Google published its own explanation of how AI Mode picks sources, and most of what gets written about it online still describes it wrong. It does not search for your exact keyword and rank pages against each other the way classic Google search does. It breaks one question into several smaller ones, searches each separately, and builds an answer out of whichever pages had the best paragraph for each piece.
That distinction changes what a B2B page needs to do. Here is the plain version of how it works, and what to actually change on a page because of it.
Query fan-out, explained plainly
When someone types a question into Google's AI Mode, the system does not run one search behind the scenes. It generates a set of related, smaller questions, sometimes 5, sometimes 20 or more depending on how broad the original question is, and runs each one as its own search. Google's own AI optimization guide describes this directly: the systems may issue multiple related searches across subtopics before generating a response.
Take a question like "how do I choose a fractional CMO." Google's system might fan that out into separate searches for fractional CMO cost, fractional CMO versus a full-time hire, warning signs of a bad fractional CMO, and how long a typical engagement runs. Each of those runs as its own search, pulling from whichever pages answer that specific piece best. The final answer stitches the best pieces together into one response.
AI Mode fans out further than AI Overviews, and it keeps going
AI Overviews, the summary box that sits inside a normal Google results page above the ten blue links, and AI Mode, Google's separate full-screen conversational search experience, both use fan-out, but not to the same degree. AI Overviews typically runs one round of fan-out per search and shows it alongside regular organic results. AI Mode issues more searches per question. Reporting on Google's underlying system describes up to 16 sub-queries fanned out from a single question, run at the same time, and AI Mode does not show organic results at all, only the synthesized answer and its citations.
The part that matters more for a B2B site: AI Mode is conversational. A user can ask a follow-up question in the same session, and that follow-up triggers a fresh round of fan-out scoped to whatever was just asked. Ask "how do I choose a fractional CMO," then follow up with "what about for a manufacturing company specifically," and the second question fans out into its own separate set of searches, pulling from different pages than the first answer did. A page that never gets pulled into the first answer can still get pulled into the third or fourth follow-up in the same conversation. This is why a page built around one broad topic and a narrower page built around one specific sub-case (manufacturing, professional services, a specific price range) can both earn a citation in the same conversation, just at different turns.
A page does not need to match the exact search
This is the part most SEO advice has not caught up to. A page gets selected by answering one of the sub-questions the system generated, not by matching the exact wording of what the user typed. Search Engine Land's guide to optimizing for query fan-out makes this explicit: the related queries are chosen for relevance and for covering different angles of the topic, not for being synonyms of the main search.
What that means in practice: a page about fractional CMO pricing does not need to also be titled or optimized around "how do I choose a fractional CMO." It needs to be the best available answer to the pricing sub-question specifically. That is a lower bar to clear than trying to own the entire broad topic, and it is why a focused page can now outperform a page trying to cover everything.
Passages get cited, not pages
AI Mode does not lift a whole page into its answer. It pulls a specific passage: a paragraph or a short section that directly answers one sub-question. This means a page with one genuinely strong section and five mediocre ones can still get cited, for that one section. It also means a great headline and a weak body underneath will not get selected, because the system is reading the actual content of each section, not judging the page from the top.
The practical result is that every H2 section on a page is competing on its own, not riding on the strength of the page as a whole. A page built as one long argument that only makes sense start to finish loses to a page built as a series of sections that each stand alone and answer one clear question completely.
What to actually change on the page
Three changes matter more than anything else here. First, open every section with a direct answer to the question the heading asks, then explain the reasoning underneath. A reader, or a system extracting one paragraph out of context, should get the answer in the first sentence or two of the section, not buried at the end after three paragraphs of setup.
Second, pick fewer sub-questions and answer them completely rather than gesturing at every possible angle of a topic. A page trying to cover ten related questions shallowly loses to five separate, focused pages that each answer one question in full. This is the same logic behind splitting a broad pillar page into specific sub-pages, and it applies just as much to blog posts.
Third, name real things. Real tools, real numbers, real named frameworks. A passage extraction system has an easier time lifting a section that says "HubSpot Marketing Hub Pro" and "$4,000 to $10,000 per month" than one that says "a marketing platform" and "a competitive price." Specificity is what makes a passage worth quoting on its own.
This does not replace normal SEO, it adds a layer on top
Google's own guide is direct about this: the same fundamentals still apply. Helpful, accurate content. Clear site structure. Fast-loading pages. Real expertise behind what is written. Query fan-out changes how content already meeting that bar gets selected and cited. It is not a replacement for doing the underlying work well, and a thin page will not get cited just because its structure happens to match a fan-out sub-question.
The takeaway
Stop writing pages aimed at one exact search phrase. Google's own system is not searching that way anymore. It is asking several smaller questions on the reader's behalf and picking the best answer to each one, wherever it finds it.
Pick the handful of sub-questions your page can answer better than anyone else. Answer each one completely, in a section that stands on its own. That is what gets picked up, not a page that tries to be everything to everyone.
Related guides: GEO vs SEO for B2B SMBs and AI Overviews vs traditional SERP.