---
title: "How AI Overviews Choose the Sources They Quote"
seoTitle: "How AI Overviews Choose the Sources They Quote"
description: "Google's AI Overviews pull answers from a handful of pages. How those sources get chosen, and how a small site earns a place among them next to big brands."
datePublished: "2026-07-09T16:06:00Z"
dateModified: "2026-07-09T16:06:00Z"
category: ai
imageAlt: "Iron Goo blog featured image on how an AI overview picks the handful of sources it cites and how a small site earns a slot."
tags: [ai-overviews, ai-search, aeo, citations, smb-ai]
faq: true
---
Count the links. Ask Google something a buyer would ask, watch the AI summary form at the top of the page, and then count the sources it actually cites underneath. Three. Sometimes four. Rarely more than a handful. Below that little box sit ten ranked results, and behind those ten sit a hundred more pages that all rank for the question. Almost none of them get cited. The ai overview sources, that short list of links the summary leans on, are not the hundred ranked pages restyled, and they are not even the top ten. They are a separate, much smaller pick, and the gap between "ranks for this" and "gets cited for this" is where most owners quietly lose.
I started counting after a client asked why a page that sat at position four for a buying question never appeared under the summary, while a smaller competitor's page did. So I read the cited list, source by source, the way you would read a guest list and ask why these few names and not those. The cited pages were not the highest-ranked. They had something else in common, and once you see it, the whole thing stops being a black box.
## How do AI Overviews choose which sources to quote?
An AI overview cites a small set of pages that each state the exact point the summary needs in a form it can lift, that agree with the other cited pages on the facts, and that come from a source the engine already has reason to trust. That cited set is a separate, smaller selection than the ranking list.
That is the short answer. The rest of this is what each of those three things means in practice, and why a small specific site can land on that list when a bigger one that ranks above it does not.
## The cited list is not the ranking list
Here is the assumption that costs people: the sources under an AI summary are just the top organic results with a new coat of paint, so the way in is the same old climb up the rankings. Rank higher, get cited. Owners chase position the way they always have, often against national brands they have no realistic way to outrank, and they treat a citation slot as the prize at the top of that same ladder.
It is a different ladder. The ranking list answers one question: which pages are most relevant and authoritative for this query, in order. The cited list answers a narrower one: which few pages can I quote to build this specific summary, where every sentence needs a source that actually states it. A page can be excellent for the first job and useless for the second. It can rank at position three because it is a thorough, well-linked page on the topic, and still get skipped by the summary because nowhere on it does it say the one precise thing the summary needed in a sentence the engine could lift clean.
So you can rank and not be cited. Owners experience that as a contradiction, because they have collapsed two separate judgments into one. The engine has not. It ranks pages for relevance and authority, and then, when it builds an answer, it runs a second, stricter selection over a smaller field to decide which sources the answer is actually made of. Treat those as one race and the citation list will keep looking random. It is not random. It is selected on different grounds.
::::comparison{title="Two different lists"}
:::side{label="Ranks but not cited"}
A strong, broad page that covers the whole topic and sits high in the ten blue links. Authoritative, relevant, well-linked. The summary respects it but cannot lift a single sentence from it that states the exact point cleanly, so it does not make the cited set.
:::
:::side{label="Cited under the overview"}
A page that names the precise point the summary needs, in one liftable sentence, agrees with the other sources, and comes from somewhere the engine already trusts. It may rank lower. It still earns a slot, because the answer can be built from it.
:::
::::
## What the cited sources share
Read enough of these cited lists and the same three traits keep showing up. They are the real selection criteria, and none of them is "ranked first".
**They state the exact point in a liftable form.** The summary is assembling sentences, and every sentence wants a source that says that thing plainly. A page that answers the question directly, in a clean declarative line near the top, hands the engine a passage it can quote with confidence. A page that buries the same fact three scrolls down inside a wall of qualifiers, or only implies it, makes the engine do work it would rather hand to a page that just said it. This is why a [liftable, readable answer the engine can quote without rewriting](/blog/quotable-answer) beats a denser, smarter page that never states the point in one piece. The summary is not grading your essay. It is looking for a sentence it can borrow.
**They corroborate the other cited sources.** An engine assembling a factual answer does not want to stake the whole thing on one page. It wants agreement. When several independent sources state the same fact the same way, that fact is safe to put in a summary, and the pages that stated it become the natural citations for it. A page that agrees with the consensus on the specifics is a candidate. A lone page making a claim nothing else backs is a risk the engine will route around, however well it ranks. Being one of several sources that say the same true thing is worth more here than being the single cleverest page that says something no one else confirms.
