In November 2024, an entertainment site called Giant Freakin Robot shut down. Its founder said monthly traffic had gone from 20 million views to a few thousand. The travel blog The Planet D closed after losing roughly 90% of its traffic. Stereogum, publishing music writing since 2002, lost 70% of its ad revenue and came back as a paid membership site.

None of them got penalized by Google. None of them lost their rankings. Their readers just stopped needing to click.

That is zero-click search, and in the United States it is now the normal ending of a Google search rather than the exception. This article covers four things: what the term means, what the measurements actually say, who is fighting about it in court, and what still earns a click in 2026.

One quick clarification

Security people use "zero-click" for something else: a zero-click attack is spyware that reaches a phone without the owner tapping anything. Same words, different subject. This article is about search.

What Zero-Click Search Means

A zero-click search is a search that ends without a click on any website. The user types a question, reads the answer on Google, and leaves. Your page can rank first and still get nothing, because the visit became unnecessary.

The result page stopped being ten blue links a long time ago. Today the same page carries a knowledge panel, a featured snippet, a "People also ask" box, a calculator, a currency rate, the weather, a map, a sports score, and above all of it an AI Overview. Every one of those answers the question in place.

There is a second kind of zero-click that people forget. The user does click, but clicks something Google owns: a YouTube video, a Maps listing, the Shopping tab. That click counts in Google's totals and never reaches your site. In 2024, out of every 1,000 US searches, only 360 clicks went to the open web, according to SparkToro, the research company founded by Rand Fishkin. In the EU it was 374.

Which answers still need your page? Ends on the search page the summary can carry all of it a definition a date or a number a unit conversion a three step how-to a company's phone number Still needs your page the summary cannot carry it your price a tool that runs what you measured yourself what it was like to do it this week's local rule
Before you read the rest, sort your own pages into these two columns. The left column is where the losses are.

The Numbers Behind the Panic

The broadest measurement comes from SparkToro, using Similarweb's clickstream panel. Between January and April 2026, 68.01% of US Google searches ended without a single click to any website. In 2024 the figure was 60.45%. A decade ago it sat around 45%.

Three details in that study matter more than the headline number:

Three details

  • An AI Overview cuts click-through by roughly 60%. The summary does not reduce clicks a little. It takes most of them.
  • Repeat searches rose 7.2 points. Instead of sending you out, Google gets you to search again. The session stays inside Google.
  • AI Mode accounted for only 0.34% of searches. The damage is not coming from the separate conversational tab. It comes from the summary that appears whether you ask for it or not.

What the panel cannot see

This is panel data, not Google's own numbers, and Google's mobile app sits outside the panel, where zero-click behavior runs even higher. The panels also changed hands over the years, from Jumpshot to Datos to Similarweb. So treat the decimal point with suspicion and the direction as settled. Every independent measurement points the same way.

Pew Research Center went at it from the user side. In July 2025 it published an analysis of real browsing data from 900 US adults: 68,879 Google searches, of which 18% produced an AI summary. When a summary was present, users clicked a result on 8% of visits. Without one, 15%. Only 1% clicked a source inside the summary. And 26% ended their browsing session right there, against 16% on a normal result page.

One Pew finding deserves its own line, because it explains who gets hurt. 60% of searches that begin with a question word, the who, what, when and why queries, produced an AI summary. Informational publishing lives on exactly those queries.

Clicked a result 8% with a summary 15% without one Ended the session 26% with a summary 16% without one Pew Research Center, July 2025, 68,879 real searches
Pew Research Center, July 2025: with a summary on the page, clicks fall by nearly half and more people quit the session.

What the Universities Found

Vendor studies are useful, but they are not peer reviewed. Three academic measurements from 2026 back up the same picture with methods you can check.

Three academic measurements

  • Washington University in St. Louis (Xu, Iqbal and Montgomery, arXiv, May 2026) tracked 55,393 trending queries over 40 days. AI Overviews appeared on 13.7% of them, and on 64.7% of question-form queries. About 30% of the domains cited did not appear in the page's own first-page results. Of 98,020 individual claims in those summaries, 11% were not supported by the pages they cited. And 50.6% of the cited pages carry display ads, which is how the loss becomes a revenue loss.
  • University of Washington (Khosravi and Yoganarasimhan, arXiv, 2026) used the staggered country rollout as a natural experiment. Search referrals to English Wikipedia fell 5.45% against the German edition and 4.82% against the French one, which works out to roughly 100 million fewer visits a month.
  • Stanford (Liu, Zhang and Liang, arXiv, 2023) audited generative search engines and found only 51.5% of generated sentences fully supported by their citations. The summary is confident. It is not always right.

