68% of Google searches now end without a click

Between January and April 2026, 68.01% of Google searches in the United States ended without a click on anything: no organic result, no ad, nothing. Two years earlier that figure was 60.45%. In 2019 it was 49%.

The mechanism is not what most people think. It is not that everyone has switched to chatting with a bot. It is that Google now answers the question at the top of the results page, and for most searches that is enough. The blue links are still there. They are just increasingly decorative.

Here is what the evidence actually shows, what it means if you run a business rather than a publication, and, most importantly, which channels still put customers in front of you.

What 68% actually looks like

Per thousand US searches, roughly 276 clicks now reach the open web. Google’s own properties absorb about 27% of all clicks. Paid results take another slice. What is left over for everybody else’s websites is under a third of what people are searching for.

The two-year change is the part worth sitting with. The share of searches producing at least one click fell 9.51 percentage points between 2024 and 2026, a 22.9% relative decline. Roughly one click in five, gone, in twenty-four months.

First, the number doing the rounds is wrong

There is a claim circulating on YouTube, in newsletters and in more than one conference talk, that Google announced conversational AI at google.com has reached 58% of searches ending with no click.

No such announcement exists. Google has never published a zero-click rate, and its public position is the exact opposite of that claim.

The 58% is Ahrefs’, from February 2026, and it means the clickthrough rate of the number-one organic result falls 58% when an AI Overview is present. That is a very different claim from “58% of searches get no click”, and, awkwardly for the doom narrative, the real zero-click figure is higher than the myth.

The distinction matters commercially. Google’s chat-style AI Mode passed a billion monthly users and still accounted for just 0.34% of US Google searches in the same period. Both facts are true. If you believe everyone has moved to a chatbot, you will spend money in the wrong place. The damage is being done by AI Overviews sitting above ordinary search results, on the ordinary results page, which most people are still using.

The evidence moved from correlation to proof

Until this year, every zero-click study was observational: measuring what happened, unable to prove why. Google leaned on that hard, describing third-party research as resting on “flawed methodologies.”

Then somebody ran the experiment.

Researchers at the Indian School of Business and Carnegie Mellon gave 1,065 US Chrome users a browser extension and randomly assigned them to see AI Overviews or have them hidden. Everything else was held constant. Hiding the AI Overview raised outbound clicks from 0.38 to 0.61 per search, meaning AI Overviews cut organic clicks by 38%, and the zero-click rate fell from 72% to 54%.

The finding that should worry publishers most is the one about people. Satisfaction, information quality and ease of finding information were near-identical in both groups. Users are not being short-changed. They are being served. Which is precisely why none of this is going to reverse.

What it looks like from the searcher’s side

The cleanest behavioural data belongs to Pew Research Center: independent, non-commercial, and based on real browsing rather than modelled estimates.

Two details deserve attention. People who saw an AI summary were more likely to stop browsing entirely: 26% ended their session, against 16% without. And of all visits to a page carrying an AI summary, 1% included a click on a citation link inside it. The citations are visible. They are almost never used as a door.

Ranking first is now the worst place to be standing

Ahrefs compared 150,000 keywords that trigger AI Overviews against 150,000 that do not, using aggregated Search Console data from December 2023 and December 2025.

Ahrefs are careful with their verb and so should everyone quoting them: the AI Overview correlates with the lower clickthrough rate. These are two different keyword sets, not the same queries before and after, and AI-Overview keywords skew informational. That is exactly why the randomised experiment above matters: it is the piece that establishes cause.

In their words: “for every 100 clicks you could historically earn for a top-ranking page, Google now keeps 58.” Their April 2025 study put position one at 34.5%. It nearly doubled in eight months.

This has been independently replicated in another market entirely. SISTRIX, analysing more than 100 million German keywords, found position-one clickthrough falling from 27% to 11%, around a 60% drop, and estimated 265 million organic clicks per month lost in Germany alone. Different vendor, different country, different dataset, near-identical result.

The smaller you are, the harder this landed

This is the finding that never makes the headlines, and it is the one that should decide your budget.

Chartbeat’s analysis of thousands of sites found Google search referrals down 60% for small sites, 47% for mid-sized, and 22% for large ones over twelve months. If you run a small business site and your Google traffic has roughly halved in a year, you are the median, not the exception.

Over the same period, AI chatbots sent under 1% of publisher page views. Nothing has arrived to replace what was lost.

Google’s side of it

On 6 August 2025, a year ago this month, Liz Reid, VP and Head of Google Search, published a post stating that “total organic click volume from Google Search to websites has been relatively stable year-over-year” and that Google is “actually sending slightly more quality clicks to websites than a year ago.”

The post contains no numbers. No time series, no definition of “slightly”, no breakdown by site size.

It should be said that “total organic click volume” being flat is entirely compatible with everything above: an aggregate can hold steady while redistributing violently away from small sites toward large ones, which is exactly what the Chartbeat data shows. Google also acknowledges this, noting that traffic is shifting between sites.