**They come from a source the engine already trusts.** The summary is the engine putting its own name on an answer, so it pulls from places it already has reason to believe. That trust is built the slow way, off the page: a business that resolves cleanly as one real, identifiable entity, described consistently everywhere it appears, mentioned by other credible sources. None of that is a trick you add at the end. It is the groundwork that makes your page eligible to be quoted at all, the difference between a page the engine is willing to cite and one it merely tolerates in the rankings.
Notice that none of the three is rank. A page can clear all three and sit at position six. Another can rank second and miss all three. The cited list is selected on these traits, not on position, which is exactly why position alone never explained who got cited.
:::callout{type="key" title="The real test"}
A page earns a citation when the engine can lift one clean sentence from it, that sentence agrees with what other trusted sources say, and the page comes from a business the engine already treats as real and credible. Rank helps you get read. These three decide whether you get quoted.
:::
## Why a small site can beat a big one for a slot
This is the part owners do not expect, and it is the reason any of this is worth your attention. A small, specific site can land on the cited list ahead of a national brand that ranks above it. Not often by luck. By design.
Big brands win rankings because they have the authority, the link profile, and the breadth to cover a whole topic. But breadth is the weakness here. A large page that covers "small business accounting" across every angle rarely states any single narrow point in the clean, liftable form the summary wants, because it is busy being thorough about everything. It is broad where the summary needs precise.
A small site that does one thing can be precise. Take a bookkeeping service that works only with independent cafes. When a buyer asks an assistant how a cafe should handle tips on the books, the giant accounting brand has a page that mentions tips somewhere inside a sprawling guide. The cafe-bookkeeping site has a page whose whole job is that exact question, answered in a plain sentence at the top, consistent with what other credible sources say about tip reporting. The engine assembling that answer has an easy choice. It does not need the biggest page. It needs the page that states the needed point, agrees with the others, and comes from somewhere it trusts. Specificity plus corroboration beats raw size for that slot.
That is the realistic path in, and it is not "rank higher". It is: pick the precise questions where you can be the clearest, most quotable, most corroborated source, and own those. Be the page the answer is built from, not the hundredth page that ranks.
The same selection logic runs underneath the whole answer, not just the headline. An assistant rarely answers a buying question with one search. It [splits the question into a fan of narrower sub-questions and looks for a source to cite for each strand](/blog/query-fan-out), which means there is not one citation slot to win but several, one per sub-question. A small site has no chance of being the best source for the broad headline. It has a very real chance of being the obvious, liftable, corroborated source for one specific strand of it. That is where slots actually open.
:::callout{type="tip" title="Where to aim"}
Do not fight for the citation on the broad question a national brand owns. Find the narrow sub-question inside it where you can state the answer most plainly and back it with consensus. Win the strand, not the headline.
:::
## This is AI search, not one company's quirk
It would be easy to file all of this under "a Google thing" and move on. It is not. AI Overviews is the named feature you can watch most easily, because Google shows the cited links right there under the summary. But the behavior underneath it is how AI search and AI answers work generally. The assistant inside a search product, a standalone answer tool, a chat assistant pulling live sources: each writes an answer by gathering sources, preferring ones that state the point cleanly, that agree with each other, and that come from places it trusts.
The surfaces differ. AI Overviews, Gemini, Claude, and other assistants each show their sources differently, or sometimes barely at all. The selection logic rhymes across them, because they all face the same problem: assemble a trustworthy answer from the open web without staking it on a single shaky page. Solve to be cited under one well, by being precise, corroborated, and trusted, and you are solving for the others at the same time. Optimize for "Google's box" alone and you will miss how much of this is just how machines build answers now.
## What this changes about your own pages
The shift is small to say and large to act on. Stop treating a citation as the prize for ranking first, and start treating it as a separate thing you earn on its own terms. Look at the questions your buyers actually ask an assistant. For each one, ask whether your page states the answer in a sentence the engine could lift, whether that answer agrees with what other credible sources say, and whether your business reads as real and trustworthy everywhere it appears. Where the answer is no, that is the work, and it is different work from chasing position.
One page done this way is a proof of concept. The real return comes from being the clear, corroborated source across the whole cluster of questions in your niche, so the engine keeps finding you whichever strand of the answer it is building. That is a site-level strategy, not a one-page fix, and it is exactly what [the deep AEO guide for building cited presence across the sources an answer is assembled from](/guides/seo/seo-for-ai-search-and-aeo) is for. When you would rather have someone build that presence into the source set for you than learn the whole discipline first, that is what the [AI search optimization service](/services/aio) does.
Go read that guide next, and bring one question your buyers ask. That single precise question, answered in a form the summary can lift and corroborate, is where your first citation comes from.
