Query Fan-Out: Why AI Shows Your Product One Time and Not the Next

A shopper asks ChatGPT for "the best sunscreen for sensitive skin." Before the assistant writes a word, it quietly asks a set of smaller questions on the shopper's behalf. Your product either answers most of them or it answers almost none. That hidden step decides whether you make the recommendation list.

It also explains a question we hear from ecommerce leaders every week: why does one query return us and the next not? The mechanism has a name. Google calls it query fan-out, and it runs underneath Google AI Mode, ChatGPT and Perplexity shopping answers alike.

What query fan-out is

Query fan-out is the step where an AI search system takes a single question, splits it into several sub-questions, runs each against its index, and assembles one answer from the pages that satisfied them. Google introduced the term when it launched AI Mode, describing the system as "breaking down your question into subtopics and issuing a multitude of queries simultaneously".

The questions shoppers bring to these engines are also longer than a classic search. iPullRank's research with Similarweb puts the average prompt submitted to AI search at 70 to 80 words, against 3 to 4 words on Google. A longer prompt carries more constraints, and each constraint becomes a sub-question the engine has to answer.

The practical effect is that the engine stops matching your page to the words the shopper typed. It matches your page to the things the shopper meant.

How one sunscreen question fans out

Take "best sunscreen for sensitive skin" and watch it decompose. A modern engine does not search for that phrase. It generates a fan of sub-questions that might look like this:

• Is mineral or chemical sunscreen better for sensitive skin?
• Which sunscreens are fragrance free and non comedogenic?
• What SPF is recommended for daily facial use?
• Are these formulas reef safe?
• Which sunscreens are safe for rosacea or eczema?
• What is the texture and finish under makeup?
• Are there pediatric or baby safe options?

Each sub-question is answered from whichever pages contain that specific information. Your sunscreen might be the perfect product for sensitive skin. If your product page carries a title, a price, two marketing sentences and a hero image, it answers one of those sub-questions and silently fails the rest. The engine builds its recommendation from the pages that answered most of them. You are not in that set.

This is why a product can be excellent and still be invisible. Fan-out does not reward the best product. It rewards the best described product.

Why you show up for one question and not the next

Two shoppers can ask for the same thing in different words and trigger different fans. "Sunscreen for sensitive skin" might fan out toward ingredients and certifications. "Sunscreen I can wear under makeup every day" fans out toward texture, finish and reapplication. If your page covers the first set and not the second, you appear for one shopper and vanish for the other.

Engines also fan out differently from one another. Google AI Mode, ChatGPT and Perplexity each write their own sub-questions and retrieve from different sources, which is why a brand can be recommended by ChatGPT and absent from AI Mode on the same day. Coverage has to be checked per engine, with the same shopper questions, rather than assumed from one.

Fan-out is also only the first gate. After the sub-questions come back, the engine reads the passages, grades its own draft and may retrieve again. We cover that second pass in why AI search grades your product twice.

The comparison attributes come from the question, not from you

Merchandising teams decide which attributes to feature on a page. The engine decides which attributes to compare on, and it takes them from the shopper's question. Ask for "a coffee maker for a small kitchen that keeps coffee hot without a hot plate," and the comparison runs on footprint, carafe type and brew temperature. A page built around color options and a brand story has nothing to offer that comparison.

The fix is to write the facts shoppers ask about, in plain text on the page, in the words they use. Which facts matter is a question you answer by reading the shopper questions in your category, not by looking at your own attribute template.

What the data shows about fan-out and citations

Classic rank is a weak predictor of whether AI cites you. In Surfer SEO's study of 173,902 URLs, 67.82% of pages cited in AI Overviews did not rank in the top 10 organic results at all, for the main query or for any of its fan-out queries. Among pages that did rank, those ranking for both the main query and its fan-out queries were 161% more likely to be cited than pages ranking for the main query alone.

The same study measured a Spearman correlation of 0.77 between the number of fan-out queries a page ranks for and its odds of being cited. Retrieval happens at the passage level, not the page level. The engine pulls the specific sentence or attribute that answers each sub-question, so a page that buries its reef safe certification in an image or omits it entirely does not exist for that sub-question.

Mentions versus product cards

Another recurring question is why a brand appears as a link in the answer but never as a product card. The two come from different sources. A mention is assembled from page text the engine retrieved and graded. A product card is drawn from structured merchant data, which for Google means the Merchant Center feed and for ChatGPT overlaps heavily with the same Google Shopping data, as we showed in how your Google Shopping feed powers ChatGPT results.

That means a brand can win the fan-out on its product pages and still miss the card because the feed has a disapproved item, a guessed category or a missing attribute. Page content and feed eligibility in Merchant Center are two separate jobs, and both have to be in order to show up both ways.

What changed from traditional SEO

Ranking still matters. It feeds the pool the engine retrieves from. What changed is what "optimized" means:

• The unit of competition moved from the page to the passage or attribute.
• The target moved from one head keyword to the set of sub-questions behind it.
• The winning signal moved from a top organic rank to coverage across those sub-questions.
• The failure mode moved from "page ranks too low" to "page answers too few questions".

Some teams call this work GEO, for generative engine optimization. We call it AI visibility, because the outcome a buyer cares about is simple: when a shopper asks what to buy, is your product in the answer?

How to check your own fan-out coverage

Rank trackers report your position on the head query and miss the sub-questions you failed. You can see the gaps yourself for a handful of products:

1. Pick the products that carry your revenue. Start with a short list rather than the whole catalog.
2. Write the questions real shoppers ask. Use support tickets, reviews and site search, phrased the way a customer would say them to an assistant.
3. List the sub-questions behind each one. Price band, use case, materials, fit, compatibility, certifications, care and who it is for.
4. Check the page against each sub-question. Is the answer present in plain text, or only in an image, a PDF or a support article?
5. Run the questions in ChatGPT and Google AI Mode. Note whether you appear as a card, a mention or not at all, and repeat on a schedule so you can see change.

Frequently asked questions

What is query fan-out?

Query fan-out is when an AI search engine splits one user question into several parallel sub-questions, runs each against its index, and synthesizes a single answer. Google described the mechanism for AI Mode, and the same pattern drives shopping answers in ChatGPT and Perplexity.

Why does ChatGPT recommend us but Google AI Mode does not?

Each engine writes its own sub-questions and retrieves from its own sources. Your pages may cover the sub-questions one engine generates and miss the ones another generates. The only reliable way to know is to run the same shopper questions in each engine and compare.

How many attributes does a product page need?

There is no fixed number. The attributes that matter are the ones shoppers ask about in your category, and those come from the questions, not from a template. Start with the sub-questions behind your top shopper questions and make sure each has a plain text answer on the page.

Is traditional SEO still useful?

Yes, and it is incomplete on its own. Ranking feeds the retrieval pool, yet Surfer SEO found that most pages cited in AI Overviews did not rank in the top 10 for the main query. Coverage across sub-questions, not a single top rank, earns the citation.

Where Paz fits

This is the work Paz does for lean ecommerce teams. A Paz expert writes the shopper questions in your category, finds the sub-questions your pages and feed fail, and fixes the product content from your own sources. You approve each change, and the platform publishes it and tracks whether you show up afterwards, question by question and engine by engine. If you want to know which sub-questions you are losing, talk to us.