Small sites lost 60% of search traffic — we chose citation over clicks

Reporting on 1 October 2026 put the decline in Google search referrals to news sites at 40.2% year over year as of mid-2026. The analytics platform Chartbeat, tracking more than 2,500 news sites, found organic Google traffic down 33% worldwide and 38% in the US over the year from November 2024, with small outlets in the 1,000 to 10,000 daily pageview range losing 60%. The same report cited randomised field experiments finding clicks on external links down 39.8% when an AI Overview was present, and searches ending without a click up 34.5%. The share of US Google searches that led to no external click rose from 60.45% in 2024 to 68% in early 2026. A separate report the same day said Google is running a pilot that pays roughly 100 publishers for content used in AI answers, with amounts ranging from under 1,000 dollars a year to more than a million.

A zero-click search is a search that ends without a visit to any external site, and in early 2026 it accounted for 68% of Google searches in the US.

“Neither the 40% nor the 60% means anything without a baseline. We started on the side that has none.”

Put side by side, the two reports settle into one frame: AI took the clicks, and payments cover part of the loss. That is the story of the side with something to lose. Neither the 40% nor the 60% means anything without a baseline, and we started on the side that has none. A site that opened in mid-August 2026 has no traffic to win back and no leverage to get onto a list of 100. So the question changes. Not how to get the clicks back, but what can still be had where there is no click. The answer we settled on is citation. Someone who reads the result and never arrives can still leave with a name and one sentence of ours inside the answer.

First limit: we have no way to confirm a citation. Search Console reports impressions and clicks; it does not report whether one of our sentences went into an AI Overview. The only check available to us is a person searching and looking, and that comes out differently by query, by region and by hour. We picked a goal we cannot measure.

Second limit: a citation leads nowhere next. A click buys the chance to show a second screen; a citation ends where it starts. Between a name appearing once and someone opening the app there is no path, and we have not built one. We do not know that one can be built.

Third limit: being easy to quote does not make a sentence right. Write the definition as one sentence, put that sentence in the meta description and in the same place in the body, and a machine finds it easier to lift. Easy to lift and accurate are separate conditions. When the form is right, a wrong sentence is lifted just as readily. From the moment we adopted this structure, the cost of getting one sentence wrong went up.

So what we changed was not the length but where the sentence sits. Every page gets a one-sentence definition in the form "X is Y", and that sentence goes into the meta description and into the second line of the body, identical down to the character. With the same sentence in both places, whatever summarises the page has fewer things to choose between. Questions in the FAQ structured data come only from queries that actually arrived, never invented, spacing variants included exactly as they came. The same sentence goes into llms.txt. And when we write an article, we do not answer the same question twice. One piece takes one question and the next takes a different one. Two pieces on one question and both go down.

What remains: a sentence written on our own domain is still our own claim. Writing the definition cleanly does not produce third-party confirmation. The reason we lose on searches for the founder name is not the markup either; it is that little outside material states the same facts. Page structure does not fix that. And if the decline in clicks keeps going, the value of a citation can fall with it. We write where the room is left, and we do not yet have evidence that it is the right room.

WorldWideSomething studies the psychology of people through HCI and AI and designs apps out of it. Beff, the proactive AI friend app, is public; LiveLike, a job-experience chat, and AStream, an AI BJ live broadcast, are not public yet. We test this structure on the public side first and move only what works.