Zero-click is old news. Zero-visit is the real crisis.

Executive summary

The SEO industry has spent two years litigating "zero-click search" — the share of queries that end without anyone clicking a result. That debate is now settled enough to be boring: the numbers are bad and getting worse. SparkToro and Datos put U.S. zero-click at 68% of Google searches in early 2026. Inside Google's AI Mode, Semrush measured 93% of sessions ending with no outbound click at all.

But fixating on the click is fighting the last war. The deeper shift is from zero-click to zero-visit: a search economy where the user not only doesn't click your link — they never see your page, because the answer is assembled before they'd ever need to. And in January 2026 the U.S. Patent and Trademark Office granted Google a patent (US12536233B1) describing the endgame in plain language: Google scoring a destination page in real time and, above a threshold, serving its own AI-generated replacement page — in organic or paid results. The page you optimized may not just go unclicked. It may go unrendered.

This piece connects three data streams that are usually reported separately — the click collapse, the publisher traffic decline, and Google's own product-and-patent roadmap — into one conclusion: click-based SEO measurement is becoming structurally obsolete, Google's "traffic is stable" defense is unfalsifiable by design, and the business model underneath (ads that "must be answers") depends on the session not ending in an outbound click. (Confidence: High for the data; Medium for intent attribution, which is argued, not admitted.)

A timeline of the disappearing visit

Part 1 — The click collapse is no longer debatable

The single cleanest dataset is Pew's, because it's a representative panel rather than a tool's keyword sample. In March 2025, across 68,879 real searches by 900 U.S. adults, users clicked a traditional result on 8% of searches that showed an AI summary, versus 15% on those that didn't — nearly halved. They clicked a link inside the summary 1% of the time. And 26% ended their browsing session on a page with a summary, versus 16% without. (Confidence: High — primary, representative panel.)

The tool studies, with their commercial caveats noted, point the same direction and add the time dimension:

• Ahrefs widened its lens to 300,000 keywords and compared December 2023 to December 2025: a 58% CTR reduction for top-ranking pages when an AI Overview is present — nearly double the 34.5% it measured in April 2025. Their framing: "For every 100 clicks you could historically earn for a top-ranking page, Google now keeps 58." (Confidence: High.)
• Seer Interactive tracked the same metric over time: organic CTR on AIO queries fell from 1.76% in June 2024 to 0.61% in September 2025 — a 61% drop. (Confidence: High.)
• Semrush measured the new surface directly: across ~69 million sessions, 93% of AI Mode searches end with no outbound click, versus 83% for AIO-triggered searches and ~60% for traditional ones. (Confidence: High.)

Screenshot to include: A line chart, "Organic CTR with AI Overviews present, 2024→2026," overlaying Ahrefs (−34.5% → −58%) and Seer (1.76% → 0.61%) with a flat dashed line labelled "Google: 'relatively stable.'" The visual gap is the story.

Part 2 — Google's defense is unfalsifiable, and that's the tell

Against this, Google's position has been remarkably consistent and remarkably un-evidenced. Liz Reid's August 2025 post is the canonical text: "Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year," paired with a new metric — "quality clicks" — defined as those where "users don't quickly click back." And: "user trends are shifting traffic to different sites, resulting in decreased traffic to some sites and increased traffic to others." (Confidence: High — verified against the primary Google blog post.)

Examine the rhetorical structure, because it's doing a lot of work:

1. "Stable total clicks" is asserted but never substantiated. Google has not published the aggregate dataset behind it, nor defined the "representative queryset" it used to reject Pew. A claim that can't be checked can't be refuted — which is convenient, not reassuring.
2. "Quality clicks" moves the goalposts from a countable thing (clicks) to an uncountable one (click quality). Even if per-query clicks halve, Google can claim the survivors are better — and no external party can audit it.
3. "Traffic just shifts" frames a system-wide contraction as a reshuffle. But SparkToro/Datos show absolute clicks to any destination fell 22.9% since 2024. That's not redistribution; it's net loss. (Confidence: High.)

Here's the contradiction Google has never reconciled: at Search Central Live Warsaw in June 2025, its own Martin Splitt said the quiet part out loud — "if things are being shown in AI overviews, you will probably see less clicks … clicks are likely to stagnate or fall a little bit." (Confidence: High.) Two Googlers, the same phenomenon, opposite tones: the developer advocate concedes clicks fall; the VP says volume is stable. They can't both be the headline.

Counter-argument, taken seriously: Google's strongest point is real — per-query CTR and aggregate traffic are different things. If total query volume grows fast enough, total clicks can hold even as each query gives fewer. Pichai claims people are "coming back to Search more," and AI Mode passing 1 billion users is genuine volume. So "stable total clicks" could be true while every independent CTR study is also true. The problem is epistemic, not arithmetic: Google asks the ecosystem to accept "stable" on faith while withholding the data, after publicly attacking the one representative study (Pew) that contradicts it. Trust-me-it's-fine is not a methodology. (Confidence: Medium.)

Part 3 — The patent that names the endgame

Every story above is still about clicks. The reason to retire that frame sits in a patent.

In January 2026 the USPTO granted Google US12536233B1 — "AI-generated content page tailored to a specific user," filed July 2024. As reported by Search Engine Land, it describes Google scoring a candidate landing page against a user's query and context in real time and, when a generated alternative would score higher, serving Google's own AI-generated page in place of the publisher's — in paid or organic placements. (Confidence: Medium — the patent grant is verifiable; whether and how Google productizes it is speculative. Patents describe possibilities, not roadmaps.)