The “quality clicks” defence, that AI Overviews filter out visitors who would have bounced anyway, is the one claim that has now been tested experimentally, and the randomised trial found no significant difference in click quality between the two groups.

Meanwhile Google’s search advertising revenue reached $63.27bn in Q2 2026, up 17% year on year. Whatever is happening to your clicks, it is not happening to Google’s business.

Why local search is the exception

Now the useful part. The collapse is not evenly distributed, and where it is weakest is where money changes hands.

Whitespark ran 540 manual queries across three US cities and six service industries. For genuinely local-intent searches (“plumber Houston”, the kind a person types when they intend to hire someone), the local pack appeared 93% of the time and an AI Overview only 15%. For informational searches, those figures almost exactly invert.

This is why “SEO is dead” and “local SEO still works” can both be true. AI came for the question, not for the purchase.

You will see it reported that AI Overviews appear on 68% of local searches. That is the same study’s average across all three intent types, and it is the single most misquoted number in local search. The number that matters to a business taking bookings is 15%.

Your Google Business Profile is now an AI data source

Google has said this on the record. Announcing Maps grounding in the Gemini API in October 2025, covering more than 250 million places, Google wrote that “Google Maps uses data from places, and other related content like user reviews, as sources to help generate the response.”

At the same time Google has been steadily removing your ability to write the answer yourself. The Business Profile website builder was switched off in 2024. Chat and call history followed in July 2024. Q&A was deprecated across late 2025 and early 2026, replaced by an AI “Ask” button that generates its own answer from your profile, your reviews and your site.

Rio SEO’s data across more than 239,000 US business locations shows calls down 12.9% and website clicks down 7.7%, but direction requests up 6.4%. Physical-visit intent is intact. That is hard to reconcile with a simple “AI killed local” story.

Two mechanical reasons calls fell that have nothing to do with demand: Google’s AI local packs carry no call button, and since July 2025 Google’s AI will place the call itself on a searcher’s behalf, a call that never appears anywhere in your profile’s statistics.

But the local shelf is getting shorter, and more of it is for sale

Honesty cuts both ways.

Sterling Sky found that AI-powered local packs show one or two businesses instead of three, and across 322 US markets surfaced 5,943 unique businesses where traditional three-packs surfaced 18,330, about a third as many, with fewer businesses shown in 88% of markets.

Over the same year, local pack ads grew from 1% to 22% of tracked queries and Local Services Ads from 11% to 31%. A meaningful share of the organic local visibility that disappeared was not taken by AI. It was sold.

Nearly half of consumers now ask AI first

BrightLocal’s 2026 survey of 1,002 US consumers puts generative AI third on the list of places people look for a local business, named by 45%, behind Google at 71%, ahead of Apple Maps at 27%.

You will see that 45% reported as “up from 6% last year.” Resist it. The comparable 2025 question asked how many used AI to find reviews, which is a different behaviour, and that figure had itself gone down from 9% the year before. A seven-fold jump straight after a decline usually means the question changed, not the country. The 2026 level is worth knowing on its own; the growth rate is not evidence.

What the right-hand panel shows is more useful anyway: among people who do use AI, 82% read the AI’s summary of a business’s reviews and 23% will rely on that summary alone. For roughly a quarter of those prospects, the machine’s paraphrase of your reviews is your reputation.

Reviews stopped being social proof and became source material

This is the change most businesses have not registered. Your reviews are no longer just something a human reads before choosing you. They are training data for the sentence an AI says about you.

Researchers varied seven business attributes independently and measured how each changed the probability of being an AI’s recommendation. Three findings are worth acting on:

• Star rating dominates. Moving from 3.9 to 4.7 stars raised selection probability by 31.7 percentage points.
• Review volume matters far less than volume-chasing implies. A forty-seven-fold increase in review count, from 45 to 2,100, bought 8.3 points, roughly a quarter of the rating effect.
• Owner responses had no measurable effect at all (+0.11 points, confidence interval spanning zero).

That last one is independently corroborated: a panel of 47 local search experts ranked owner responses 122nd out of 187 local ranking factors. Two completely different methods, same answer.

But responses absolutely move people: 89% of US consumers expect them, 80% are more likely to use a business that replies, and 50% are put off by generic templated replies. Reply because a human is reading. Never buy it as an SEO product, and never automate it.

The figure that ties the whole article together is the last one: 97% of the US consumers who ask an AI for a recommendation then check it against real review platforms, and 42% always do. AI has become the discovery layer. Reviews became the verification layer. Being recommended by ChatGPT does nothing for you if the profile the customer checks next is three years stale and sitting on 3.8 stars.

You still need a website, but not for the reason you have been told

Here is the uncomfortable half of the evidence, stated before the comfortable half.

Your own site is a minority of what AI cites. Estimates vary with how you define “owned”: 13.7% in Muck Rack’s 25-million-link analysis, 14.3% in an independent academic study of 167,551 citations, but every serious dataset puts it in the minority. Note the bottom row: paid and advertorial content is 0.3% of citations. There is no door here to buy.

