The Zero-Click Rate Is 46.96%. The Average Is the Least Useful Part.

Why three credible sources give three different answers

The two figures the industry already quotes are SparkToro at roughly 60% and Similarweb at roughly 37%. iPullRank now adds 46.96%, sitting between them.

The instinct is to decide which one is right. The study’s framing is better: all three are “right on average, but the average is the least useful part of the picture.” They measure different panels, different periods and different definitions of a session. iPullRank reports at the session level specifically because that is how SparkToro and Similarweb define it, which is what makes the comparison meaningful at all.

We have not independently audited SparkToro’s or Similarweb’s methodology, and we are citing those two numbers only as figures iPullRank quotes for contrast. Treat them accordingly.

The spread is the finding

Here is what one blended benchmark conceals, all from the same dataset:

Cut	Range
By country	about 41% in Japan and Vietnam to 54% in the Philippines, a 13-point spread. The US sits at 49%
By device, UK	65% on mobile against 46% on desktop, a 19-point gap
By device, Indonesia	55% on mobile against 49% on desktop, a 6-point gap
Widest single contrast reported	46% on desktop to 66% on US mobile

The mobile-kills-the-click effect that everyone repeats as a universal law turns out to be concentrated in English-speaking markets. The UK gap is 19 points, the US 17, Canada and Australia 15. Indonesia, the Philippines, Vietnam and Japan all sit near 6.

There is a stranger cut underneath that one. In Indonesia, users on google.com zero-click 11 points more often than users on google.co.id. In the UK, Australia and Canada it is a 6-point gap. Same country, same users, different Google property, materially different behaviour. The study’s suggested explanation is that Google ships its newest answer surfaces on google.com first.

If your reporting rolls all of that into one organic-CTR number, you are averaging across populations that do not behave alike.

The 14% that was never yours to lose

The finding we expect to be misread hardest: only about 14.0% of Google clicks go to a site the searcher named in the query. The other 86% are discovery clicks, landing on sites the user never asked for by name.

Read one way that sounds alarming. Read correctly it is the opposite. Brand-recall clicks were never winnable through search optimisation in the first place — those users were always going to arrive. The discovery clicks are the entire addressable game, and they are the large majority.

The study labels its own 14.0% as a conservative floor and says the true navigational share is “likely in the high teens, but almost certainly under 20%.” We are repeating that caveat because the source volunteered it.

Who Google can replace, and who it cannot

The study ranks major destinations by how often Google intercepts a search that was heading to them with its own on-page answer.

Hardest to intercept: OpenAI at 11.1%, GitHub at 12.7%, Reddit at 14.2%, LinkedIn at 16.5%, Amazon at 16.6%.

Easiest: Wikipedia at 30.5% and IMDB at 33.4%.

The pattern is not about size — Wikipedia and IMDB are among the most-visited destinations on the web. It is about whether the query has a factual core that a box can satisfy. “What year did this film come out” is answerable in a snippet. “Show me this specific repository” and “what do real people think about this” are not. Sites that hold conversation, transactions or proprietary artefacts are structurally more defensible than sites that hold facts.

Worth noting what is absent: traditional news publishers do not appear in the top 25 Google-click destinations at all.

ChatGPT is now a search destination, and it buys ads

ChatGPT ranks sixth among all sites for clicks arriving from Google search. About 4.75% of its Google-referred traffic comes from paid ads, against 0.2% for YouTube, effectively zero for Wikipedia and under 2% for Amazon. Together, AI tools pull roughly 720,000 navigational Google searches a week in this panel.

That is a destination category that did not meaningfully exist a year ago, now being defended with paid search.

The caveats the study states about itself

We are reproducing these because they materially bound what the headline number means, and because a study that publishes them earns more trust than one that does not.

The panel is 94.5% desktop, so the overall headline is effectively desktop-weighted. The demographic breakdown covers one week only and is 72% male, with the 25-34 and 35-44 bands making up 97.4% of that slice. And the study cannot separate AI Overview impact from Knowledge Panel impact from featured-snippet impact — it measures whether a click happened, not which surface prevented it.

That last limit matters if you were hoping this settles the AI Overviews argument. It does not.

What follows from this

Stop reporting a single zero-click benchmark. Segment by country, device and query type at minimum. A 13-point country spread and a 19-point device gap mean a blended figure describes none of your actual audiences.

Check which side of the interception line your content sits on. If your traffic depends on queries with a factual core, you are in the Wikipedia and IMDB position and the trend is against you. If it depends on discussion, transactions or things only you hold, you are closer to the Reddit and GitHub position.

Do not treat the 14% as a loss. It is the share you were never competing for. Your addressable surface is the 86%, which is a better starting position than the headline implies.

Wait before quoting the number at all. One panel, desktop-weighted, one 15-month window. It is the largest publicly available direct measurement of this behaviour, and it is still one measurement.

The honest version is less quotable than 46.96%: zero-click is real, it is rising, and it is not one number. Anyone selling you a single benchmark is selling you an average of populations that do not resemble each other — or your audience.