How AI Search Decides Who to Cite: Query Fan-Out, Passages and Entities Explained

TL;DR

• Clicks are falling: when an AI Overview appears, the top-ranking page gets 58% fewer clicks, according to a 300,000-keyword Ahrefs study.
• Queries are longer: users describe whole situations, and Google’s AI Mode splits them into subtopics with a technique Google calls query fan-out.
• AI cites passages: systems retrieve the specific section that answers a sub-question, not the whole page.
• Entities matter: brands described consistently across the web are easier for AI systems to recognize and trust.
• Three technical basics decide eligibility: crawler access, structured data that matches the page, and clear authorship.

Contents

1. Ranking is no longer the same as traffic
2. From keywords to conversational queries
3. Query fan-out: one question, many searches
4. Passages, not pages
5. Context and entities
6. Three technical requirements
7. How to measure AI visibility
8. Are keywords dead?
9. Vyara’s take

Many websites now rank first on Google and still lose traffic. The reason is structural: search engines increasingly answer questions directly and cite a handful of sources, instead of sending users down a list of links. Here is how AI search selects those sources.

Ranking is no longer the same as traffic

Two independent studies show how much AI summaries reduce clicks:

Pew also found that 88% of AI summaries cited three or more sources. Being one of those sources is the new goal.

From keywords to conversational queries

Traditional search relied on word matching. A user typed a short phrase, and the engine looked for documents where those words appeared prominently: in the title, the URL and the opening paragraph.

AI search handles a different kind of input. Users describe a situation in full sentences: their company size, current tools, constraints and budget. No single page matches that query word for word.

Query fan-out: one question, many searches

Google explains how AI Mode handles such questions. According to Google’s May 2025 announcement, “AI Mode uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf.”

Each subtopic may be answered by a different source. The final response combines them. As a result, a page can be cited for a sub-question it was never explicitly optimized for, and a page that ranks first for the main phrase may not be cited at all.

Modern retrieval also compares meaning rather than exact wording. Both the query and the content are represented as vectors, and relevance is measured by semantic similarity. Repeating a key phrase adds no extra relevance.

Passages, not pages

AI systems retrieve content at the level of passages. A long article is split into sections, and the system selects the section that best answers a specific sub-question.

This has a direct consequence for writing:

• each section should open with a direct answer;
• the first two or three sentences under a heading should make sense without the rest of the page;
• phrases such as “as mentioned above” weaken a passage, because it no longer stands on its own.

Context and entities

Context. Language models interpret a term through the words around it. A term mentioned without a definition, example or supporting data carries little meaning. A passage that surrounds it with specific numbers, names and related concepts signals expertise on the topic.

Entities. AI systems draw on knowledge graphs that represent brands, people and organizations as entities. A brand described consistently on its website, professional profiles, industry directories, interviews and press coverage is easier to identify with confidence. Inconsistent information, such as an old company name that was never connected to the new one, makes a source less reliable. Consistent brand mentions across independent sites increasingly complement traditional links.

Three technical requirements

Authorship matters even more on voice assistants and wearables, where users hear a single answer rather than a list of options.

How to measure AI visibility

Monthly search volume shows how many people search. It doesn’t show which brands AI systems recommend. A practical starting point:

1. List 20 to 200 real questions customers ask before buying.
2. Run them in several AI assistants, in clean sessions without history.
3. Record whether your brand is mentioned, whether it is cited with a link, and which competitors appear instead.
4. Repeat monthly, because AI answers change frequently.

Traffic should also be read alongside revenue. Fewer visits combined with stable sales often means visitors arrive better informed.

Are keywords dead?

No. Keyword density is. Search systems are hybrid: they combine semantic retrieval with classic lexical matching. If customers use a specific term for a product or service, content that avoids that term may not be matched to their questions, even when it describes the same thing.

Vyara’s take

“We’re no longer competing for a position. We’re competing to be cited. Speak your customer’s language, even when their term isn’t the most precise one, but write the knowledge only you have. SEO and GEO help you get found. Knowledge is what gets you cited. One doesn’t work without the other.”

Further reading: From keywords to questions: why ranking #1 on Google no longer brings you clients – Vyara Stefcheva’s full analysis on stefcheva.com.

Sources: Ahrefs via Business Wire – AI Overviews cost websites 58% of clicks · Pew Research Center – Google users are less likely to click on links when an AI summary appears · Google – AI in Search: Going beyond information to intelligence · Google Search Central – Succeeding in AI search

About the Author

Vyara Stefcheva

Vyara Stefcheva is the founder and editor of Egorithms and an independent marketing expert. She follows how technology, AI and algorithms change the way people find, choose and buy, and writes about what that means for businesses and for the people who use them.

She has spent her career building marketing and business strategies for more than 300 brands across three continents. She built and exited a marketing agency and now works independently. She has taught more than 20,000 people through lectures, courses and corporate training, and speaks at conferences about AI and the future of the internet.

On her YouTube channel she tests new AI and marketing tools as they launch and gives each an independent verdict: adopt, watch or ignore.

See author's posts