How Google's AI Overviews Actually Decide What to Cite

For most of 2025, businesses trying to get cited by AI Overviews were guessing. Consultants sold checklists built on speculation, agencies invented new acronyms, and nobody could point to a single sentence from Google that confirmed any of it. That changed on June 29, 2026, when Google published its own official guide, "Optimizing your website for generative AI features," directly on Search Central (Google Search Central, updated June 29, 2026). It answers the question plainly, and the answer is less complicated than the industry around it.

This is a companion piece to our broader breakdown in How AI Actually Recommends Businesses. That article covers the mechanics across every major AI tool. This one goes deep on the tool that touches the most searches: Google's AI Overviews and AI Mode.

Google said the quiet part out loud

The core message of the guide is that generative AI features on Google Search are "rooted in core Search ranking and quality systems," built using retrieval-augmented generation and a technique Google calls query fan-out, where a single question gets broken into several related searches before an answer is assembled. In plain terms, AI Overviews are not a separate system with separate rules. They pull from the same index, the same crawling, and largely the same quality signals as regular Search results (Google Search Central).

Search Engine Journal's coverage of the release put it directly in its headline: Google's new guide calls AEO and GEO "still SEO" (Search Engine Journal, 2026). That is a meaningful statement from the company running the largest AI answer surface in the world. A great deal of the AEO industry has been selling a new discipline. Google is saying it built on the old one.

What Google says does not matter

The guide is unusually direct about what businesses can stop worrying about. On structured data specifically, Google states: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." It goes further, saying sites don't need to create "new machine readable files, AI text files, markup, or Markdown" such as the llms.txt files a segment of the AEO industry has been recommending, and there is no requirement to break content into small chunks for AI to read more easily (Google Search Central). We cover the schema question in more depth in a separate article on what a controlled 2026 study found about schema and AI citations, and the short version lines up with what Google is saying here.

Google also pushes back on chasing inauthentic mentions across the web to manufacture the appearance of authority, calling that approach ineffective. The guide's framing is consistent throughout: less technical gaming, more content that earns its place.

What actually predicts a citation

A separate large-scale analysis, covering 30 million cited sources across ChatGPT, Google AI Mode, Gemini, Perplexity, and AI Overviews, gives a clearer picture of what shows up in practice. Reddit was the single most-cited source across all five systems, with YouTube, LinkedIn, Wikipedia, and Forbes rounding out the top five. Within that, Google's own AI features leaned differently than the others, favoring platforms like Facebook and Yelp alongside its usual web index, reflecting Google's access to first-party review and local business data that other AI tools don't have (Search Engine Land, March 2026).

The pattern underneath all of it is consistency: independently published, verifiable, specific content beats brand-authored claims. Google's guide names this directly, urging sites to create "non-commodity content that's helpful, reliable, and people-first," and cautioning against content that simply recycles "what others on the internet have already said, or could easily be produced by a generative AI model" (Google Search Central).

The prerequisite nobody skips

One detail in the guide matters more than it might first appear: a page has to be eligible for regular Search snippets before it can appear in an AI Overview at all. AI Overviews draw from Google's existing index, so a page that isn't well indexed, isn't crawlable, or doesn't already perform reasonably in organic Search has effectively no path into an AI Overview citation. Standard technical SEO, indexation, and site health remain the floor. AI visibility is built on top of that floor, not instead of it.

What query fan-out actually means for a small or mid-size business

Query fan-out sounds abstract until you see it in practice. When someone asks an AI system "who's the best functional medicine provider for thyroid issues," Google's systems don't just search that exact phrase. They quietly generate and search related questions too, things like "what does a thyroid specialist actually do differently" or "how do I know a functional medicine provider is legitimate," then pull the strongest answer to each sub-question into one combined response. A business only shows up in that combined answer if it has content that answers one of those fanned-out questions specifically and well. A vague homepage claiming "personalized thyroid care" answers none of them. A named patient describing exactly what changed in their treatment answers one of them precisely.

What this means for a founder-led business

Strip away the acronyms and Google's own guidance points in one direction: publish specific, verifiable, people-first content and stop treating AI visibility as a technical workaround. For a founder-led business, the most valuable version of "non-commodity content" is rarely a blog post restating industry knowledge everyone already has. It is a real customer, named, describing a real result, in language nobody could have generated by prompting a model. That is the content shape Google's own guide is describing, whether or not it uses those words.

In practical terms, that points to three priorities ahead of any technical AEO checklist:

• Make sure the business is properly indexed and technically healthy in regular Search first, since nothing else matters if that foundation is missing.
• Replace generic claims on key pages with specific, attributed customer outcomes that answer the real sub-questions a prospect, or Google's query fan-out, would actually ask.
• Publish those stories consistently rather than once, since a single case study answers one fanned-out question while a growing library answers dozens.

This is the same mechanism behind the Trust Flywheel: every authentic story you publish becomes a piece of evidence an AI system, or a skeptical prospect, can actually check. You can see what that looks like across more than 1,500 businesses at our case studies page, and dig deeper into the full comparison of how different AI tools weigh sources in our breakdown of ChatGPT, Claude, Perplexity, and Google AI Overviews side by side.

FAQ

Is Google's June 2026 guide an official statement, or industry speculation?

It is official. It was published directly on Google Search Central, the same channel Google uses for its regular Search documentation, and is dated June 29, 2026.

Does this guide apply to AI Overviews only, or AI Mode too?

Both. Google's guidance covers its generative AI search features broadly, including AI Overviews and AI Mode, since both draw on the same underlying ranking and retrieval systems.

If schema markup isn't required, should a business remove it?

No. Google recommends keeping structured data in place for its established purposes, like rich results eligibility in regular Search. The guidance simply clarifies that adding schema is not a shortcut to AI Overview citations.

Does being cited in a Google AI Overview require ranking on page one?

Not exactly, but it requires being indexed and eligible for standard Search snippets. Research shows a meaningful share of AI Overview citations come from outside the top 100 organic results, so ranking isn't everything, but basic indexation and technical health are non-negotiable.

What is the single biggest thing Google's guide asks businesses to do differently?

Stop producing content that repeats what is already available elsewhere or that a generative model could produce on its own, and instead publish content built on direct, verifiable experience.

How does this connect to customer stories specifically?

A named customer describing a specific, checkable outcome is close to the exact opposite of recycled, AI-producible content. It is unique, attributable, and grounded in real experience, which is precisely what Google's guide says it wants to see more of.