The zero-click number isn't the story. The composition is.

Sixty-eight percent. That's the SparkToro/Similarweb figure for the first four months of 2026 — the share of US Google searches that ended without a click anywhere. Not to your site. Not to anyone's site. Not even to a paid result.

Every SEO publication will run with that number this week. Most of them will frame it as the AI Overviews story. A few will frame it as the death-of-the-open-web story. Both framings miss the more interesting question, which is what the remaining 32% is actually made of, who's catching it, and why the businesses still growing revenue from search are the ones that stopped trying to win the average click a long time ago.

Because the headline number isn't new. Zero-click was 45% in 2016. 49% in 2019. 60% in 2024. The trajectory has been linear-to-accelerating for a decade. What's new in 2026 isn't that the number crossed two-thirds. What's new is the composition of the remaining third, the structural reasons it's collapsing toward branded and transactional intent, and what that means if you're a business currently planning your 2027 marketing spend around traffic forecasts that haven't been redrawn since 2023.

Most strategy decks being shown to UK businesses right now are operating on assumptions the data has just contradicted. That's the piece I want to write.

The number behind the number

Let's start with what 68% actually represents, because the framing matters.

The 68% number isn't an AI story. It's a Google product strategy story that AI accelerated.

It is not the share of searches where Google shows an AI Overview. AI Overviews appear on roughly 20% of searches and, when present, reduce click-through rate by about 60% — significant, but not the dominant driver of the broader number. Rand Fishkin is explicit about this: AI Mode itself was a third of one percent of searches in his data. A rounding error.

What 68% actually represents is a decade of Google steadily building features that keep users on Google. Featured snippets. Knowledge panels. People Also Ask. Local packs. Shopping carousels. Travel modules. Sports scores. Weather. Currency conversion. Definitions. Calculators. Then, layered on top, AI Overviews. Each of these features individually was defensible — Google was answering a question users were asking. Collectively, they shifted the equilibrium of the page from "ten blue links plus some chrome" to "an answer page that occasionally surfaces links if you keep scrolling."

That shift was already mostly complete before the AI conversation got loud. The 2024 number was 60%. AI Overviews accelerated it by 7.5 percentage points in two years. That's meaningful. It's not the whole story.

The reason this matters is that everyone framing 2026 as the year AI killed search traffic is going to draw the wrong conclusions and propose the wrong fixes. The fix is not "optimise harder for AI Overview citations." The fix is reckoning with the fact that Google has been a walled garden in slow motion for a decade and the walls just got higher.

The 68% number isn't an AI story. It's a Google product strategy story that AI accelerated.

What's still in the 32%

This is where it gets interesting. If two-thirds of searches end on Google, somebody is still earning the third that escapes. Who?

Cyrus Shepard at Zyppy did the homework on this — 400-plus winning and losing sites, profiled for shared characteristics. The pattern is consistent and worth sitting with, because it tells you exactly what kind of search query still produces a click in 2026 and what kind doesn't.

Winners overwhelmingly offer a product or service. They own proprietary assets — original data, original research, original tools, original photography. They stay tightly focused on a niche rather than sprawling across topics for traffic. They carry a recognisable brand. And, crucially, they let the user actually complete a task on the page they land on.

That last point is the one I'd underline. Nearly 84% of the winning sites in Shepard's analysis allowed the visitor to do something — buy, book, calculate, configure, sign up, request, compare. The losers were predominantly informational sites: blogs, content hubs, "ultimate guides," editorial pieces written to rank.

The 32% that still produces clicks is, with very few exceptions, transactional and branded. Someone searches "book hotel rome november," they click through. Someone searches "[your brand] login," they click through. Someone searches "best running shoes for plantar fasciitis," they read the AI Overview and stay on Google.

This has been true for longer than people want to admit. The category of search that produced informational traffic — the long tail of "how do I do X" and "what is Y" queries that built the entire content marketing industry between 2010 and 2022 — is the category Google has been quietly absorbing. AI Overviews are the final stage of that absorption, not the beginning of it.

The implication is direct. If your search strategy was built around informational queries — the model where you publish helpful content, rank for related questions, and convert a percentage of the visitors who arrive — that strategy is now operating against a 68% headwind and rising. It can still work. It works for a narrower band of queries than it did three years ago, and the band is shrinking.

The "traffic down, revenue up" pattern

Here's the part most of the industry isn't talking about properly.

Search is no longer how you earn traffic. It's how you capture demand you've earned elsewhere.

