You do not rank in an AI Overview, you get cited in one, and the difference decides how you optimise. Google assembles an Overview by pulling passages from several pages and writing a short answer around them, then linking the sources it used. The pages it pulls from are frequently not the ones sitting at position one, which is the single most useful thing to understand here: a page well outside the top three can be cited if it answers the specific sub-question cleanly, and a page at number one can be skipped if its answer is buried. So the work splits in two. Be eligible, which means Google can crawl and parse you. Then be quotable, which means one passage on your page answers the question so directly that lifting it into an Overview requires no editing.
Answer the sub-question, not the topic
An AI Overview rarely answers one clean query. Google breaks the question into smaller ones, gathers a source for each, and stitches the result together. That is why broad, comprehensive guides underperform here while narrow pages punch above their weight. A 4,000 word guide to a topic may contain your answer somewhere in the middle, competing with everything else on the page for attention. A focused page that names the sub-question in an H2 and answers it in the next sentence is far easier to pull. When you plan content for Overviews, write down the actual sub-questions a buyer asks in sequence, and make sure a passage exists that answers each one on its own.
Write passages that survive being lifted
The test for every important passage is simple: cut it out of the page, paste it somewhere with no surrounding context, and see whether it still makes sense. If it opens with “this” or “it” or “as mentioned above”, it fails, because a model pulling that passage into an answer inherits the confusion. Name the subject inside the sentence. State the claim first, then the evidence. Attach the number and its source in the same breath rather than three paragraphs later. Roughly 150 words of direct answer near the top of a section, written so it needs no neighbours, is the shape that gets quoted.
Give the passage something checkable
Overviews lean toward sources that carry verifiable specifics: a figure, a date, a named standard, a named tool. Vague competence reads the same as everyone else's vague competence, and there is no reason to cite one over another. Concrete detail also protects you, because a claim with a source attached is one a quality rater can confirm. Name the real things in your field, the platforms, the standards, the tools, the places, rather than writing around them in generic language. Recognisable named entities are part of how a model decides your page is genuinely about the subject and not a thin imitation of it.
Make it parseable, not just present
Structure carries real weight because it tells a machine where the answer starts and stops. A question-shaped H2 followed immediately by its answer, short paragraphs, ordered steps for anything sequential, and FAQ markup where a page genuinely answers a set of questions, all reduce the work required to extract you. Server-render the content so it exists in the HTML rather than appearing after JavaScript runs. And check the page in Search Console: a page Google has not indexed cannot be cited, so indexation is the floor, not an afterthought.
Track citations separately from rankings
This is where most teams go wrong in measurement. Overview citations move independently of position, so a ranking report will not tell you whether the work is landing, and a page can gain citations while its position stays flat. Pick the ten questions that matter most commercially, run them in Google monthly, and record whether an Overview appeared and who it cited. That list is your real competitive set for AI search, and it usually differs from the one your rank tracker shows. Watch it change over a quarter rather than a week, because citation sets settle slowly.
The short version
Be reachable, answer the narrow sub-question, write passages that stand alone, attach checkable specifics, and structure the page so extraction is easy. None of that requires you to outrank anyone first, which is exactly why it is worth doing while your competitors are still waiting for their rankings to improve. The same discipline goes by the name answer engine optimization, and it applies well beyond Google, to every surface that resolves a question instead of listing links. If you want to know which questions in your category already trigger an Overview and who is being cited instead of you, that is the first thing our free AI visibility report shows.