So why does the website still matter? Because it is the eligibility requirement, and Google says so in writing. From Google Search Central: “To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet.”

No indexed page, no citation. And being cited is worth a great deal:

Brands cited in an AI Overview earned roughly 120% more clicks per impression than brands that were not, while still landing well below a page with no AI Overview at all. Seer state explicitly that they cannot claim causation and that one account skews the “not cited” figure, so treat it as a strong signal rather than a promise.

And if you are wondering how much your ranking still matters, the honest answer is that nobody can tell you precisely:

Ahrefs themselves say their two datasets are not directly comparable, because their citation parser improved between studies. And beware the denominator: studies asking “did at least one citation on this page come from the top 10?” report over 90%, because that is a different question. Ranking still matters more than any other single input. It has simply stopped being a guarantee.

What actually feeds an AI answer about your business

Nothing on the left is new work invented for the AI era. It is a website, a profile, reviews and a reputation: the same four things that made a business findable in 2015. What changed is that a machine reads them now instead of a person, and it will paraphrase them to a customer who may never see a search result at all.

GEO: what is real, and what is being sold

“Generative engine optimisation”, also sold as AEO or AI SEO (the terms mean the same work), is the fastest-growing line item in the industry. Some of it is real. A surprising amount of it has been tested and failed.

llms.txt. Ahrefs checked 137,210 domains: 97% of published llms.txt files received zero requests in a month. SEO audit tools fetched them more often than AI systems did. The decisive control: no AI bot ever requested a missing one, proving they do not look. It costs ten minutes and does nothing. Nobody should be billing for it.

“AI schema markup.” Google’s own documentation says “there’s also no special schema.org structured data that you need to add” to appear in AI Overviews or AI Mode. Ahrefs then ran the causal test on 1,885 pages that added JSON-LD, matched against 4,000 controls, and found AI Overview citations fell 4.6%. Schema still earns its keep for rich results and for stating plainly who you are. It does not buy you into an AI answer, and it should not be sold as though it does.

“Your AI ranking position.” This one is worth understanding properly.

There is a worse than 1 in 100 chance that asking the same AI the same question twice returns the same list of brands, and worse than 1 in 1,000 for the same order. This is not a bug a vendor can patch: language models process your prompt in a batch alongside everyone else’s, and the arithmetic shifts slightly with batch size, so a measured “AI ranking” partly reflects how busy the servers were.

Share of voice, measured across hundreds of prompts run repeatedly and tracked as a trend, is a reasonable metric. A position number is not. If someone sells you rank 3 in ChatGPT, they are selling you server load.

And before anyone rebuilds their business around AI referral traffic:

You will also see AI traffic described as converting 54% better than everything else: Adobe’s finding across more than a trillion retail visits. It is published, it is large, and it disagrees with the peer-reviewed result. Adobe’s own figure had the opposite sign a year earlier. Growth here is real and fast. The volume is not there yet. Anyone selling you certainty on this point does not have it.

So what should you actually do?

Ranked by how well the evidence supports it, not by how much it costs.

1. Keep a real website, indexed and readable without JavaScript. It is the eligibility requirement for being cited at all, in Google’s own words, and it is where the sale actually happens. You can check what AI systems can see of yours with our free AI Visibility checker.
2. Do not block the assistants that send people to you. Training crawlers and answer crawlers are different things, and blocking the second means AI tools cannot recommend you. Our robots.txt tester will show you which you are currently blocking.
3. Treat your Google Business Profile as a data feed, not a listing. Categories, services, hours and attributes are now literally the source material for what Google’s AI says about you. Check how your local visibility looks.
4. Protect the star average above everything else in reviews. It is the one review signal with causal evidence behind it in both revenue and AI recommendation. Ask every customer, every time, never only the happy ones. A review QR code at the counter outperforms a follow-up email nobody opens.
5. Reply to reviews for the humans, not the algorithm. And write the replies yourself; half of consumers are actively repelled by templated ones.
6. Earn mentions on sites you do not own. Journalism, industry publications, “best of” lists, genuine directories. This is 84% of what AI cites, and it is the single strongest correlate of AI visibility measured so far.
7. Do the ordinary SEO properly. It remains the largest observable input to being cited, and it is the same work it always was.
8. Build channels that do not depend on a click. Email lists, repeat customers, referrals, direct demand. Publishers who did this in 2024 are the ones still standing.

Where this leaves us

The blue link is not dead, but it is no longer the destination: it is one of several places a machine looks before it decides what to tell somebody about you. Being findable now means being consistent everywhere that machine looks: your site says the same thing as your profile, which is corroborated by your reviews, which is echoed by what other people have published about you.

That is not a new discipline requiring a new budget line. It is the old work, done properly, for a new reader.

If you would like a straight answer about where your business currently stands in all of this, including which parts you can safely ignore, that is what our AI content and search structure work is for, and you are welcome to start with a conversation.