A new Pew study shows that Google’s AI Overviews are siphoning traffic from websites. If smaller publishers can’t survive, the internet and AI itself could pay the price.
If you’ve used Google in the past year, you’ve likely encountered its AI Overviews: brief, AI-generated summaries that appear at the top of the search page. These AI summaries may seem mundane, but a new Pew analysis of nearly 69,000 Google searches shows they’re quietly reshaping how people use the internet. So let’s dive into the results.
AI Overviews are killing click-through rates. When no summary was shown, 15% of users clicked on a traditional link. That share dropped to just 8% when an AI Overview appeared, and only 1% clicked on a link inside the summary itself. A different study found that the top organic link can lose as much as 79% of its traffic if it sits below an Overview.
The ripple effects don’t stop there. 16% of users ended their browsing session when no summary appeared, but that jumped to 26% when an AI Overview was present.

How often do AI Overviews appear? In Pew’s sample, 18% of all searches triggered a summary. Longer, conversational queries were far more likely to do so: 1‑2 word searches produced a summary only 8% of the time, while 10+ word searches did so 53% of the time. Overviews appeared in 36% of full‑sentence searches and in 60% of queries that started with an interrogative or question word.

For years, the economics of the internet has rested on a three-way exchange of value: Users gain by finding something they’re looking for and clicking through to that publisher’s site; search engines like Google gain by selling ads to firms that want to reach users; and web publishers gain by all the traffic they receive. This traffic fuels ad revenue, subscriptions, and discovery. But AI Overviews interrupt that flow, often satisfying users before they ever leave Google’s page. Web publishers are thus dropped from the exchange. Google may see the bypassing of publishers as efficiency, but for content creators, it’s an existential threat. It means no traffic and no income. (Google no doubt has its own existential worries, since it may fear that without AI Overviews users will go directly to ChatGPT or DeepSeek and abandon Google entirely.)
Worse still, web publishers can’t opt out. Any attempt to block Google from scraping their content for AI Overviews also removes them from the company’s regular search results. On the other hand, Google could use this arm-twisting to its advantage in the courtroom. Since publishers do indeed have the “choice” to opt out—in return for being invisible—Google could argue that publishers are tacitly agreeing to this new system. This may give Google a massive legal advantage over rivals, like ChatGPT, which hoover up everybody’s content without any publisher opt-out option and, for that reason, are tied up in costly copyright battles .
IMO, this system is destined to implode. Even before AI Overviews, media publishers were complaining that Google was shrinking their revenue by profiting unfairly from their content. In the age of AI, that dynamic is likely to accelerate. Yet if publishers start collapsing under reduced traffic and strangulated ad revenue, Google, along with all the other AI giants, will have fewer and poorer sources to pull from. As the web fills with AI‑generated content, the quality of those summaries will degrade. Ultimately, when AI begins training mostly on its own recycled output, the result will be gibberish. AI scientists call this “model collapse.”
Model collapse is what will happen when the AI machine (the parasite) kills most of the human creators and publishers who actually experience and think about the world (the host). And once the host dies, so does the parasite.