Google's AI Overviews roughly tripled on brand-name searches in September

For years, the one Google search a company could count on owning was its own name. In September 2026, AI Overviews on branded queries roughly tripled, according to DemandSphere data cited by Search Engine Roundtable. The publication also saw the AI summaries showing up much more often on searches for large brand names.

An Overview gets built from sources across the web. A company that ranks first for its own name still holds the top link, but that does not necessarily mean it controls the entire experience surrounding its brand. For the head of digital or brand director who reports on search, that changes what a good branded result means.

The branded jump was one of several September changes to Google Search. Google also launched its fourth confirmed spam update of the year. Together they leave brand teams with a practical question: what should they measure when Google answers the brand question itself?

Brand searches now get an AI summary too

The branded surge adds to already fast general growth. Comscore data published in September and reported by Search Engine Land showed AI Overviews appearing on 39.4% of U.S. desktop Google searches as of June 2026. That share stood at 25.8% in July 2025.

Branded queries used to sit outside that trend. Someone typing a company name usually wanted the company's site, and the company's site usually came up first. With AI Overviews roughly tripling on branded queries in September, the brand-name results page, the one search page most companies assumed they owned, is now partly written by Google from other people's websites.

Most of the work that shapes that summary happens away from the company's own site. Writer's July 2026 enterprise guide, by Alaura Weaver, calls SEO mostly a first-party game, where a company optimizes the pages it owns. It describes generative engine optimization, or GEO, as mostly a third-party game, built on brand authority, presence across outside sites and original data. The guide puts the split at 80% strategic, meaning positioning, ecosystem presence and brand authority, and only 20% technical.

LSEO's roundup lists the factors gaining weight for branded search: a clear entity, reputation, mentions from authoritative sources, reviews, coverage by third parties, and accurate information across the wider web. Read against the 80/20 split, that list suggests the branded Overview is mostly a reputation problem with a technical layer, not the other way around.

The research behind the GEO label supports the same approach. In a guide on Progress Software's Sitefinity blog, Zach Stone of DeltaV Digital points to the November 2023 paper by researchers from Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi that coined the term. That study found targeted optimization could lift visibility in generative responses by up to 40%. Citations, statistics, expert quotations and authoritative language did best. Keyword stuffing did nothing. Those are familiar SEO habits, now applied to answers that a brand doesn't write.

The technical 20% still has a clear first priority in Stone's view. He calls structured data the most impactful action this year, because large language models rely on it even more than traditional search engines do. His test is simple: a human reader, Googlebot and an LLM visiting the same page should each come away with the same understanding of it.

The clicks behind the answers

Click-through studies collected in the Writer guide measured the effect on individual keywords. Ahrefs analyzed 300,000 keywords, comparing December 2023 with December 2025, and found that where an AI Overview appears, click-through rate for the top-ranking page falls by up to 58%, from 7.3% to 1.6%. Seer Interactive's separate study of 3,119 informational queries found organic CTR on AI Overview queries fell 61%, from 1.76% to 0.61%.

Two methods and two keyword sets landed in roughly the same place. The top result keeps its position and loses well over half its clicks. Both studies measured general and informational keyword sets rather than brand names, so for brand teams they work as a reference point for what an Overview can do, not a forecast for their own name.

Site-level numbers point the same direction. Chartbeat figures cited by Search Engine Journal show that Google Search referrals to publishers in its network dropped 40.2% year over year between July 2025 and July 2026. Traffic from Google Discover also fell substantially. A separate experiment with more than 1,000 participants, which Search Engine Journal and Search Engine Land both covered, found that people placed in Google's AI Mode were less likely to click through to outside websites.

That experiment also complicates a comfortable assumption. The AI Mode group reported lower trust and satisfaction than people using conventional Google Search. So a missing click can't simply be read as a happy user who got what they needed, which matters for any brand inclined to treat a lost session as harmless.

LSEO's roundup is careful to say this does not mean SEO traffic is disappearing universally. Even so, zero-click behavior was common before the branded surge: Arc Intermedia, writing in December 2025, cited research that nearly 60% of Google searches end without a click. The branded data still makes one point clear: traffic is no longer a complete measure of organic visibility. A brand search that ends with a reader satisfied by Google's summary produces no session.

Metrics beyond the click report

Two measures keep coming up. Stone argues that reference rate, meaning how often content gets cited in generative answers, matters more than click-through rate for AI responses, and that securing citations in AI Overviews is now worth more than chasing the clicks those Overviews replaced. Writer's guide points to share of model, a term coined by Jack Smyth and Tom Roach. It measures how often a brand appears in AI-generated answers compared with competitors, as a successor to share of voice.

Unlike paid share of voice, Writer argues, share of model is earned. ChatGPT now sells clearly labeled "Sponsored" cards below its answers, but the guide says those ads don't influence the model's response. Or, in Weaver's phrasing: "You can buy placement. You can't buy a recommendation."

Few companies track either one. Writer cites McKinsey research finding that just 16% of brands systematically track their AI search performance. For a team whose organic report is still sessions plus rank positions, the September data suggests traffic can no longer be the only measure of organic visibility.

The gap matters because of what Writer calls "silent shortlists," preferences buyers form in AI conversations before they ever visit a website. By the time someone lands on the site, they may already have put the brand on the shortlist or crossed it off. A branded Overview is one more place where that shortlist could take shape.

The rest of LSEO's roundup points the same way. Brands, it argues, need to know whether they appear in AI answers, get mentioned by name, earn citations, make recommendation shortlists and influence searches or conversions that happen later. Most of that list never shows up in an analytics dashboard. Google did add new AI and multimodal reporting capabilities to Search Console in September, according to the same roundup, alongside expanding AI Mode and pushing further into agentic commerce.

A spam update that ran almost two weeks

Any analysis of September rankings also has to factor in Google's spam update. Google launched it globally and in all languages on September 24. It was the fourth confirmed spam update of the year, following March, June and August. Google said the rollout could take up to two weeks and used almost all of that window, completing on October 8 after roughly 13 days and 16 hours. By comparison, each of Google's three earlier spam updates this year wrapped up in under three days.

Search Engine Roundtable reported heavy volatility on September 23 and 24. It saw another big swing over the spam update's first weekend and more fluctuation as the rollout continued. The month was already noisy before the update began. Industry observers flagged unconfirmed volatility around September 4-13, then again on September 15. That means a branded-query report built on September data likely reflects several ranking changes that overlapped.

Google also disclosed research on SAFE, short for Scaled Abuse Forensics Examiner, a system built to detect forms of AI-generated abuse. Search Engine Journal pointed out how the timing lined up with the spam update. Google hasn't confirmed that SAFE was responsible for it.