Search your company name. Google may already be answering for you.

Search your company name in a logged-out window. By the last week of September, Google was showing an AI Overview on most of the branded queries the measurement firms track. On 1 September that share, in DemandSphere's set, was 26%. On 27 September it was 90%. It did not climb in a straight line, and Google has not announced a change. The practical fact is simpler: the sentence under your name is no longer guaranteed to be your title tag.

We wrote a GEO playbook in March, when the fear was "rank does not carry into the answer." September's measurements split that fear in two. On Google's own AI Overviews, rank is still the main road in. On ChatGPT, it mostly is not. Treating those as one channel is how this year's budgets get wasted.

The branded jump is real, and it wobbled

DemandSphere tracked branded keywords once a day and counted an AI Overview whenever one appeared, whether or not it mentioned the brand. Through most of September the rate sat between about 27% and 35%. It spiked to 57% on 18 September and was back near 31% the next day. Then it jumped for good: 69% on 26 September, 90% on the 27th, 80% on the 28th, 82% on the 29th.

Other counts land in the same place and disagree about the size of the jump, which is what you want from independent data. Ahrefs' US desktop crawls, as reported by Semrush on this story, already had AI Overviews on about 65% of branded keywords in August and about 83% on 29 September. Chris Long at Nectiv checked 100 enterprise brands with SerpApi and found an overview on 93 of them, including companies much smaller than the usual giants. Search Engine Land's write-up is careful about the limit that matters: none of these studies measured clicks. A summary above your name can steal the visit. It has not been shown, in this dataset, that it did.

So do not brief the board that "branded CTR fell 40% in September." Do this instead. In Search Console, filter queries that contain your name. Compare click-through rate — not raw clicks — for 8–14 September against 29 September–5 October. Demand was not flat, so clicks alone will lie to you. If CTR on brand terms dropped after the 26 September jump, you have a local fact. If it did not, you have a monitoring habit and no crisis.

Google's overview is a search result with a narrator

On 15 May 2026 Google published Optimizing your website for generative AI features on Google Search. The line worth taping to the monitor: optimizing for generative AI search is optimizing for the search experience, and thus still SEO. They define the machinery in two steps.

First, retrieval-augmented generation. A normal ranking pass pulls pages from the index. The model writes from those pages and links to them. Second, query fan-out. The model issues related searches — their example is a weedy lawn splitting into herbicides, non-chemical removal, and prevention — and each of those searches returns its own pages. A citation can arrive because you rank for the question the model asked next, not only the question the human typed.

Eligibility is boring, and boring is the gate. The page has to be indexed and allowed to show a snippet. noindex, nosnippet, and max-snippet:0 keep you out. Since 31 August 2026 there is also a Search generative AI control in Search Console, rolled out worldwide with a performance report. The guide says a site must be included there to be eligible. The control is on by default. Opting out drops impressions and traffic from AI Overviews and AI Mode, and Google has said it is not a ranking signal for ordinary results. For a company that wants to be named, leave it on. Check that nobody turned it off during a panic in August.

The new report shows impressions from AI Overviews and AI Mode, by page, country, and device. It does not show clicks, queries, or a separate position. Those impressions already sit inside the regular Web totals. Add the two reports together and you will invent traffic.

What 400 overviews actually cited

Kyle Eggleston pulled 400 US results that showed an AI Overview, across ten industries, from Ahrefs snapshots taken between 1 and 24 September. He matched 1,561 outside links against the organic results on the same page. This is one person's study, one country, desktop snapshots, and it counts the links Google shows first rather than every source in an expanded panel. It is also the cleanest public count we have of the question clients keep paying to have answered with a vibe.

Of those 1,561 links, 51% pointed at a URL already in the organic top 10 for that exact query. Counted by domain, 63.7% came from a site already on page one. In 347 of the 400 overviews, at least one cited page was a top-10 result.

Position 1 is cited more often than not. By position 4 the odds have fallen by more than half. That is not a new ranking system. It is the old click curve, read by a model.

The other half of citations are the fan-out. They are not obscure pages. In his sample, pages cited from outside the top 10 still ranked for something: almost all of them ranked in the top 100 for at least one keyword. Surfer, in a separate 10,000-keyword study he cites, found pages that ranked for both the main query and the neighboring ones were cited 51% of the time, against 20% for pages that ranked for the main query alone. The move is to answer the next questions on the same strong page. The move Google's guide treats as spam is minting a thin URL for every variation so you can "cover fan-out." That is scaled content abuse, and it is a bad bet even if nobody files an action, because forty thin pages do not become a source.

Two cuts in the same study should change what you publish:

• Question-shaped queries triggered an overview 86% of the time in his wider pool of 3,804 keywords. Queries with local intent triggered one 23% of the time. If your entire business is "plumber near me," this is mostly not your problem yet. If buyers ask "how do we automate intake," it is.
• On software queries, YouTube was 31.6% of citations. On finance queries it was 1.8%. Health barely cited Reddit at all. If you sell software, a short how-it-works video is not a brand exercise. It is a third of the slots in that sample.

Landing pages were cited slightly more often than articles (37% versus 31% of the organic results he checked). The overview is not allergic to commercial pages. It is allergic to pages with nothing specific enough to footnote.

What to stop buying

Google's May guide has a section titled, without poetry, "Mythbusting generative AI search." From Search's point of view, answer-engine optimization and generative-engine optimization are names for SEO. The tactics they tell you to ignore:

ChatGPT is the part of the March post that still stands. Ahrefs' August 2025 overlap study, about 15,000 prompts, found roughly 8% of citations from ChatGPT, Gemini, and Copilot also sitting in Google's top 10 for that prompt, versus about 29% for Perplexity. Rank is the floor for Google's AI surfaces and a weak predictor for ChatGPT. You need both habits. You do not need a vendor who pretends they are the same habit with a new acronym.

What to do before the next crawl

1. Search the brand, logged out, and save the overview. What does it claim? Which URLs does it cite? If the cited page is a two-year-old directory listing and your homepage says nothing a model can lift, that is the bug. Fix the sentence: what you do, for whom, and where. The same sentence is what an agent visitor needs when there is no human to infer it from the hero video.

2. Open the Generative AI performance report. Note which URLs are earning impressions. Do not add them to your Web totals. Confirm the inclusion control is on.

3. Run the branded CTR comparison described above. One before-and-after, then monthly. Guessing from industry charts is how you fund the wrong project.

4. Pick one money page and answer the next three questions on it. Not forty new URLs. A heading and a direct first sentence each. Publish a number you actually have — a price range, a timeline, a constraint — because a fact only you know is a fact the model can only get from you.

5. If you sell software or a how-to, put the demonstration on YouTube as well as on the site. In that vertical the overview is already spending a large share of its links on video.

6. Keep the non-Google log. Once a month, ask ChatGPT and Perplexity the ten questions a buyer asks before they call ("how much does X cost," "X versus Y," "who does X for a small team"). Record whether you are named. That sheet is still the dashboard for the channel where rank does not save you. It is also the unfashionable half of growth engineering: measurement first, then the page, then the follow-up when the warmer click arrives.

Google AI Overviews  →  rank the page, then give it a sentence worth citing
ChatGPT and similar  →  be the source the model can corroborate, rank optional
Your brand name       →  read the overview, fix the sentence, watch branded CTR

Anyone selling a file drop, a schema package, or a bag of mentions as "AEO" is selling around the measurement. The work that shows up in both the blue links and the overview is older than the acronym: a page that ranks, says something specific, and has other pages willing to point at it. That is the work. The overview just grades it in public, including on your own name.