Optimizing for Searches That Never Send a Click

Here is the uncomfortable claim this page defends: for most informational queries, your website's job is no longer to receive the visit. Its job is to supply the answer that gets shown to someone who will never arrive. That sounds like defeat. Measured correctly, it is closer to a change of currency. The businesses that adapt are not fighting to recover 2019's click volume; they are competing to be the name inside the answer, and to capture the smaller pool of clicks that still happen, which now carry more intent per visitor than they ever did.

This is the thesis piece behind a lot of our other work, so it earns the full treatment: what the studies actually say, why the studies get attacked, what AI Overviews change, and what to build.

What a zero-click search actually is

A zero-click search is a query that ends on the results surface. The searcher asked, the engine answered, nobody's website got a visit. The answer can come from an AI Overview, a featured snippet, a knowledge panel, a local map pack, a currency converter, or a sports score module. From the searcher's chair this is simply good service. From a publisher's chair it looks like the referee started keeping the ball.

Two things make the 2026 version different from the featured-snippet era. First, scale: answer surfaces used to sit on a minority of queries, and now the no-click outcome is the majority outcome. Second, synthesis: an AI Overview does not lift one snippet from one page, it composes an answer from several sources and attributes fragments back to them. That second change is what makes zero-click a strategy question rather than a formatting trick. When the engine writes the answer, the competition is over who it reads and who it names.

The 2026 numbers, stated precisely

The reference dataset right now is SparkToro's June 2026 study, built on Similarweb desktop and mobile panel data covering US Google searches from January through April 2026. Its headline: 68.01 percent of Google searches ended without a click to the open web, up from 60.45 percent in 2024. That 7.56-point jump in two years is the fastest movement SparkToro has recorded since it began tracking the metric.

The pressure source is measurable too. Ahrefs trigger data cited in the same study puts AI Overviews on more than 20 percent of all Google searches, and Ahrefs' click-through analysis found that when an Overview appears, CTR to the results below drops by roughly 60 percent. Pew Research Center reached a compatible conclusion from a different direction: across 68,879 real searches from its March 2025 panel, users clicked a traditional result on just 8 percent of searches that showed an AI summary, against 15 percent when none appeared, and users were more likely to end the entire session after seeing a summary (26 percent versus 16 percent).

Different panels, different months, different definitions. Same slope.

The fight over the numbers, and why it matters to you

Quoting a zero-click statistic without its dispute history is how agency blogs end up repeating numbers they cannot defend, so here is the dispute history. Google publicly rejected the Pew study, calling it built on "a flawed methodology and skewed queryset" that does not represent real Search traffic. Google executives have argued through 2025 and 2026 that overall clicks to the web are stable and that AI features send "higher quality" clicks. The panel studies have real limitations Google is pointing at: clickstream panels over-represent certain user types, and a search that ends in a phone call from the map pack is a business win recorded as a zero-click.

Here is the part the dispute does not touch. A 2026 randomized field experiment with 1,065 desktop Chrome users, the strongest study design yet applied to the question, found that showing an AI Overview cut outbound organic clicks by 39.8 percent and raised zero-click outcomes by 34.5 percent. Independent measurements now range from roughly 15 percent CTR loss to nearly 90 percent depending on query set and method, and not one of them found clicks increasing. When every flawed instrument points the same way, the reasonable engineering conclusion is that the effect is real and the magnitude is uncertain. Plan for the direction, argue about the decimal later.

AI Overviews are the accelerant, not the origin

Zero-click behaviour predates generative AI; knowledge panels and snippets were absorbing queries a decade ago. What AI Overviews changed is the class of query that gets absorbed. A snippet could answer "how tall is the CN Tower." An Overview can answer "should I repair or replace a 15-year-old furnace in southern Ontario," which is precisely the research-stage question that used to earn a service business its first visit from a future customer.

The absorption is climbing the funnel, in other words, and it is doing so on the exact queries where local businesses built their content strategies. The one lever a business controls inside that surface is citation: Overviews name and link the sources they draw from, and the sources named are frequently not the pages ranked directly below. Getting selected into that citation set is its own discipline, and we maintain a dedicated playbook for it in our guide to AI Overviews optimization. The short version: eligibility comes from ranking-adjacent signals, but selection comes from having a passage the model can lift cleanly.

The click that survives is worth more than the click you lost

Run the funnel math on what a zero-click surface actually filters out. The visitor who only wanted a definition, a price range, or a yes-or-no was never going to become a customer; the answer box now serves them without costing you server time or attention. The person who clicks through an AI Overview citation has read a synthesized answer, seen your business named as its source, and decided they need more than the summary. That is not a colder version of your old traffic. It is a pre-qualified version of it.

Matt has been watching this in client analytics since Overviews went broad:

This reframing has a hard edge, though. Brand impressions inside answers only convert later if the brand is actually named. Anonymous absorption, where the Overview uses your explanation and cites a bigger publisher, is the true loss scenario. Which is why the strategic response is not "get clicks back." It is "be the citation."

