Google's AI Overviews cited the specialist pages, not the general ones

How I measure

I ask Google a fixed list of questions a client might ask in the firm's field: 25 questions on 1 Sep, 35 by 28 Sep. The scans run from an Australian location on desktop, two or three times per question. Each run records whether an AI Overview appears and which sites it links. None of the questions includes the firm's name. [1][1] LayerOps scans through DataForSEO's live Google results API, with AI Overview, location Australia, desktop. 21 Aug and 1 Sep 2026: 25 questions, 2 runs each. 17 Sep 2026: 35 questions, 2 runs each. 28 Sep 2026: 35 questions, 3 runs each. Dates are Sydney time.

What I found

"Cited" means at least one run linked a page on the firm's site. [1]

• Pages in both groups were written to the same standard: every fact sourced to a primary source and reviewed by the firm's professionals before publishing. Primary sources appeared in the AI Overviews for both groups too.
• In the specialist topics, the firm's pages cover narrow questions in detail. One of the cited pages is an original-data page: figures nobody else had put in one place, each with its source.
• In the broad topics, the firm's pages cover ground that official sources, and many other sites, already cover. The broad topics scored 0 in every scan where they were measured: 21 Aug, 1 Sep, 17 Sep and 28 Sep 2026. [1]
• On service-plus-city searches there was mostly no AI Overview to be cited in.

It wasn't the markup

On 28 Sep 2026 I read the structured data on 19 of the site's pages: 12 that the 28 Sep scan cited and 7 it didn't. All 19 carried Article, FAQPage, Person and Organization markup. The one difference points the other way: 3 of the cited pages had no HowTo markup, while all 7 uncited pages had it. [2][2] LayerOps read of the JSON-LD on 19 of the client's URLs, 28 Sep 2026.

That matches Google's own guide to its generative AI features. It says "there's no special schema.org markup you need to add". It asks for "non-commodity content", and describes commodity content as "often based on common knowledge", against content that "provides unique expert or experienced takes that go beyond common knowledge and the ordinary." [3][3] Google Search Central, "Optimizing your website for generative AI features on Google Search", https://developers.google.com/search/docs/fundamentals/ai-optimization-guide. Page last updated 10 Jul 2026. Opened 28 Sep 2026.

Limits

• One site, 35 questions, two or three runs each. The specialist count moves between scans: 8 of 12 on 21 Aug, 10 on 1 Sep, 8 on 17 Sep and 10 on 28 Sep 2026. Answers also vary between runs: in the specialist topics the firm was cited in 18 of 24 runs on 1 Sep and 21 of 36 on 28 Sep. [1]
• Topic and page age are tangled. The new topic's pages went live on the day of the 1 Sep scan. Its first citation came on 28 Sep, on one question, in 2 of 3 runs. If that topic keeps climbing, page age explains more of this than I've said.
• This covers Google's AI Overviews only. I track ChatGPT separately.

What I do with it

Before I write a page, I ask what it can say that official sources and the other sites answering the same question don't: a first-hand process, original data, or a sourced detail nobody has put in one place. If the answer is nothing, I don't write the page.

The AI Visibility Audit applies the same test to a firm's existing pages, and it's how the weekly work chooses what to write next.