Two thirds of US searches end without a click. Position is no longer a traffic report

The useful part of the SparkToro figure is not its size, it is what it does to your reporting. If 68.01% of US Google searches ended without a click across the first four months of 2026 ^[1], then clicks happened on 31.99% of them ^[11], and an average-position column is now describing a surface where most impressions were never in the market for a session.

Google's own line is that AI Overviews and AI Mode are grounded in the core Search ranking and quality systems ^[3]. Read that as an accounting change rather than a reprieve: the ranking work still lands, but part of the return arrives as being the source named in an answer, not as a row in analytics. Nobody hands you that number. The checklist this comes from is entirely page-side advice, on answer placement ^[7], question-shaped headings ^[8], first-hand evidence ^[9], and crawlability and mobile hygiene ^[10]. Instrumentation is left to you.

Two counters are worth building. First, citation rate across the query set you actually care about: how often your domain is the source cited in the answer, tracked over time. Second, what an arriving reader does after the answer. That one follows directly from the advice to put the direct response in the opening paragraph rather than behind a long introduction ^[7]. A visitor who clicked through has already read that paragraph in the results, so everything below it is the only part still capable of earning attention. Bounce rate measured against impressions will not see this. Depth of visit on the sections beneath the answer will.

The headline number deserves less respect than its second decimal invites. The same writeup that reports 68.01% warns that cross-study comparisons need care because panels differ ^[2]. A hundredth of a percentage point out of a panel estimate is a direction of travel, not a baseline you can calibrate a target against ^[1][2]. Your own clicks-per-impression ratio is a panel of one, it is yours, and it moves for reasons you can go and look at.

What will not help is buying a technical shortcut. Google says site owners need no AI-only markup and no llms.txt file to appear in Search ^[5], and its recommended list is the familiar one: valuable original content, clear organization, crawlability, good page experience, accurate technical implementation ^[4]. Which pushes the differentiator onto material that cannot be regenerated from other people's pages. A real theme and the rule that broke its layout, a before-and-after screenshot, a note on what failed during testing ^[9]. That costs tester hours, and the fact that it costs tester hours is the reason it survives being compressed into a two-sentence summary.

The reframing in the source is that a page should be discoverable, understandable, worth citing, and valuable enough that someone continues reading after the short answer ^[13]. Only the first of those four is what a rank tracker measures. Position still tells you whether you are eligible to be quoted. It stopped telling you whether anyone showed up.