01 The consumer data
Most clicks go to a site nobody named
Across 10.2 billion Google searches, only about 14% of clicks landed on a site the searcher had typed.
iPullRank measured the Datos clickstream panel: 13.1 billion search events from 9.1 million opted-in people between October 2024 and December 2025. Nearly half of those searches, 46.96%, ended without a click of any kind. The study went up July 30; Search Engine Land pulled a second finding from it on Monday.
The smaller number is the one that should change how you think about your website. Only about 14% of Google clicks went to a site whose name appeared in the search. The other 86% were discovery clicks, where Google introduced someone to a business they never asked for by name.
iPullRank also scored the top 25 destinations by how often Google answers the question on the results page so the click never happens. Wikipedia loses about 30.5% of its would-be searches that way, Reddit only 14.2%. Authority is not what separates them: a reference book can be replaced by someone who has read it, and a courthouse cannot.
Your firm sits on one side of that line page by page. A page of general legal definitions is a reference book. A page carrying your own fee structure and your own account of how a matter actually moves through your county’s court cannot be summarized away, because it exists nowhere else.
Read the panel’s shape before quoting the figures. The sample is 94.5% desktop, so the headline rates are desktop-weighted, and the demographic slice runs one week and skews male. Large and directional, not precise.
What to do about it
Pick the page you would most want an AI to name and ask whether a machine could have written it without you. If it could, that page is competing against the answer box on the answer box’s terms.
Then add what only your firm holds: what your fee actually covers, how long a matter like this really takes in your jurisdiction, what happens in the first meeting. None of it can be reconstructed from somebody else’s copy.
Source
iPullRank, “What 13 Billion Google Searches Reveal About Zero-Click Behavior,” Cate Dombrowski, July 30, 2026 · ipullrank.com · Secondary: Search Engine Land, Anu Adegbola, August 17
02 The research & evidence
The answer used your paragraph and skipped your name
Of 17 passages Google surfaced but declined to cite, not one contained a hard number or a novel claim.
Bart Magera ran 40 prompts through Google AI Overviews and Bing Copilot Search on a single day, July 29, logged 265 citations, and hand-coded 112 passages. Advanced Web Ranking published it on August 13. Google produced an AI Overview on all 40 of the 40 queries, citing 16.8 sources per answer to Bing’s 4.9.
He separated two things most studies collapse. A cited passage gets quoted with a visible link back; an absorbed passage supplies the information and gets nothing. Every lawyer knows the second feeling: an opinion that adopts your argument and cites the other side’s brief.
One page ranked first organically on the exact question its content answered, saw the AI Overview track its definition almost clause for clause, and went uncited while lower-ranked competitors were named.
The coded table says what separated the two groups, and it is not what content teams optimize for. Cited and uncited passages were equally extractive (76% against 71%) and equally likely to be written as definitions. What did separate them: a visible 2025 or 2026 date (80% of cited against 53% of uncited), naming the subject in the first sentence (96% against 82%), and not being a straight restatement of the consensus (61% against 82%).
The bottom two rows carry the finding. Not one of the 17 uncited passages contained a hard number, and not one contained a claim the consensus did not already hold. Everything the engine used without credit was interchangeable with five other pages saying the same thing in the same order.
Hold the percentages loosely. Forty prompts, one person coding them, one collection day, and a Thailand IP with US locale pinned, and absorption is inferred from the answer rather than observed inside Google’s retrieval. Magera publishes the screenshots and coding sheet and invites anyone to re-judge them, which is more than most studies offer.
What to do about it
Open that same page and read the first sentence under every heading. If it starts with “this approach” or “it,” the machine has to work out who you are talking about, and at 96% against 82% the data says it would rather not.
Then put an honest updated date on the page and add one thing the consensus does not contain: a figure out of your own matters, or a dated observation about your local court. Being extractable gets your work used; being distinctive is what gets your name attached.
Source
Advanced Web Ranking, “What Gets Quoted and What Gets Absorbed: A Passage-Level Study of AI Citations,” Bart Magera, August 13, 2026 · advancedwebranking.com
03 The platform change
Your video posts now report to Search Console
Google turned social and video accounts into verifiable properties, with the search queries behind them attached.
Google added a Search Console property type called platform properties, announced on the Search Central Blog on July 7 and rolled out globally on July 29 with a guide to reading the numbers. Search Engine Land worked through the practical use yesterday.
You verify an Instagram, TikTok, X or YouTube account as a property and get the reporting website owners have always had: clicks, impressions, click-through rate, average position, and the search queries that led people to those posts on Google Search and Discover.
