// On this page
Google AI Overviews now sit above the regular results on roughly half of all searches, and the handful of sources they cite collect the attention that used to flow to the blue links. Here's the uncomfortable part: ranking number one no longer guarantees you're in the answer. Only 38% of pages cited inside an AI Overview rank in the top 10 for that query. The page that gets quoted isn't always the page that ranks. It's the page that's easiest to extract.
So this is a structure problem, not a rankings problem, and structure is repeatable. Once you understand how Google's system pulls an answer apart and reassembles it, you can build pages that hand it exactly what it wants. This is how to do that, section by section, with the patterns citation research keeps surfacing.
How do Google AI Overviews choose what to cite?#
Google AI Overviews extract short, self-contained passages from pages that read as clear and authoritative, then stitch them into a synthesized answer. The system pulls content in roughly 130 to 160 word blocks, favors text that answers the query early, and prefers definitive phrasing over hedging. The unit that wins a citation is a passage, not a whole article.
This is the mental shift that changes everything else. Google isn't ranking your page and handing the top result to its AI. It's reading your page in chunks and asking, of each chunk, "is this a clean, quotable answer to one of the sub-questions I'm working on?" Duane Forrester's framing captures it: the model moves from crawl-index-rank to chunked-retrieved-synthesized. Your job is to make every chunk independently liftable, because the model rarely quotes the whole thing. It quotes the paragraph that stands on its own.
Two findings from citation research anchor the rest of this guide. First, position matters: an analysis of AI citations found 44.2% came from the first 30% of a page's text. Put the answer early. Second, language matters: cited text is nearly twice as likely to use definitive phrasing as hedged phrasing (36.2% versus 20.3%). Make the claim, then support it.
Where should you put the answer on the page?#
Put the direct answer in the first one or two sentences of every section, and lead the whole page with a complete answer. Google's system front-loads its extraction: 44.2% of AI citations come from the first 30% of page text. Burying the answer under a long preamble is the most common reason a strong page never gets cited.
Most marketing writing is built backwards for this. We open with context, warm the reader up, build to the point. That structure is fine for a human who chose to read, and invisible to a machine scanning for a liftable answer in the first few lines. The fix isn't to delete the context. It's to lead with the answer, then add the context underneath for the reader who wants it.
Practically: under every question-heading, the first paragraph is the answer, written so it could be quoted with nothing above it and still make complete sense. No "it depends," no throat-clearing, no "before we dive in." The answer, stated plainly, in the first 40 to 60 words. Then your examples, your nuance, your voice. The machine takes the top; the human gets the rest. The same answer-capsule discipline underpins good prompting too, which is why the 4-Layer Prompt Framework leans on the same instinct: say the important thing first, qualify it after.
How should you phrase your headings for AI Overviews?#
Phrase every major heading as the exact question a person would type, not an abstract or branded label. AI Overviews trigger mostly on informational queries, and Google matches question-shaped headings to question-shaped queries. A heading like "How much does X cost?" gets extracted far more reliably than a clever one like "The price of progress."
This is where brand voice and machine-readability sometimes pull against each other, and for headings, machine-readability wins. Save the wordplay for the body. The heading is a matching surface. When someone asks Google "how do I structure content for AI Overviews," a page with that literal question as an H2 is a cleaner match than a page with a heading like "winning the answer box." Write headings the way your audience asks, then let your personality live in the paragraphs underneath.
The deeper reason this works is query fan-out. Google's AI Mode and AI Overviews decompose one query into many sub-questions, search each, and synthesize. A page with eight question-headings covering eight angles of a topic can get cited for sub-questions it never literally targeted. Before you write, list the questions a reader actually asks about your topic (People Also Ask and Reddit are good sources) and make the high-value ones into headings.
What kind of evidence makes content more citable?#
Specific statistics, named sources, and verifiable claims make content far more citable, because the AI takes a reputational risk when it quotes you and prefers sources it can stand behind. The Princeton GEO study found adding statistics lifted AI-search visibility up to 41%. Vague, unsupported assertions get passed over.
Think about what the model is doing when it cites you. It's putting your claim into an answer with Google's name on it. A sentence like "structured content performs better" gives it nothing to stand on. A sentence like "pages with question-based headings are cited more often, according to a 2026 AirOps analysis" gives it a verifiable, attributable fact it can quote with confidence. The attribution is the feature, not the decoration.
So write with numbers and names. Replace "many marketers" with the actual percentage. Replace "studies show" with which study. Where you have original data nobody else has, lead with it, because the model preferentially cites the origin of a statistic, and being the source is the strongest position on the page. If you can't source a claim, cut it rather than assert it.
Does schema markup help you get into AI Overviews?#
Schema markup helps by making content easier for Google to parse and by feeding the Knowledge Graph that informs which entities the AI trusts, but it does not guarantee citation. Article, FAQPage, and HowTo schema clarify what your content is and how it's structured. Schema supports citation; it doesn't buy it.
The honest framing matters here, because schema gets oversold. Adding FAQPage markup does not force Google to cite you. What schema does is reduce ambiguity: it tells the system, in a machine-readable way, that this block is a question and that block is its answer, that this page is an article by this author about this topic. That clarity makes extraction more reliable. Think of it as removing friction from the model's reading, not as a switch that turns citation on.
Pair the schema with the on-page structure it describes. FAQPage schema whose answers match real question-headings and real answer paragraphs on the page is coherent and trustworthy. Schema that describes content the page doesn't actually contain is the kind of mismatch that erodes trust. The schema and the visible page should tell the same story.
Why isn't your content getting cited even though it ranks?#
If your content ranks but isn't cited, the causes are usually structural: the answer is buried instead of front-loaded, headings are branded instead of question-shaped, claims are hedged instead of definitive, or passages are too tangled to extract. Ranking proves authority; citation requires extractability. The fix is almost always restructuring, not rewriting.
This is the most useful diagnostic on the page, so be honest with yourself about it. Pull up a page that ranks well but never shows in an AI Overview, and read its first section the way a machine would: is there a complete, quotable answer in the first two sentences, or do you have to read three paragraphs to find the point? Is the H2 a question someone would type, or a phrase only you would write? Are the claims specific and sourced, or soft and general?
Usually the content is good and the packaging is wrong. You don't need new expertise; you need to surface the expertise you already have into liftable passages. Lead each section with its answer, convert headings to questions, harden the hedged sentences into claims with sources, and break tangled paragraphs into self-contained ones. The same article, restructured, becomes citable. For the broader picture of how every engine differs, the guide on AEO vs SEO vs GEO maps the full landscape, the cited-by-AI checklist turns this into a page-by-page audit, and the rest of the AI search guide covers visibility across every answer engine.
Where to start this week#
Take your highest-traffic article that ranks but doesn't get cited, and restructure one section as a test. Lead with a complete answer in the first two sentences, change the heading to the exact question a reader would type, and add one specific sourced statistic. That's a 20-minute edit on a page that already has authority, which is the fastest path to a citation. The biggest wins come from upgrading content that already ranks, not from publishing new pages from scratch. Do one section, watch whether it starts showing up, then run the same three moves across the rest of the page.