic overnight.

Brands hold up better than everyone else

One of the more striking findings is that branded searches seem to behave differently. Amsive research cited in the same roundup found that branded queries with an AI Overview saw an 18% increase in click-through rate, while generic queries lost 34% to 46% in the studies above.

This makes intuitive sense. If someone searches for a company by name and the AI summary mentions it, the summary works almost like an endorsement. The user clicks through to the official site. Generic information, though, is exactly what a summary can replace.

My own view is that this creates a worrying two-tier web. Established brands with loyal audiences cope. Smaller independent sites, which often produce the specialized, obsessive and useful content that makes the internet interesting, look like the ones most exposed. That's a cost to everyone, not just to publishers.

Google's side of the story

Google rejects the idea that it is starving the web. As summarized in the article, its defense has three parts: usage is up, clicks are higher quality, and overall traffic is stable. In a May 2025 post introducing the global expansion, Google said AI Overviews were driving over a 10% increase in usage for the types of queries that show them. In an August post, Liz Reid, Google's head of search, said click volume to websites was relatively stable year over year and that people were seeing more links on the page.

A Google spokesperson also told the BBC that the company sends billions of clicks to websites every day. And Google questioned Pew's methodology, arguing the study period overlapped with unrelated algorithm testing.

Are these claims compatible with the independent data? Possibly, in a narrow sense. More queries and more links can coexist with fewer clicks per query for certain content types. Google's 10% figure applies only to queries that trigger AI Overviews, not to search overall. And "stable" in aggregate can hide big winners and losers. The roundup itself points out that Google hasn't provided specific metrics defining "higher quality," so publishers can't verify the claim. That gap is part of why trust is low.

The wider fight over content and AI

Search is only one front. Publishers are also fighting over how their text is used to train and feed AI systems in the first place. We recently covered how Meta dropped terms that could have let it scrape news sites for AI, which shows how sensitive this topic has become. And readers who want to see how different AI search products handle sources can look at our comparison of Perplexity and Google AI Mode.

The legal and regulatory route is already open. According to the roundup, in July 2025 the Independent Publishers Alliance, Foxglove and the Movement for an Open Web filed a complaint with the UK's Competition and Markets Authority. They argue that AI Overviews misuse publisher content, and they want temporary measures stopping Google from using it in AI answers without compensation. The article also warns that courts and regulators move slowly, and market realities may outrun them.

The underlying question is the same everywhere: if AI systems learn from journalism and then answer questions so well that people no longer visit the journalism, who pays for the next article? Nobody has a clean answer yet. The article mentions content licensing as one emerging path, citing partnerships between OpenAI and News Corp and The Atlantic.

What publishers are doing, and what I would watch

The roundup describes publishers rushing to cut their reliance on Google. David Higgerson of Reach told the BBC that the company needs to find where audiences are elsewhere and build relationships there, mentioning WhatsApp alerts reaching millions of people and newsletters. The article adds that email newsletters, mobile apps and podcast subscriptions are traffic sources not affected by AI Overview disruption. Stuart Forrest stressed quality, with the aim of being the source that gets cited rather than rivals. The article cautions, though, that quality alone may not be enough if users never leave the results page.

None of that is easy for a small outlet. Newsletters and apps are a sensible hedge, but in my view they are not an instant replacement for a channel that supplies 20% to 40% of referral traffic at major publishers.

For the next year I would keep an eye on three things. First, whether Google shares transparent, verifiable click data. Second, whether regulators, such as the UK's Competition and Markets Authority, force changes in how publisher content can be used in AI answers. Third, whether users start to notice that answers get thinner or less reliable as the sources behind them lose funding. That last one is my own worry, and a slow-burning one: AI answers depend on a healthy supply of human-made reporting and expertise.

A question for your own habits

It's easy to treat this as an industry story, but it lands in your daily routine. Think about the last thing you searched for. Did you click through to a source, or did you read the summary and move on? Did you check who wrote it? Convenience is real, and I use these tools too. But every habit is a small vote for what kind of web survives.

So here's my invitation: next time an AI answer satisfies you, consider paying attention to the sources it cites, and maybe support one outlet directly. The future of the information we all rely on may depend on whether enough of us decide that the original is still worth a click.