The query list is the part worth an hour. A short video answering one narrow question can pull steady Google search demand that never touches your website, and until three weeks ago nobody outside Google could see what people typed to reach it.
Resist filing this under social media reporting. It is search demand data arriving from a surface you do not own, and a query showing volume against a 40-second clip is a query your practice-area page has not answered yet. The catch is that you get the traffic report for a building you do not own: the visibility belongs to the platform’s URL rather than yours, so measuring it is useful and owning the page that answers the question is still better.
What to do about it
If your firm posts video anywhere, verify the account in Search Console this week. It takes minutes and the historical data appears on its own.
Then read the query list as a content brief: anything with real search demand against a clip and nothing on your site answering it is a page you have not written.
Source
Google Search Central Blog, “Platform properties roll out globally, plus a new social and video performance guide,” July 29, 2026 · developers.google.com · Secondary: Search Engine Land, August 18
04 The operational change
A fifth of your AI traffic is filed wrong
Nine months of first-party tracking put AI Overview misattribution to the Direct channel at 22.4%.
Alex Galinos has captured AI Overview referrals for a transportation brand since September 2025 and published the results in Search Engine Land yesterday: 51,200 tracked events across 1,661 cited snippets, running through June 2026.
The method rests on a fragment Google sometimes appends to the destination URL when someone clicks a cited snippet inside an AI Overview. A custom GA4 dimension fires whenever a session arrives carrying it, which is the only first-party signal available while Search Console stays quiet.
Then the finding that rewrites a report: 22.4% of AI Overview sessions were attributed to Direct rather than Organic Search, which is 11,468 events over nine months, with a monthly range running from 16.8% in April to 29.3% in May. AI Overviews exist only inside Google Search, so every one of those was a deposit posted to the wrong account.
The share refuses to sit still. AI Overviews drove 7.53% of organic sessions across the period, peaked near 16–17% in February and March, and fell to 2–4% by the time he wrote it up. Treating a peak month as the new baseline is forecasting off a spike.
One brand, one industry, and Galinos names the rest of the limits himself: the fragment is shared with featured snippets and People Also Ask, and the metric compares events against sessions. The direction holds. The decimal does not.
What to do about it
Ask whoever handles your analytics to add a GA4 custom dimension for that URL fragment. It is a one-time setup and the data is already sitting in your property waiting to be surfaced.
Until then, read your Direct channel as partly organic rather than as a mystery. If Direct grew this year with no campaign to explain it, you have been under-reporting your own search performance.
Source
Search Engine Land, “What 9 months of AI Overview data and 51,000+ tracked events reveal,” Alex Galinos, August 18, 2026 · searchengineland.com
05 The money
Google shortened the window on brand credit
Branded Searches now defaults to seven days, and Performance Max quietly came off the eligible list.
Google updated its documentation for Branded Searches, the conversion type that counts when somebody searches your firm’s name on Google or YouTube after seeing your ad. Hana Kobzova spotted the change yesterday.
Three specifics matter. The default conversion window is now seven days, where the original 2025 announcement described a 30-day view-through window; advertisers can set it anywhere from one to thirty. Performance Max, listed as eligible when the metric launched, is no longer on the list, and YouTube and Demand Gen are.
Google now files Branded Searches under the Consideration goal and treats it as a primary conversion action, but keeps it out of the standard Conversions column and out of bidding entirely; it reports in Results and All Conversions. Brand mapping has to be configured or nothing gets measured.
This lands directly on top of the clickstream finding above. When a branded search ends in a click, that click is roughly 1.3 times as likely to be an ad as on an unbranded search, 4.4% against 3.3%. Those are businesses paying to sit above results they already own, and the stopwatch on that spend just went from thirty days to seven.
If your firm runs video or Demand Gen to build name recognition, the effect is arithmetic. A seven-day default shows less credit for the same money, and a partner reading the dashboard next month without knowing the window moved will conclude the campaign got worse.
What to do about it
If you run YouTube or Demand Gen, check the Branded Searches window before comparing this month against last: two numbers measured over different windows are not a trend. Note where the metric lives now, too. It reports, and Google’s own documentation says it does not bid.
One standing reminder: state bar advertising rules apply to a YouTube ad and to an AI-surfaced recommendation exactly as they apply to a billboard. Comparative claims, results and testimonials carry the same disclosure obligations here, and your compliance review should see the creative before it runs.
Source
Search Engine Land, “Google updates Branded Searches conversion measurement,” Anu Adegbola, August 18, 2026 · searchengineland.com · Secondary: PPC News Feed, Hana Kobzova