The strategic response: be the answer, not just a result

Once the majority outcome is no click, three prizes remain on the table, and content should be engineered for all three at once:

• The citation. Your brand named inside the AI Overview, the ChatGPT response, the Perplexity footnote. This is the new first impression, and it compounds: engines that repeatedly retrieve you start treating your domain as the entity for the topic.
• The impression. Even uncited, appearing on the answer surface builds the recognition that later shows up as branded search. Watching branded query volume rise while generic clicks fall is the classic signature of zero-click exposure doing quiet work.
• The high-intent click. The reader the summary could not satisfy. Pages should assume this visitor arrives pre-briefed and get them to depth, proof, and a next step fast, rather than re-explaining what the Overview already told them.

None of this replaces conventional ranking work; it sits on top of it, because retrieval still begins at the index. Within our own methodology this page's territory is Vector 4, Embed: writing the answers engines extract. The broader engine-by-engine practice, covering ChatGPT, Perplexity, Gemini, and Copilot rather than Google alone, is laid out in our AI search engine optimization guide.

Structure for extraction: write for where machines actually read

Extraction has mechanics, and the mechanics are documented rather than guessed. The Stanford study behind the "lost in the middle" finding (Liu et al., 2023) tested how language models use long inputs and found a U-shaped curve: models handle information best at the start or end of what they read, and accuracy drops sharply for material buried in the middle. In the authors' words, "performance is often highest when relevant information occurs at the beginning or end" of the input. A model deciding whether your page answers a question is subject to exactly this bias, which converts directly into editorial rules:

• Answer first, elaborate second. The opening sentence under every heading should resolve the heading's question outright. Context, caveats, and colour come after. Burying the verdict in paragraph four places it in the exact zone the research says machines read worst.
• Make every passage self-sufficient. Engines quote fragments, so a paragraph that relies on the two above it for meaning cannot be cited without distortion. One claim, its number, and its source, together in one place.
• Give headings honest question shapes. Retrieval matches queries to sections, not just to pages. A heading that states its topic plainly is a retrieval hook; a clever one is camouflage.
• Date your evidence inline. "68 percent (SparkToro, June 2026)" is liftable and checkable. "Most searches" is neither.

We ran our own investigation into which regions of a page different AI engines actually pull from, and the positional findings lined up with the academic curve closely enough to change how we draft every client page. The full write-up is in our research note on where AI engines read.

Label the answers with markup engines can parse

Structure in the prose gets you read; structure in the markup gets you understood. FAQPage, Article, and Organization schema tell retrieval systems what each block is, who wrote it, and which entity stands behind it. Since Google's May 2026 update the non-negotiable rule is symmetry: the questions and answers in your schema must match the visible ones word for word, because asymmetric pairs now cost the rich result. A minimal, valid FAQPage block looks like this:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "What counts as a zero-click search?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "A search where the user gets the answer on the results page and never visits a website."
    }
  }]
}

Markup is not a ranking bribe. It is disambiguation: it removes the guesswork between what your page says and what a machine records your page as saying, and in an answer economy that gap is where citations get lost.

Measurement has to change before the strategy can

A zero-click strategy run on click-era dashboards will always look like failure, which is how good programs get cancelled. The metrics that describe reality in 2026:

• Impressions versus clicks in Search Console. Impressions holding or rising while clicks fall is exposure moving onto the answer surface, not demand disappearing.
• Branded search volume. People who met you inside an answer come back by name. Branded queries are the delayed echo of uncounted impressions.
• Citation presence per engine. For each revenue-relevant question, log whether Google's Overview, ChatGPT, Perplexity, and Gemini name you, monthly. Each engine retrieves differently, so presence in one says nothing about the others.
• Lead intent, not lead volume. If the surviving clicks are pre-qualified, close rate and time-to-close should improve even as sessions decline. That pairing is the proof the model is working.
• Direct and dark traffic drift. Some answer-surface visitors arrive later by typing your URL. A slow rise in direct visits alongside falling organic sessions belongs in the same story.

The full instrumentation, including how to log AI citations without expensive tooling, is in our walkthrough on how to measure AI visibility.

The honest counterweight: what zero-click optimization cannot do

Three limits, stated plainly, because a thesis you cannot falsify is marketing rather than engineering. First, some business models genuinely lose: ad-supported publishers and affiliate sites monetize the visit itself, and a citation pays them nothing. Their problem is structural, and answer-surface presence only softens it. Second, citation selection is probabilistic. Extraction-friendly structure, sourced claims, and clean markup raise the frequency with which engines pick you; nothing forces a specific model, on a specific day, to name you. Anyone quoting a certainty here is selling one. Third, the data itself is contested territory, as covered above. We treat the 68 percent figure as the best current panel estimate with known biases, not as gospel, and we re-verify it before every planning cycle because these numbers have moved 7 points in two years.

What survives all three caveats is the core trade: the work that earns citations (real expertise, extractable structure, verifiable sourcing) is identical to the work Google's quality systems reward in ranked results. There is no scenario in the current data where that investment strands.

Frequently asked questions

Sources

Where to go from here

If your analytics show the slide and you want a second set of eyes on what it means for your specific market, talk to us. We will tell you which of your queries are already absorbed, which are defensible, and what the citation play looks like for your niche. And if the answer is that your situation does not need an agency, we will say that too; a practice built on answer-surface trust cannot afford to earn distrust in its own funnel.