Two in Three Google Searches Now End Without a Click. Most Marketing Playbooks Still Chase It.

On July 22, 2025, the Pew Research Center reported that Google users clicked through to a website in 8% of searches that surfaced an AI summary, against 15% of searches that did not. When a link appeared inside the summary itself, people clicked it 1% of the time.

The study followed the real browsing of roughly 900 U.S. adults through March 2025. It measured one thing cleanly. Put a synthesized answer at the top of the results page, and the click that funds most of marketing roughly halves.

That click is not coming back.

Since Google rolled out AI Overviews in May 2024, the share of searches that end without any click climbed from 56% to 69% by May 2025, according to Similarweb data. SparkToro’s clickstream analysis put it at 68% of searches across the first four months of 2026, up from 60% in 2024. Rand Fishkin, who co-founded and runs SparkToro, was blunt about the direction:

“Thanks to AI features, instant answers, UI elements that keep searchers in the results, and shifting user preferences, Google is becoming a walled garden.”

The discovery channel is moving from ten blue links to one answer. Most marketing was built to win a spot in the ten.

For fifteen years the organic playbook was stable. Rank a page, earn the click, capture the visit, measure the funnel from there. AI search breaks the second step. The engine reads the page, lifts the answer, and hands the user a paragraph with a citation the user rarely follows. The page still ranks. The visit never happens.

Some of the lost traffic is reappearing somewhere new, in a smaller and stranger form. Adobe Analytics found that visits to U.S. retail sites from generative-AI sources rose 1,200% in February 2025 against July 2024, and that those visitors converted at higher rates than the site average. The volume is tiny next to classic search. The intent is not. Someone who lands on a site after asking ChatGPT a buying question arrives further down the decision than a keyword tourist.

That combination, collapsing clicks and higher-quality AI referrals, is why a new acronym is eating marketing planning meetings. Generative engine optimization, GEO (or answer engine optimization, AEO, when the target is Google’s own AI answers), is the work of getting cited inside the synthesized answer instead of ranked beneath it. Conductor’s 2026 survey of more than 250 enterprise digital leaders found 94% planning to increase AEO/GEO investment and ranking it their top marketing priority for the year. That is the same budget pressure we tracked when CMOs poured 15.3% of spend into AI capabilities most teams could not yet scale.

The companies with the most to lose are already saying so. HubSpot, which built one of the most-cited content operations on the web, told investors its organic traffic was declining and pointed at AI Overviews answering queries before anyone clicks. Its response was to ship its own answer-engine-optimization tooling. When the company that wrote the modern inbound playbook starts rewriting it, the shift is not a forecast.

The mechanism is a change in who does the reading. In the old model the searcher read your page. In the new one a model reads it, on the searcher’s behalf, and decides in a single pass whether your sentence is the one worth quoting. Optimizing for a human scanning ten headlines and optimizing for a model extracting one factual claim are different jobs. Keyword density, bait titles, and pages engineered to keep a visitor scrolling do nothing when the visitor is an API. What gets cited is clear, specific, and easy to lift: a direct answer near the top, named entities, numbers a model will trust.

Here is the uncomfortable part for anyone running a content budget. The traffic reports will keep looking survivable for a while before the strategy catches up. Non-brand keywords hold their rankings and lose their clicks at the same time, so the dashboards read fine right up until the pipeline does not. Teams that grade themselves on sessions will cut the wrong things. Teams still buying “SEO content” by the article are funding a channel whose payout mechanism changed underneath them. This is the sameness trap we described in AI content saturation, now with the discovery layer stripped out.

Four moves follow.

One, put citation share on the scorecard. Track how often ChatGPT, Perplexity, and Google’s AI answers name you for the questions your buyers ask, and weight that number the way you once weighted rankings.

Two, rebuild your best pages to be quoted, not just visited. Lead with the direct answer, mark up entities and data, and write the one sentence you want a model to lift.

Three, retune measurement toward quality. AI referrals arrive in smaller numbers and convert harder, so a channel that looks like a rounding error on a traffic chart can carry real pipeline.

Four, stop filing this under SEO housekeeping. The discovery channel your whole funnel depends on is being rebuilt around a single answer, and the brand that owns that answer owns the category. That is the last durable advantage in an AI market, and it is worth defending while the click keeps shrinking.

The ten blue links are not dead. They are just where fewer and fewer people look. Budget for where they look now.