The clicks aren't dying. They're emigrating.

A 60% decline in clicks. The strangling of the open web. The end of search as we know it. The commentary around Google's AI Overviews has reached the pitch of a fire alarm, and like most fire alarms it is mostly telling you that somebody has burnt some toast.

The clicks are real, the decline is real, and I am going to show you the numbers. But the conclusion at the end of this piece is that search traffic is moving, not ending, and before you trust that conclusion you deserve to know who is selling it to you.

A declaration of interest

I sell generative engine optimisation. I have also spent 17 years doing SEO, which means I have both a new service and an old reputation riding on "search still matters." When a consultant who sells AI visibility concludes that AI visibility is the future, you should apply exactly the suspicion you would apply to an SEO platform announcing the apocalypse it happens to track. I wrote a while back that a source that always concludes the source is the best is not a source, it is an advert. The razor cuts both ways, and it cuts me too. All I can do is show the working and let you check it.

Grading the evidence

Almost every number in this debate comes from somebody with a dog in the fight, so let's sort them by how much you should trust them.

Top shelf: real people, observed behaviour. Pew Research watched 900 US adults make nearly 70,000 real searches. Where the results page had no AI summary, about 15% of searches produced a click on an organic result. Where a summary appeared, that fell to 8%. And links inside the AI Overview itself were clicked roughly 1% of the time. Pew sells nothing in this market. Sit with those three numbers before you read anyone else's.

Middle shelf: vendor studies, useful but motivated. SEO Stack analysed 100 million keywords of warehoused Search Console data and found clicks falling while impressions hold, which is genuinely valuable evidence, wrapped in language like "abomination" by a company that sells AI Overview tracking. Ahrefs put the click reduction on affected queries at 34.5% in April 2025 and 58% by February 2026. Cloudflare reports Google's crawl-to-referral ratio moving from roughly 2:1 a decade ago to 14:1 by mid-2025, while selling a pay-per-crawl product that monetises exactly that grievance. SparkToro says 68% of US searches now end without a click, built on clickstream panels whose representativeness Google disputes and whose data sources have changed between studies. None of this is fabricated. All of it is published by people whose revenue improves when you are frightened.

Bottom shelf: the defendant's testimony. Google's official position is that total organic click volume is "relatively stable" and that it now sends "slightly more quality clicks" than a year ago. No data published, no methodology, no definitions you can audit, from the one party with full visibility and the most to lose. Meanwhile Chartbeat data in the Reuters Institute's 2026 trends report has publisher traffic from Google down by roughly a third. A company marking its own homework and declining to show the paper is not lying, exactly. It is just not evidence.

Strip the incentives out and a stable core remains. Where an AI Overview appears, clicks to the web roughly halve. How much of the whole that represents is contested. The direction is not.

Where the traffic is actually going

Now the part the doom pieces leave out. Similarweb counted 1.13 billion AI referrals to the world's top 1,000 sites in June 2025, up 357% year on year. Adobe measured AI-driven traffic to US retail up nearly 700% over the busy Christmas period. The assistants are no longer a curiosity at the end of the funnel. They are a referral channel with a growth curve you could ski down.

Honesty requires the next sentence, so: that volume does not replace what Google is removing, and it will not for years. A billion monthly referrals across the entire top 1,000 sites is a rounding error against Google's scale, and anyone implying the lost traffic comes back, click for click, is selling something. Possibly me. Check above. I am not telling you that.

What the assistants replace is value, and the data on this is consistent and striking. Similarweb puts the conversion rate of ChatGPT-referred visits at 11.4%, against 5.3% for organic search. The visitor who arrives from an AI recommendation has already had the comparison done for them, the shortlist drawn, the question answered. They turn up at your door pre-sold. That trade, fewer visitors of higher intent, is the actual shape of what is happening, and it is a trade most businesses should take, provided they can see it happening. Which, as we will get to, most currently cannot.

What DuckDuckGo's record month actually measures

While the open web argues about Google, DuckDuckGo quietly posted record query volume in May 2026, with its sharpest growth landing right after AI Overviews became the default experience. The tempting read, and plenty have reached for it, is that users are fleeing AI search for the old ten blue links.

That read does not survive contact with the rest of the data. Gemini's referral traffic grew 388% year on year in late 2025, far faster than DuckDuckGo's queries, and ChatGPT serves billions of prompts a day. People plainly do not hate AI answers. And DuckDuckGo itself ships AI answers and an AI chat product. The difference is a single setting: you can turn them off.

