Getting cited in AI Overviews.
Getting cited in AI Overviews means a section of your page is mentioned as a source inside the summary, and this follows a selection logic different from classic ranking.
The weakening link between ranking and citation invalidates most teams’ measurement habits. A page can rank third and never be cited; another can rank ninth and be mentioned every time. What separates the two is less the page’s authority than its structure and its phrasing.
This page lists the things that make that difference. None of them is magic, and all of them already count as good writing — but in the answer engine era their cost and their payoff have both grown.
The cited unit: the passage
This is the most fundamental concept. When a model composes a summary, it does not evaluate the page as a whole and decide “let me cite this page”; it finds the SECTION that contains the answer and takes that. The unit of optimization is therefore the section, not the page.
The practical consequence: a long, comprehensive page struggles to be cited if its sections are not self-contained. Conversely, a mid-length page where every section fully answers a single question is mentioned far more often.
What defines the section boundary is the heading structure. The model derives where things end from the document structure; in a sea of text without headings there is no boundary, and the model cannot tell where to cut. That is why the heading hierarchy is as much a citability item as an accessibility one.
The first-sentence rule
The first sentence of a section determines that section’s chance of being cited out of all proportion. The reason is simple: the model looks for a short statement that is correct on its own, and the first sentence is the closest candidate for that role.
A good first sentence does three things. It answers the question without repeating it. It needs no outside context. And it cannot be misread when quoted alone. When all three hold, the sentence becomes citable.
- Bad: “There are a few important points on this topic.” — says nothing.
- Bad: “As we explained above, this method works.” — a back-reference, meaningless without context.
- Bad: “So how is this done?” — a question, not an answer.
- Good: “The contrast ratio is the ratio of two colors’ relative luminances and takes a value between 1:1 and 21:1.” — a definition, complete, independent.
- Good: “WCAG 2.2 was published on 5 October 2023 and added nine new success criteria.” — a fact, a date, a number.
What the last two examples share is that they are useful to a human as well. Writing for citability does not conflict with writing for readability — they are two names for the same discipline.
Structured data: it eases the match
FAQ and how-to schemas translate a page’s question-and-answer pairs into machine language. This makes it easier for a model to find which question the page answers — it does not have to infer it from the text; it reads it explicitly.
There is one critical rule, and it is often violated: a question in the schema must also be visible on screen. A question-answer block that exists only in the schema is both a violation of search engine rules and a half-measure for the user — the model cites it, the visitor cannot find it on the page.
The FAQ sections on this site’s service pages are built exactly that way: the questions live both in the schema and on screen, and the verification suite measures on every run that the two match.
Multimodal content
Industry reports show that pages carrying images and video alongside text, supported by structured data, get selected at a visibly higher rate than text-only pages. The size of the gap varies by source, but the direction is consistent.
The reason is probably twofold: the content is richer, and pages that carry visuals tend to be the pages more effort went into. So the relationship may not be causal — but the practical advice stays the same.
One important warning: adding an image is not enough by itself. If the information the image carries is not in its alternative text or the text around it, the model cannot read that image and the page behaves like a text-only page. The machine counterpart of an image is its alternative text.
Freshness and identity
Answer engines check how fresh a piece of information is before using it — especially in fast-moving fields. Having the publication and update dates both on screen and in machine-readable form pays off directly on this scale.
Author identity carries weight the same way. Who wrote the page, what that person works on, and what the claims rest on — the three together determine the content’s credibility. Declaring them in the schema and showing them on screen gives the same information to both readers at once.
And the most skipped item: citing sources. When what a claim rests on is written down, the claim weighs more for human and model alike. A measured number, a dated event and a named source — all three are what make a page citable.
Writing citably is the technical name for writing honestly.
What competitor analysis teaches
Looking at who is already cited on your target queries is more instructive than any general advice — because it shows directly what the model prefers for that query. The method is manual and simple: ask the question, open the mentioned sources, and find the cited passage.
What you examine is not the whole page but the cited passage itself. How many words? Right under a heading? Does it answer the question directly? Does it carry a number, a date or a definition? The answers to these four questions tell you which form works for that query.
- Measure the cited passage’s length — usually one or two sentences; long paragraphs are rarely taken whole.
- See its relation to the heading — the passage is almost always right beneath one.
- Note its concreteness — passages carrying numbers, dates, definitions or step lists are chosen disproportionately.
- Look at the source type — official documentation, industry publication or community post? It varies by query type.
Done once a month, this sweep produces a pattern, and that pattern shapes the form of your own content. Real behavior in your field replaces general advice.
Being cited may not bring clicks
To be honest: being mentioned inside the summary does not always bring visitors. The user may have the answer and never click the link. This is the most debated side of answer engine optimization.
Still, being mentioned has value, and it comes in two layers. The first is measurable: industry reports show cited pages earning a higher click-through rate than uncited rivals. The second is hard to measure but real: being named as an authoritative source pays back in later searches and direct visits.
That is why tying informational content’s success to clicks alone is now an incomplete measurement. Brand mentions, citation frequency and direct traffic — looked at together, the three make the picture more honest.