Wikipedia did not wait for outside researchers. In October 2025 the Wikimedia Foundation published its own count: human pageviews down about 8% year over year, after it re-ran its bot detection and reclassified the traffic. The foundation named the cause itself, which is search engines answering directly, using Wikipedia's own text.

Why It Hits American Publishers Hardest

The open web in the US runs on advertising. An ad only pays when a human loads the page. Take away the visit and the business model goes with it, no matter how good the writing was.

What US publishers lost Giant Freakin Robot 99% 20M views a month to a few thousand · closed The Planet D 90% traffic after AI Overviews · closed Stereogum 70% ad revenue · now membership Business Insider 55% search traffic, 2022 to 2025 · cut 21% of staff People Inc. 40% Google traffic, year over year · Google was 70% of traffic Chartbeat panel 38% US search referrals, 2,500+ sites · one year Each bar reports a different measure. Read the label under it.
Six businesses, six different measures, one direction. Two of them no longer exist.

The pattern repeats at every size. Business Insider lost 55% of its organic search traffic between April 2022 and April 2025 and cut 21% of its staff. People Inc., the publisher formerly known as Dotdash Meredith, told investors its Google traffic was down 40% year over year; five years ago Google sent roughly 70% of its traffic, and by 2026 that share was down to about a fifth. Its CEO, Neil Vogel, said turning Google off entirely was "100% on the table."

The aggregate numbers agree with the anecdotes. Chartbeat measured Google search referrals to more than 2,500 sites and found them down 33% worldwide and 38% in the US between November 2024 and November 2025. Digital Content Next, which represents large publishers, put the median decline among its members at 10% in May and June 2025. The Reuters Institute asked 280 media executives in 51 countries what they expect: a 43% drop in search referrals over three years, with one in five bracing for more than 75%.

Where “Google Zero” came from

The Verge's editor-in-chief, Nilay Patel, coined the phrase in 2024 as a thought experiment: what would you do the day Google stops sending you traffic at all? Two years later it reads less like a thought experiment and more like a planning assumption.

The Fight in Court

American publishers did what American businesses do. They sued.

The fight, in order Feb 2025 Chegg sues Google first suit over AI Overviews Jul 2025 Pay Per Crawl Cloudflare starts charging crawlers Sep 2025 Mehta rules no opt-out for publishers Sep 2025 Penske sues and the RSL standard launches Aug 2026 “Seems really unfair” the judge pushes back
The legal and technical pushback arrived in the same eighteen months.

The docket

  • Chegg v. Google, February 2025. The education company filed the first major antitrust suit over AI Overviews, arguing that Google requires its content to appear in search at all, then uses that content to answer the question itself.
  • Penske Media v. Google, September 2025. The owner of Rolling Stone, Billboard, Variety and The Hollywood Reporter sued in federal court in Washington, D.C. The claim is antitrust, not copyright, and that choice is deliberate: the harm is the diverted traffic, not the copying.
  • United States v. Google, remedies, September 2, 2025. In the Justice Department's search monopoly case, Judge Amit Mehta ordered behavioral remedies and rejected a Chrome divestiture. On publishers he was blunt: they cannot opt out of having their content used in AI Overviews while staying in search, which leaves them, in his words, between a rock and a hard place. He declined to order a fix, saying the record did not support one.
  • August 2026. Hearing the motions to dismiss in the Penske and Chegg cases, the same judge said out loud that the situation "seems really unfair."

Academics have looked at the obvious remedy and found it wanting. In the Journal of European Competition Law & Practice, Spencer Cohen and Todd Davies (Oxford and UCL) argue in "Opt-out remedies will not fix AI overviews" that letting publishers opt out solves nothing, because a publisher that opts out loses the citation traffic too and still competes against the summary. Their proposed alternatives are an opt-in structure, an end to the feature, or licensing fees.

What does any of this mean for your business this quarter? Very little, and that is the point. Court cases move in years. Your traffic moved last month.

The Other Pushback: Charging for the Crawl

While the lawyers argued, the infrastructure companies moved. In July 2025 Cloudflare launched Pay Per Crawl, letting publishers charge AI crawlers for access, and in September it switched to blocking AI crawlers by default for new sites. That same month a group including Reddit, Yahoo, Medium, Quora, Ziff Davis and The MIT Press published Really Simple Licensing, a standard for stating a price for machine access to content.

Our view: this is a strategy for content licensors, not for most businesses. If you sell heat pumps or legal services, blocking the crawler does not get you paid. It gets you out of the answer, and your competitor stays in it. Cloudflare's own setting is worth checking for the opposite reason. If you run behind Cloudflare and the AI crawlers are blocked by default, you are invisible in the answers by accident. We wrote about that setting here.