There's a quiet pattern emerging in the businesses that are navigating this well: traffic is down, revenue is flat or up. HubSpot is the named example Rand uses. I've seen it now with several clients across very different verticals — a UK B2B SaaS, an e-commerce brand in the home category, a professional services firm. The shape is consistent. Sessions from Google down 20-40% year on year. Conversion rate up. Revenue from organic flat or modestly up.

What's happening?

The traffic that's disappearing is the traffic that wasn't converting anyway. Informational visits that landed on a blog post, read the answer, and left. Top-of-funnel sessions that bounced. The "research mode" visitors who needed five more touchpoints before doing anything. Google has eaten that traffic, and for most businesses, that traffic was never the revenue.

What's remaining is the high-intent visitor. The person who searched for the product, the brand, the specific service in the specific location. The person who arrived already qualified.

This isn't a happy ending. Two things to flag before anyone takes that as good news.

First, the businesses showing this pattern are almost universally the ones that built brand and direct demand alongside their SEO. If your only acquisition channel was Google organic — and a meaningful percentage of UK SMEs are still in this position — losing the informational top of funnel is not survivable, because you don't have a brand layer above it catching the demand Google is now intercepting.

Second, the pattern doesn't extrapolate forever. The same logic that ate informational queries will, eventually, eat some transactional ones. AI Mode is at 0.3% of searches now. When it's at 5% — and it will be — agentic flows will start completing tasks without a click too. The 32% isn't a stable equilibrium. It's a smaller, more defensible territory than the 68%, but it's still being contested.

The lesson isn't "informational SEO is dead, focus on transactional." The lesson is that the share of total demand that flows through clicks of any kind is shrinking, and the businesses that survive are the ones building demand they can capture through multiple channels, including channels that don't depend on a click happening at all.

Search is no longer how you earn traffic. It's how you capture demand you've earned elsewhere.

The infrastructure category nobody priced in

While everyone was watching the click number, two adjacent shifts happened that change what "showing up in search" even means.

The first is Stripe Projects, launched April 30. The protocol lets AI agents create accounts, buy domains, upgrade plans, and deploy infrastructure on a user's behalf. Cloudflare, Vercel and Netlify shipped as launch partners. The category Stripe is building toward is agentic infrastructure purchasing — not retail commerce, but capability commerce.

The second is the Apple-Gemini deal. Siri AI, announced at WWDC, runs on a custom 1.2 trillion-parameter Gemini model Apple is reportedly paying around $1 billion a year to license. It's built into Spotlight on iPad and Mac, where people already type questions. Apple's own support documentation says web answers "may include links to sources" without committing to when, how often, or how anyone would measure them.

Read those two together. One says: agents are now buyers of capability on behalf of users, and the merchants are platforms. The other says: a billion-device install base just got an answer layer that may or may not surface the sources it cites, with no measurement framework.

Both are evidence of the same structural shift — discovery is moving away from the open web in two directions at once. Upward into AI conversational surfaces that synthesise rather than link. And sideways into protocol-level transactions where the user never sees a search result at all because the agent has already chosen the vendor.

This is the part I'd want every business owner reading this to internalise. The visibility problem isn't just "will my page rank in an AI Overview." It's "will an agent choose my service when a user asks Siri to handle the task." The shape of that visibility problem is not the same as SEO, and pretending it is — by relabeling SEO as GEO and selling the same playbook with markup on top — is going to leave a lot of businesses unprepared for the actual question.

The actual question is whether your brand exists in the training data, the citation graph, and the structured commerce protocols at a frequency and clarity that makes you the obvious choice when a model is asked to choose. That's a different problem from ranking. It overlaps with SEO. It is not SEO.

What the measurement gap actually looks like

The other thing nobody is saying clearly enough: we cannot measure most of this yet.

Apple's documentation explicitly admits it. Web answers from Siri "may include links to sources." There is no published rate, no benchmark, no panel data. A site could be cited in Siri's answers every single day across a hundred thousand users and have no way of knowing. The same is broadly true of ChatGPT, Claude, Perplexity, Gemini's conversational surfaces, and the long tail of vertical agents being built on top of these models.

The closest we have is log-file analysis — measuring bot crawl frequency from named user agents — and the emerging category of citation monitors that scrape AI tools at scale and report back on share of voice. The first is honest but incomplete. The second is mostly rebadged scraping at enterprise prices, and the metric it measures (frequency of citation in synthetic prompts) is correlated with but not equivalent to the metric that matters (real users seeing your brand in their actual answers).

This is the second-order problem nobody wants to confront. We're being asked to optimise for surfaces we cannot measure, using tools that are early, expensive, and methodologically shaky, while the surfaces we can measure (Google organic) are paying out less revenue per ranking position than they used to.