So the record month is not a vote for traditional search. It is a vote for being asked. Google made AI summaries the default for everyone, everywhere, without a dial. DuckDuckGo made them optional. The backlash is not against AI in search. It is against not being consulted, and the lesson for anyone building products is older than search itself: people will accept remarkable changes if you give them the courtesy of a choice.

The clicks leaving first

"Moving, not ending" is true in aggregate and cold comfort for specific business models, so let's be honest about who is actually emigrating feet first: every business whose product was the pageview.

Early in my SEO days, a colleague noticed that local builders had essentially no presence online. So he threw together a website about swimming pool installation. Glossy photos, confident copy, your area covered. He had never installed so much as a paddling pool. He simply knew how to rank, and every enquiry that came in was sold on to someone who could actually dig the hole. In 2009 that was a smart trade, and I will admit a sliver of professional admiration for it even now.

But notice what that business actually was: an arbitrage on the real providers' web illiteracy. The affiliates, the aggregators, the lead-gen sites and the comparison listicles all run versions of the same trade, and in fairness they did a real service for years. When the companies that dig the holes were invisible, somebody findable had to stand in the doorway. The visitor still got their pool.

That arbitrage is now closing, because the machines doing the recommending are weighting exactly the things a middleman cannot fake: demonstrated expertise, a real entity, evidence you have actually done the work. Pew's cruellest number is the 1%. If your entire model needs the click, a citation is not compensation, and no repositioning deck changes that. For businesses built on intercepting traffic, this genuinely is an apocalypse, and I will not pretend otherwise.

For everyone else it is a disintermediation, and the world has rarely mourned a disintermediated middleman. We mostly resent the agents wedged between us and the thing we want, taking a cut and frustrating both sides (it's exactly why you'll never find a popular estate agent). If the actual provider of the service has a web presence good enough to be found, recommended and trusted, why shouldn't they collect their own customer? The honest answer is that for fifteen years many of them couldn't be bothered, and the middlemen ate well off the laziness. The grace period is over.

Not a conspiracy, just a squeeze

A lot of the commentary wants this to be deliberate: Google strangling organic traffic to force everyone onto ads. I don't buy it, and I say that as someone who will happily criticise Google all day. Featured snippets have been answering questions on the results page since 2014, and they were eating clicks long before anyone had heard of a transformer. AI Overviews are an escalation of a fifteen-year trend, not a heel turn.

The simpler explanation is the usual one. Google wants to remain everyone's first port of call, and for the first time in two decades it has a genuine reason to worry that it won't be. If people learn to start their questions in a chat box, Google loses the one habit its entire business sits on. So it answers in page, keeps the user, and the sites that used to receive that user are not the target of the strategy. They are the verge it is driving across.

A prediction you can hold me to

That same fear is why I think the open web gets thrown a rope, and Google has already blinked once. In May 2026 it shipped a set of AI Overview changes explicitly designed to send more traffic out: a "Further Exploration" section of curated source links, inline links with hover context so users can see where they are going, and labels highlighting publications you subscribe to. You do not ship traffic-outward features unless the inward-only version was becoming a problem.

So, on the record, June 2026: within 18 months, clicks from AI Overview citations become a channel you can meaningfully measure and optimise, with Google reporting it and a real click-through rate attached, because Google needs the open web fed. A summary engine that starves its sources is a very expensive way of becoming wrong, and Google knows it. If I am wrong about this, the piece stays up and you are welcome to wave it at me.

Measure what moved

Which leaves the practical question, and the answer is blunt: stop treating traffic as the score. Traffic was always a proxy for customers, and the proxy is breaking. The things worth measuring now are presence, sentiment and conversion. Which AI answers mention you, and which casually recommend your competitor. What the models actually say about you when nobody from your company is watching. What an AI-referred visitor does once they arrive, because at double the conversion rate, each one is worth counting properly.

Five years ago this job was guessing which keywords had volume and wedging them into paragraphs until the page read like a ransom note. The job now is knowing which questions your market asks an assistant, and whether the answer mentions you, which is a thing you can find out in about thirty seconds. Fewer visitors are coming. Better visitors are coming. Make peace with the first, get measurably good at the second.

The clicks are not dying. They are emigrating: to assistants, to consented search, to fewer and better visits that arrive already convinced. Traffic was never the point. Customers were. The shop window has moved again, and the only genuinely fatal strategy is to keep staring at the old one.