What Google Says About All This

Google disputes the framing. On August 6, 2025, Head of Search Liz Reid published a post arguing that total organic click volume to websites has been "relatively stable" year over year and that click quality has gone up slightly, where quality means the user does not immediately bounce back.

The useful part of that post is not the defense. It is the list.

Google's own winners and losers

Gaining traffic: forums, videos, podcasts, posts with authentic voices, in-depth reviews, original reporting, first-person analysis.

Losing traffic: sites people visited for a quick answer.

Both sides can be right at once. The total can hold steady while the distribution changes underneath it, and Google's own sentence describes exactly that. A click is now a filter. What gets filtered out is not bad content, it is content of a certain type: the answer that fits in two sentences stays on the search page, and the answer that requires experience still costs a click.

Scale explains the urgency. At I/O in May 2026, Sundar Pichai put AI Overviews at 2.5 billion monthly users and AI Mode at 1 billion. OpenAI said in February 2026 that ChatGPT had reached 900 million weekly users. Whatever the lawsuits decide, that is where the questions are being asked now.

Is Blogging Over?

Honest answer: it is worth less than it was, and pretending otherwise helps nobody. People can now get an answer to almost any question without leaving the chat window, and every one of these systems keeps learning from its own conversation history, which is an archive nobody else has. If a site is built purely on information that anyone can restate, it is in trouble right now.

But look again at Google's winners list, then at what the Reuters Institute recommends to publishers, and then at the last three pages you actually clicked on yourself. They point at one thing: give people what a summary cannot produce.

Six things a summary cannot give

  • Firsthand experience. You ran the test, you made the mistake, you were in the room. A model can quote you. It cannot have been there. We wrote about experience as the core of content separately.
  • Original data. Your survey, your measurements, your Search Console numbers. Every figure in this article came from somebody who counted something.
  • A tool that runs. A calculator, a checker, a lookup. A summary cannot execute anything, so the click goes to whoever built the tool.
  • Current and local facts. Today's price, this state's rule, this week's change.
  • Community. The reason forums came back from the dead.
  • A stated opinion with its reasoning. Not a hedge. A position you are willing to defend.

Notice what is missing from that list. Nobody is telling you to write longer or to add keywords.

What Actually Changed in Commerce

Here is the part that gets lost in the panic. The market did not shrink. There are X consumers and Y suppliers, and both counts are unchanged. Nobody stopped needing a plumber, a CRM, or a pair of boots.

What changed is the distribution. It now splits between suppliers who invested in SEO and GEO and those who did not. Visibility used to be measured by rank. Now the measure is blunter: when your customer asks the question, does your name appear inside the answer? If it does not, your competitor's does, and the buyer continues with that name.

Buyers (X) same count as before Suppliers (Y) invested in SEO and GEO named inside the answer no investment absent from the answer same count as before most of the demand what is left
Same number of buyers, same number of sellers, different route between them.

For an agency this made the work larger, not smaller. SEO is still the base, and Google's own guidance says optimizing for generative AI features is still SEO. GEO is the work of being cited by generative engines; the term comes from a paper presented at KDD 2024 by researchers from Princeton and IIT Delhi. Above both sits AIO, our umbrella for strategy, measurement, test cycles and authority. We compared how often each term gets used in a separate piece.

We take careful steps here for a plain reason: new fields are easy to game. Paid "best of" lists, manufactured mentions, text rewritten "for the AI" all look clever this year and become liabilities the next. We would rather build the method on ground that holds.

Next Threshold: Agents That Buy

Information sites felt it first. Commerce is next, and the plumbing is already being laid. OpenAI and Stripe published the Agentic Commerce Protocol in September 2025. Google announced its Agent Payments Protocol the same month and a Universal Commerce Protocol in January 2026. Visa and Mastercard each shipped their own agent payment standards. That is five competing protocols and no settled standard.

For a store, this means integration work: product feed, stock, price, shipping, returns and payment all exposed in a form an agent can read and act on. That is the commerce version of what publishers are living through now. It is also next year's problem. This year's problem is whether your name shows up in the answer, which we covered in our guide to AI Overviews, and the ecommerce SEO guide still covers the ground underneath it.

What to Do About It

Four principles, in the order we go through them with a new client. The first one is not optional; the rest do not work without it.

1. Know the One Thing You Are Excellent At

First we need to know what a company's outstanding, genuinely excellent thing is. What is its character, and what separates it from its competitors? Saying "we are very good" does not make anyone good. There is a difference, and there is objective proof of being good: a measured time, a certification, a production capacity, a test result, a customer count. AI systems use that proof when they describe you. With no proof, the system writes a generic sentence, and in that sentence you and your competitor are the same company.

