Writing for Search When Two-Thirds of Searches Never Click

Every few years SparkToro runs the same analysis: take a large panel of US browsing data, count Google searches, and count how many of them ended in a click to somewhere that is not Google. In 2024 the answer was that 60% ended with no click at all. The 2026 run, covering January to April, puts it at 68%. Only about 276 of every thousand searches reached the open web. The rest were answered on the results page, sent to another Google property, or abandoned.

That is the steepest two-year move since the series began, and the study attributes most of it to AI Overviews, which now appear on a substantial share of searches and cut the clicks below them by well over half. If you publish on the web and Google sends you readers, this is the most important number of the year, and it deserves a more careful reading than the headline gets.

What the number measures, and what it does not

Three things worth being precise about before drawing conclusions.

It is all searches, not your searches. The 68% includes people checking the weather, converting currencies, looking up a celebrity's age and spelling "necessary". Those were never going to click and now they do not need to. The searches a small publication lives on — comparisons, decisions, "which one should I pick" — are a different population, and the study does not break them out.

"Zero-click" is two things. Some of those searches were answered on the page. Some were simply abandoned, refined, or typed by someone who changed their mind. The study is honest about this; the headlines are not. A search that was abandoned in 2024 and abandoned in 2026 has not been "taken" by an Overview.

It is the US, on the web. Similarweb's panel covers US desktop and mobile web browsing. Searches inside apps, and every other country, are outside the data. The UK trend is almost certainly similar and almost certainly not identical, and nobody has measured it with the same rigour.

None of that makes the finding wrong. It makes it a trend rather than a verdict, and the correct response to a trend is to look at your own data.

What your own numbers will show

Search Console is the tool here, and it is free. Open the Performance report, compare the last three months to the same three months a year ago, and sort queries by the change in click-through rate rather than by clicks. What you are looking for is queries where impressions held up or grew and clicks fell. That is the signature of an Overview arriving above you: Google still shows your page, and fewer people need it.

When we ran this on ToolNest, the pattern was clean enough to be uncomfortable. Queries of the form "what is X" and "does X do Y" had lost a third to a half of their click-through rate. Queries of the form "X vs Y", "best X for Z" and "should I switch from X" had barely moved. The definitional pages were being answered above us; the decision pages were not, because a decision is not something a paragraph at the top of a results page can make for you.

That split is the whole strategy, and it cost nothing to find. We wrote up the free research stack, including the Search Console part, in keyword research without a subscription.

What we changed

Not the volume. Publishing more pages into a shrinking click pool is the response everyone reaches for, and it produces more pages that get read by an Overview and nobody else. What changed was what a page is for.

1. We stopped writing the fact page. "What is a password manager" is a query Google now answers itself, well enough. A page targeting it is a donation. The pages we write answer a question with a "because" in it — which one, for whom, after what changed.
2. The answer moved to the top. Every article on the site now opens with a quick answer of a paragraph. If that is all the reader needed, they have it, and they had to click to get it. If it is not, the rest of the page earns the time. This is also the section an Overview is most likely to cite, which brings a different kind of reader.
3. Every claim got a source with a date. Overviews and answer engines cite pages they can verify against. A pricing table that says when it was checked is more citable than one that does not, and it is more useful to a human for the same reason. The sourcing rules we follow are in checking a claim before you publish.
4. Titles say what the page decides, not what it is about. "Password managers after the price rises" tells a reader there is a judgement inside. "Password managers compared" tells them there is a table, and Google has one of those now.
5. We measured citations, not just clicks. Being named in an Overview shows up as impressions with a low click-through rate. That used to look like failure. It is now a separate line, because a reader who sees the name three times before clicking is a reader who arrives already trusting it.

What did not work

Two things we tried and stopped.

Writing "for AI". There was a period of advice about structuring pages so that language models would prefer them — particular heading patterns, question-and-answer blocks in a particular order, schema for everything. We tried it on a handful of pages. It made them worse to read and changed nothing measurable. The pages that get cited are the ones that were already clear and sourced. There is no trick; there is just the work.

Chasing volume with thinner pieces. For a month we published shorter, faster articles on more queries to see if breadth compensated for the falling rate. Impressions went up. Clicks did not, because the short pieces were exactly the kind an Overview replaces. We took most of them down.

The honest outlook

The trend is not going to reverse. Google has said clearly that it intends search to answer more questions in place, and a study showing that this is working will not change its mind. Clicks from search to the open web will keep falling as a share, and a small publication that treats search traffic as a stable input is planning on a number that will be smaller next year.

What is left is the part that was always the point: pages that help someone decide, that show their sources, and that are worth the click they still get. That is a narrower target than "rank for everything" and a better one. It is also, not coincidentally, the only kind of page that AdSense's own quality guidance ever wanted from us in the first place. Our publishing rhythm changed to fit it, and the weekly workflow now has a research day that used to be a writing day.