Sit with what that means for measurement. Under zero-click, your page still exists and could be visited; the user just didn't click this time. Under the regime this patent describes, the destination page is not rendered to the user at all — Google synthesizes the experience from your content without sending anyone to your URL. There is no click to lose because there is no visit to have. That is zero-visit, and it makes the entire apparatus of click-and-session analytics a measurement of something that no longer happens.

This is also where the business model becomes legible. At Google Marketing Live in May 2026, Google unveiled four Gemini-powered ad formats engineered to live inside AI Mode answers, with VP Vidhya Srinivasan declaring "the best ads must be answers." (Confidence: High.) Connect that to Reid's defense and the shape is unmistakable: a monetization model that captures ad value on the answer itself has no incentive to send the user out. "We still send traffic" is the public line; an answer engine that ends 93% of AI Mode sessions in-product is the operating reality. Google's Q1 2026 Search ad revenue was up 19% to $60.4B, which Pichai credited to AI Mode and AI Overviews — the engine works precisely because the session stays home. (Confidence: High.)

Diagram to include: Three stacked horizontal bars. Ten blue links: user → SERP → click → your page (visit ✔). Zero-click: user → SERP+AIO → (no click) → answer (visit �’). Zero-visit: user → AI Mode → Google-generated page from your content (your page never rendered — visit ✗). Annotate each with the metric that dies at that stage: rank → CTR → the visit itself.

Part 4 — AI referrals won't save you (the math is brutal)

The optimist's rebuttal is that a new traffic source is being born: ChatGPT, Perplexity, and AI Overviews send referrals too. They do, and they're growing fast. Similarweb measured AI referrals to top sites up 357% YoY in June 2025, reaching 1.13 billion visits; ChatGPT referrals to news sites grew roughly 25x in 2025. (Confidence: High.) And the quality is real — Similarweb pegs ChatGPT referral conversion at 7.1%, second only to paid search; Ahrefs found AI-referred users that are ~0.5% of visits drove 12.1% of signups in its sample. (Confidence: Medium — Ahrefs figure is from a single-property sample.)

But scale ends the argument. AI referrals measured in the low billions per month do not come close to offsetting search-referral losses measured against trillions of queries. Digital Content Next's premium publishers saw median referral down 10% in eight weeks; Similarweb shows zero-click rising from 56% to 69% in a year. "AI sends traffic too" is technically true and materially misleading — like noting a bucket while the dam fails. The honest synthesis: the new channel is higher-converting and an order of magnitude too small, simultaneously.

What everyone is missing

The debate is calibrated to the wrong unit. "Zero-click" still implies a SERP with links a user could click. The frontier — AI Mode, and the patent — removes the link, the page, and eventually the visit. Measuring click-through on a surface engineered to prevent clicks is like measuring shelf-space in a store that's becoming a vending machine.

Google's "stable clicks" is a category error dressed as data. Total clicks and per-query clicks can both be cited honestly to opposite conclusions. The dishonesty isn't in the arithmetic; it's in asserting the favorable aggregate while withholding it from audit and discrediting the one representative study that disagrees.

The business model and the data point the same way, which is why the messaging can't. An ad system where "the best ads must be answers" monetizes the session staying inside Google. Every "we value the open web" statement is in tension with a P&L that improves when the visit doesn't happen. You don't need to allege bad faith — just read the incentives.

The strategic response is not "win the click." It's "win the comprehension." If visits are vanishing, the asset is being understood, recalled, and named inside the answer — Fishkin's "zero-click marketing." This is the bridge to brand engineering: when the visit dies, brand presence is the only thing that survives the trip through the model.

Future predictions

• GSC adds AI-surface impressions but not clicks for a long time. Google shipped Gen-AI performance reports in June 2026 showing impressions only — surfacing visibility while withholding the click data that would quantify the loss. Expect that asymmetry to persist. (Confidence: Medium.)
• "Sessions" quietly demote as a north-star metric. Brands shift primary KPIs to assisted conversions, branded search lift, and share-of-answer by 2027. (Confidence: Medium.)
• The patent gets piloted, framed as "rich results." If Google ships AI-generated destination experiences, it will be positioned as a feature, not a replacement — and publishers will discover the difference in their logs. (Confidence: Low — speculative.)
• More publisher litigation, consolidating. Penske won't be the last; expect coordinated or class action as referral data worsens. (Confidence: Medium.)

Practical takeaways

1. Stop optimizing primarily for clicks you can't get. Re-baseline reporting around impressions, share-of-answer, branded search, and assisted conversions.
2. Instrument AI referrals separately and cherish them. They're small but convert near the top of every channel. Tag and nurture ChatGPT/Perplexity/AIO referrals distinctly.
3. *Write to be quoted, not just clicked.* Self-contained, extractable, factually clean passages travel through summaries even when the visit doesn't.
4. Treat GSC's Gen-AI report as a visibility signal, not a traffic report. Impressions there are reach you're getting without a click — manage to it.
5. Build direct relationships that route around search. Email, community, and brand demand are the hedges against a zero-visit funnel.
6. Assume the destination may be synthesized. Structure content so the facts and brand survive extraction even if your page is never rendered — because the patent says it might not be.