2. Publishing and Selling Became the Same Job

Companies are now expected to be far more explanatory about the work they do. A product page is no longer enough. How do I choose one, how do I maintain it, when do I replace it, when is it the wrong fit: those answers are expected from you as well. A company that teaches, guides and helps with its content is the one that ends up cited, both in generative engines and in AI Overviews. This is not sales copy. It is an honest answer to a customer's question.

3. Your Corporate Site Is the Heart of Your Business

Remember this: your corporate site is the heart of your trade. A person knows and defines themselves best, and mission and vision define a company philosophically. Your content has to be consistent, and it must not contradict what the academic world accepts as true. If it does contradict it, then you have to explain that clearly and well. Every theory you put forward and fail to explain becomes a weight you carry.

There is a machine reason for this too. The Stanford audit found that half of what generative engines assert is not fully supported by what they cite, and the Washington University study found 11% of claims in Google's summaries unsupported by their sources. In other words, the system will describe you whether or not you have described yourself clearly. Consistency across your site, your directory listings and the press is what keeps that description accurate.

4. Get Into the Answer, Then Measure It

The practical work is covered in our AI Overviews guide: how sources get picked, what query fan-out does to your page structure, and the six steps we run. Two documents belong on your desk, both from Google: AI features and your website and the Search Console generative AI report. The report gives impressions, pages, countries and devices. It does not give clicks, click-through rate or queries, so the rest comes from testing your own target questions on a schedule.

And what not to do

  • Do not block the snippet and expect traffic. The summary still runs; your competitor is simply in the spot you gave up. The tags are documented in Google's snippet guide, so use them on purpose, not in protest.
  • Do not buy a slot on a "best of" list. A ranking with no stated variable is cheap fuel today and a liability later.
  • Do not manufacture mentions. Google's guidance says plainly that it does not work.
  • Do not rewrite your pages "for the AI." Write them for the reader and make them answerable.

Official and Academic Sources

The figures above come from the documents below. Vendor measurements (SparkToro, Similarweb, Bain, Chartbeat, Digital Content Next) are named in the text; these are the primary and peer-reviewed sources.

Google and the US government

Academic and research institutions

In Short

What to take away

  • Zero-click search is a search that ends with no click on any site. In early 2026 that was 68% of US searches, up from 60% in 2024.
  • When an AI Overview appears, click-through drops by roughly 60%. Pew measured 8% versus 15% at the visit level.
  • Question-shaped queries are hit hardest: 60% of them produce a summary, and informational sites live there.
  • Real businesses closed. Giant Freakin Robot and The Planet D are gone; Business Insider, Stereogum and People Inc. rebuilt around other revenue.
  • Three universities and Wikipedia's own foundation measured the same effect, including 11% of summary claims unsupported by their sources.
  • The courts are engaged (Chegg, Penske, Judge Mehta) but nothing there will restore traffic on your timeline.
  • What still earns a click: experience, original data, working tools, current and local facts, community, and an opinion with reasons.
  • Commerce is next. Five agent payment protocols exist and no standard among them yet.

Frequently Asked Questions

It is a search that ends without a click to any website, because the answer was already on the results page. A knowledge panel, a featured snippet, a calculator or an AI Overview can each do it. Clicks that stay inside Google, such as a YouTube video or a Maps listing, are counted separately but reach your site just as little.

What Percentage of Google Searches End Without a Click?

68.01% in the US between January and April 2026, measured by SparkToro on Similarweb panel data. It was 60.45% in 2024 and roughly 45% a decade earlier. Bain's December 2024 survey put the share of searches ending without a visit at about 60%, from a different method.

Is Zero-Click Search the Same as "Google Zero"?

They are related but not identical. Zero-click search is the measured behavior of individual searches. "Google Zero", a phrase coined by The Verge's Nilay Patel in 2024, is the endpoint: the day Google sends a publisher no traffic at all. Zero-click is the trend; Google Zero is where the trend is pointed.

Did AI Overviews Cause This?

They accelerated it rather than started it. The line runs from the Knowledge Graph in 2012 through featured snippets and "People also ask" to today, and zero-click searches were already near half of the total in 2019. AI Overviews took it from a growing trend to the majority outcome, and SparkToro measures roughly a 60% click-through drop on pages where the summary appears.

Should I Block Google's AI Crawler?

For most businesses, no. Google's AI features and its search index use the same crawl, so blocking one takes you out of both. Judge Mehta described that bind from the bench in the remedies ruling. Unless you are a licensing business, the goal is to be in the answer with accurate information and a page worth clicking.

Is It Still Worth Writing a Blog?

Only if the post does something a summary cannot. Information that anyone can restate in two sentences now stays on the search page. Firsthand experience, original data, a working tool, current or local specifics, community and a reasoned opinion still earn the click, and Google's own list of gaining sites names exactly those formats.