The honest answer is that the next eighteen months are going to be partially blind. The vendors selling certainty are selling fiction. The publications projecting precise GEO ROI numbers are extrapolating from data thin enough to read a newspaper through.

What you can do is build for legibility — make your business as easy to discover, cite, and transact with as possible across every surface that matters — and accept that you're going to be running the strategy without full instrumentation for a while. That's uncomfortable. It's also true.

What this means if you're allocating budget for 2027

Let me be concrete, because the analytical version of this argument can stay abstract forever.

If you're a UK business currently planning marketing spend for the next twelve to eighteen months, the 68% number should change three things about how you allocate.

The first is the mix between traffic acquisition and demand creation. If you've been spending 70% of your budget on activities designed to earn clicks (content, SEO, paid search), that mix is wrong now. Some portion of that — call it 20% of the total — should shift toward activities that build demand that arrives pre-qualified. Brand work. PR. Earned media. Owned channels (newsletter, podcast, community). Distribution partnerships. The activities that show up under "direct traffic" and "branded search" in your analytics, both of which are still converting well.

The second is the quality bar for content. The era when you could publish competent informational content and capture meaningful traffic for it is largely over. The content that still earns clicks in 2026 has one of three properties: it's original data nobody else has, it's a tool that does something useful on the page, or it's transactional content optimised for users who already know what they want. Generic "guide to X" content is competing against an AI Overview that will quote it and keep the click. If your content strategy is producing the latter, stop. Reallocate to the former three.

The third is honesty about measurement. Build the attribution model assuming you cannot see most of what's happening in AI surfaces. Use proxies: brand search volume, direct traffic trends, share of citation in tools you can scrape, qualitative feedback from sales conversations ("how did you hear about us"). Don't pretend you have a click-attribution model for an environment where the click is increasingly optional. You don't, and the people selling you one are not your friends.

That's the practical layer. It's less satisfying than a tactical checklist. It's more useful than one.

The honest limits

This argument has edges worth naming.

The data set behind the 68% number is a clickstream panel. Panels have selection bias. Similarweb's panel, Jumpshot's earlier panels, Datos' panel — they're not the same users, devices, or demographic mixes. The trend line is real and supported by multiple independent sources (Ahrefs' 75,000-site tracker shows a comparable 22% drop in Google's share of referrals year on year), but the absolute precision of "68.01%" should be treated as "roughly two-thirds, give or take a few points depending on methodology."

The "traffic down, revenue up" pattern is real but not universal. I've seen it across multiple client engagements; I've also seen businesses where traffic down meant revenue down, full stop. The pattern requires a brand layer above the SEO funnel. If you don't have one, building one is now the priority — but you can't pretend the headwind isn't real while you build it.

The infrastructure shift around agentic commerce — Stripe Projects, the Apple-Gemini deal — is early. I'm describing the trajectory, not the current state. None of these surfaces are at meaningful volume yet. The argument is that they will be, and the businesses that prepare now will have a structural advantage when they arrive. That's a bet. It's a defensible one. It's still a bet.

And the broader thesis — that informational SEO is in structural decline while transactional and branded search remain defensible — is contested by people I respect. Some argue AI Overviews will plateau, that user behaviour will shift back toward sources, that the regulatory environment will force more outbound linking. Possible. I don't see it in the data. But reasonable people disagree.

Where I'd land

The 68% number is going to be the most-quoted statistic of the year in our industry. It will be used to sell GEO services, AI optimisation tools, brand strategy retainers, and probably a few books. Most of those framings will be wrong, because most of them will treat the number as the story when it's actually the consequence of a longer, more boring story about Google product strategy that's been playing out since 2014.

The actual story is that the open web's role in commercial discovery has been narrowing for a decade, and the narrowing accelerated. The territory that's left — branded search, transactional search, high-intent local — is more valuable per click than the territory that was lost. It is also smaller, more contested, and won by businesses that built brand, owned assets, and direct-to-customer relationships alongside their SEO, not instead of it.

If you're reading this and you're running a business whose entire commercial pipeline depends on Google sending you informational traffic that you convert downstream, the work this year is not better SEO. The work this year is building the channels and assets that let you earn demand before someone types anything into a search box at all. That's a harder, slower, less measurable job than SEO ever was. It's also the only one that compounds in the direction the data is pointing.

Most of the industry will spend 2026 arguing about AI citation tactics. The businesses that come out of this period in good shape will spend it building brands customers ask for by name. Those are not the same activity. They never were. The 68% number just made the difference impossible to ignore.