Sites Cited in AI Overviews Pass Core Web Vitals 56.8% of the Time vs 40% Web-Wide

As AI Overviews take over the top of Google's results, every SEO team is asking the same question: what does it take to get cited? We took a different cut at it. We pulled every web page the AI Overview cited across a spread of common questions and measured its real-world Core Web Vitals, the field data Google uses to grade page experience. The result is reassuring for anyone who cares about speed. The pages an answer engine chooses to put in front of people pass Core Web Vitals more often than the web at large.

That matters because AI Overviews are becoming the first thing a searcher sees, and the handful of sources they cite are the pages that get the visibility. If those pages were slow and unstable, it would mean the machine deciding what billions of people read had no regard for the experience waiting on the other side of the link. The data says otherwise.

We captured the pages Google's AI Overview cited across 13 US informational queries spanning health, money, and general knowledge, then pulled the distinct registrable domains behind those citations. That gave us 77 cited domains, of which 74 had Chrome User Experience Report field data to measure. Each domain is assessed on its real users' 75th-percentile mobile experience, and it counts as passing only when loading, responsiveness, and stability are all good under Google's strict thresholds. YouTube video citations, which appear on nearly every query, were logged separately and left out of the page-speed count.

Of the 74 cited domains we could measure, 42 pass Core Web Vitals. That is 56.8%, a clear majority, and it sits well clear of the roughly 40% of the whole web that passes. The pages Google's answer engine reaches for are, more often than not, quick and stable for the people who land on them.

The metric split shows where that edge comes from. Cited pages post 74.3% good LCP, ahead of the web's roughly 66%, so they load fast. They match the web on responsiveness at 83.6% good INP. The one place they give ground is visual stability, at 82.4% good CLS against the web's roughly 87%. That gap traces to the ad slots and third-party embeds some of these pages carry, which nudge the layout around as it loads.

The cited sources share a trait that shows up directly in the speed numbers: they tend to be lean, content-first pages. The list runs across the CDC, the FTC, and MedlinePlus and other government sites; Berkeley, Cornell, and Harvard; Mayo Clinic and Cleveland Clinic and other hospital systems; Investopedia and Bankrate in finance media; and Wikipedia and the dictionaries in reference. Different topics, one shared shape: pages built to answer a question rather than to sell, carrying little of the commercial weight that slows a typical site down.

The pattern in the data is a correlation, and it runs in speed's favor. The pages an answer engine cites pass Core Web Vitals more often than the web at large and load faster on real devices. We are not claiming a fast page earns the citation on its own, the substance of the answer clearly matters, but citation and speed travel together: the sources these engines reach for are, more often than not, quick and stable for the people who land on them.

The fastest page in the set is Wikipedia at 0.8 seconds, lean and famously cached. A state government page at 0.9 seconds and university law and health pages at 1.1 to 1.2 seconds round out the quick end. These are content-first pages with little of the ad and tracking weight that slows commercial sites, and they show how a lean page and a fast one are usually the same page.

Cited pages pass overall far more often than the web, driven by strong loading, with visual stability their one soft spot. Fastest cited: Wikipedia (0.8s), a state .gov (0.9s), university pages (1.1 to 1.2s).

Getting cited still takes real substance: thorough, well-structured answers an engine can trust. But here is the part a speed team should hold onto: the pages that win these citations are usually fast, and speed decides what happens after the click. An AI Overview citation is a link, and when a real person taps it, they land on your page. If it appears in one second and holds still, the visibility converts. If it takes three seconds and shifts under their thumb, it leaks.

That is where the soft spot in the data becomes a to-do list. The verticals that dominate this citation list, healthcare and finance among them, are middling on Core Web Vitals, and stability is the common weak point across cited pages. Reserving space for images, ads, and embeds so the layout does not jump is the single fix that would lift the most cited pages from soft CLS to a clean pass. Earn the citation with authority, then make sure the page behind it is fast enough to keep the visitor it sends. On this evidence, the majority of cited pages already are.

Embed this research

Cite this study