Why do AI tools quote some passages and not others?
AI tools quote the passage that answers a sub-question most clearly and most credibly. They retrieve many pages, then pick passages. A passage wins when it states the answer plainly, near the top, in words that match the question, without needing the rest of the page to make sense.
That is why the same page can rank well and still lose the citation. If the answer is in paragraph six, or split across a long story, a clearer source gets quoted instead.
Search patents show the mechanics. Patent US8700620B1 scores documents paragraph by paragraph, and ranks those that cover more of the question's meaningful words, ideally all of them, higher. Patent US12306834B1 compares each retrieved chunk with the query, drops chunks scoring below a cutoff set from the other results (for example, the average minus one standard deviation), and trims further when the text exceeds the model's token budget. A focused passage that answers the whole question survives both filters.
How do you write a direct answer?
Write a direct answer as the first thing under the heading, in 40 to 60 words:
- Start with the subject, repeating the key term from the question.
- Give the answer in the first sentence.
- Add one detail that makes it useful: a number, a condition or the main reason.
- Stop. Explanation, examples and caveats come after.
Here is a direct answer from aeo.so's guide to query fan-out:
Query fan-out is the way AI search splits one question into several related searches, runs them at the same time and combines the results into one answer.
It names the term, defines it, and stands alone.
How should you structure the page?
Structure the page as a set of answered questions:
- H1 as the main question people ask, in their words.
- A direct answer right under the H1.
- H2s for each follow-up question, each with its own direct answer.
- H3s only for parts of an H2's answer.
- One topic per section. A reader, or an AI, should get the point from that section alone.
- A table of contents for long pages, linking each H2.
aeo.so's guides all follow this pattern, and its 261 tactic pages open with a highlighted answer before anything else.
When should you use lists and tables?
Use lists and tables when the content has that shape:
| Content | Best format |
|---|---|
| Steps in order | Numbered list |
| Options or features | Bulleted list |
| A comparison on several points | Table |
| A definition | One sentence, then detail |
| Numbers over time | A chart plus a sentence stating the trend |
AI answers often reuse lists and tables directly. Do not force them, though; a list of fragments with no explanation helps nobody.
How do sources and data help?
Sources and data make a passage worth citing. The GEO research found that adding citations, quotations from relevant sources and statistics raised a source's visibility in AI answers by up to 40%. Practical rules:
- Put a number with a source in each key claim.
- Link to the primary document, not a summary of it.
- Quote named experts where their words add weight.
- Publish your own data where you can; nobody else can cite it first.
What should you avoid?
Avoid what hides the answer or weakens trust: long introductions before the point, vague claims without numbers, headings that tease rather than ask, walls of text with no structure, and the same answer repeated in different words to catch keywords. aeo.so runs every page through a list of banned filler phrases for the same reason: plain, specific writing reads as more credible to people and machines alike.