Hype Index: 'AI Overviews Cut Publisher Clicks by 58%'

TL;DR: Ahrefs really did measure 58%, and it is not a measure of publisher clicks. The February 2026 study measured the desktop click-through rate of the position-one organic result on keywords that trigger an AI Overview, December 2023 against December 2025, across 300,000 keywords. That is one ranking slot on one keyword set, not a publisher's traffic. I fetched 30 articles citing the number and counted 14 distinct percentage values in circulation for "what AI Overviews do to clicks," measuring at least six different quantities. We score it 40% noise / 60% signal: the study is solid, the citation of it is a mess.

Fifty-eight percent is the number of 2026. It has a headline in The Next Web, a press release on the wire, and a permanent seat in every "state of search" deck. It also sits next to at least three other live figures that measure entirely different things, and they get swapped for each other freely.

What is the claim?

Stated the way it circulates, in the words of headlines from my own sample:

"Google's AI Overviews killed 58 per cent of publisher clicks." "AI Overviews Now Steal 58% of Your Website Clicks." "Google AI Overviews Reduce Publisher Traffic by 58%."

What did Ahrefs actually measure?

I read the primary before writing any of this. Ryan Law and Xibeijia Guan, published February 4, 2026. The method, in their words: 300,000 keywords from Ahrefs Keywords Explorer, 150,000 that show an AI Overview and 150,000 informational keywords that do not, with aggregated Google Search Console data giving each keyword's average desktop CTR per month.

The numbers behind the headline are specific. Position-one CTR on AI Overview keywords was 0.073 in December 2023 and 0.016 in December 2025. The control group of informational keywords fell from 0.076 to 0.039 over the same window, which lets them forecast where AI Overview keywords would have landed anyway: 0.037. The 58% is the gap between that forecast and the actual, (0.015677-0.03734)/0.03734. Ahrefs' own framing: "For every 100 clicks you could historically earn for a top-ranking page, Google now 'keeps' 58."

Note what that sentence contains and what it does not. It is a ratio for one position on one keyword sample. Ahrefs published the whole ladder: position one is down 58.0%, position two 50.8%, position five 32.6%, position ten 19.4%. Nobody quotes the ladder. Everybody quotes the top rung and calls it "publisher traffic," which is a different denominator with a different shape, because no publisher ranks first for everything and no publisher's traffic is one keyword set.

How many numbers are actually in circulation?

Fourteen, by my count, and they are not measuring the same thing. Four matter most, and every one has a different denominator:

• Ahrefs, 58% (February 2026): desktop CTR of the position-one result on AI Overview keywords, forecast-adjusted.
• Ahrefs, 34.5% (April 2025): the same design run a year earlier, March 2024 against March 2025, position-one CTR 0.073 to 0.026 against a 0.040 forecast.
• Pew, 8% versus 15% (July 2025): not a reduction at all. It is a click rate per visit, from a panel of 900 US adults and 68,879 searches in March 2025, of which 12,593 showed an AI summary. Users clicked a traditional result on 8% of visits with a summary and 15% without. Only 1% clicked a link inside the summary.
• Agarwal and Sen, 39.8% (SSRN, April 2026): a randomized field experiment, 1,065 US Chrome users, 68,089 searches, browser extension assigning users to see or hide the summary. Outbound organic clicks fell from 0.62 to 0.37 per search. This is the only causal estimate of the four. Flagging my own chain here: SSRN returned 403 to every fetch I tried, so unlike the other three I am citing this one from PPC Land's report of it rather than from the paper.

A CTR ratio for one ranking slot, a per-visit click rate across a panel, and a per-search click count from a randomized trial. These are not versions of one measurement. Ahrefs' own post adds four more values in a single sentence, citing Seer Interactive at 49.4% to 65.2%, Kevin Indig above 50%, Authoritas at 47.5%, and the Daily Mail at 80% to 90%. The rest of my sample supplied 15% (Amsive), 89% (DMG Media), 38% and 33% for publisher-level traffic, and 61% from Seer.

What did 30 articles do with it?

I collected 32 articles published between February 1 and August 15, 2026 that state a percentage for AI Overviews' impact on clicks, and fetched each one. Thirty resolved: one returned 404 and one returned 403 to every fetch I tried. Here is the tally, run August 15, 2026.

Twenty of the 30 contain "58%". Only 16 use it for the Ahrefs finding. The other four attach "58%" to something else entirely: the share of SERPs carrying a People Also Ask box, the year-over-year growth in AI Overview coverage, and a zero-click baseline. The number has started detaching from its own study and drifting onto whatever quantity is nearby.

Of the 16 that do cite the Ahrefs finding, 11 link to the study and five do not. Four of the 16 rename the denominator in their own headline or lead, calling it publisher traffic, publisher clicks, or "your website clicks." Twelve describe it correctly somewhere in the body as position-one or top-ranking-page CTR, which is better than I expected, though several of those still carry a headline that says something different from the paragraph underneath it.

One article headlines the figure as an antitrust revelation: "How Antitrust Filing Reveals 58% Click Decline." I could not find a filing that produces 58% independently, and every other article in my sample that sources the number at all sources it to Ahrefs. I am not going to assert that the filing's number and the Ahrefs number are the same finding, because I could not confirm it. What I can report is that the chain in that article did not resolve to a document I could read.

What is not true?

That publishers lost 58% of their clicks. Nobody measured that. The 58% is what happens to the top organic slot on keywords where Google shows a summary, against a forecast, on desktop. A site whose traffic is spread across positions three to ten on a mix of keywords, only some of which trigger an Overview, is described by a completely different arithmetic, and Ahrefs published that arithmetic in the same post.

The other thing that is not true is that the four numbers corroborate each other. They are frequently stacked in a single paragraph as if 58%, 39.8%, and Pew's 8% were three readings of one instrument. They are three instruments. If you want to know why your own numbers do not match any of them, that starts with measuring AI traffic honestly in your own analytics rather than importing someone else's denominator.

What would change my mind?

A publisher-level, position-weighted study with a control group. The Agarwal and Sen experiment is the closest thing that exists, and its 39.8% is a real causal estimate of outbound clicks per search. If a large panel of publishers released before-and-after traffic with AI Overview exposure controlled, and it landed near 58%, I would happily retire this post. The direction is not in dispute in any of the 30 articles I read, or in any of the 14 values. Clicks are down. The argument is only ever about what the denominator was.

Verdict

The study is careful, the method is published, the arithmetic is in the post, and Ahrefs even printed the ladder that undercuts the headline version of its own finding. Then 30 downstream articles turned a position-one CTR ratio into "publisher clicks," and four of them stopped attaching the number to the study at all. Like the 95% pilot-failure stat, the failure is not in the research. It is in the sentence that carries it. The same pattern is running right now on Wikipedia's traffic decline, where the caveat is in the source and not in the citation.

Verdict: 40% noise / 60% signal. Quote the number with its denominator or do not quote it.