Analyze results
Query fan-out
The sub-queries an AI engine runs behind a prompt before it writes the answer. Where AmICited gets them, how they are classified, and how to turn them into content.
Definition
Fan-out query
Why fan-out matters#
When someone asks ChatGPT “best helpdesk software for ecommerce”, the engine does not search that exact phrase once. It fans out into several searches (“helpdesk software Shopify integration”, “Zendesk vs Gorgias pricing”, “helpdesk for small online store”), reads the results, and writes an answer from the pages it found.
That means the pages that get cited are the ones that rank for, or closely answer, the sub-queries, not only the prompt. If you optimize a page only for the prompt wording, you can miss the questions that actually decide which sources are used.
Where the queries come from#
Every prompt has one fan-out set, merged from two inputs:
- Engine-reported. ChatGPT and Gemini report the sub-queries they actually ran. These are collected with each answer, never inferred, and cost nothing extra.
- Predicted. For the other engines (Perplexity, Claude, Grok and the rest), which do not report fan-out, a model predicts the sub-queries from the answers and sources collected for the prompt.
The two lists are de-duplicated, with the engine’s own wording winning where both produced the same query. The set is rebuilt automatically after scheduled runs, and you can Generate or Regenerate it on demand from the prompt detail page.
Categories#
Each sub-query is classified by how it transforms the original prompt, into one of 19 categories:
| Category | Example transformation |
|---|---|
| Reformulation, Generalization, Specification | Same question in other words, broader, or narrower |
| Decomposition, Implicit | Splits the question, or asks what it assumes |
| Comparison, Entity, Attribute | “X vs Y”, a named product, one property of it |
| Use case, Compatibility, Pricing | “for small teams”, “works with Shopify”, “cost” |
| Procedural, Definitional | “how to…”, “what is…” |
| Freshness, Geographic | “2026”, “in Germany” |
| Sentiment, Authority | Reviews and opinions, expert or official sources |
| Related, Multimodal, Other | Adjacent topics, images or video, anything else |
Each sub-query also carries a search Intent (informational, commercial, transactional and so on) with a confidence, plus search volume, CPC and competition where keyword data exists.
Reading the Fan-out queries table#
On the prompt detail page, Fan-out queries lists “the sub-queries generative engines derive from this prompt”. Filter by category or intent, or add sub-queries as prompts of their own:
- Track adds one sub-query as a tracked prompt.
- Track all prompts (N) adds every sub-query shown.
When a citation analysis is available, the table grows a heatmap: each cited domain becomes a column, ordered by its average citation position (best first), and each cell shows how close that domain’s best page sits to the sub-query. A ★ row at the top does the same for the main prompt. Rows are sorted so the sub-queries most related to the cited pages come first: those are the ones your page most needs to cover. Darker cells mean closer.
Read the heatmap column by column
Find the column for the domain that is cited first. The rows where it is dark and your column (marked you) is pale are the sub-questions the winner answers and you do not. That is your outline for the next edit.
Fan-out history#
Fan-out history shows “which sub-queries the engines started and stopped running” over the selected date range, grouped as New, Stable and Dropped, with how many days each was seen (for example “12/30 days”) and a sub-queries-per-day chart. It reads only what engines actually reported, so it covers ChatGPT and Gemini.
Use it to spot a change in how an engine interprets the question. If a “pricing” sub-query appears and stays, pricing content just became relevant to this prompt even if your visibility has not moved yet.
How to act on fan-out#
- Open a prompt you are not ranked on and read its fan-out list. Group the sub-queries by category.
- Check prompt coverage to see which sub-queries your pages already answer (Covered, Partial, Uncovered).
- Add sections, not keywords. A sub-query like “Zendesk vs Gorgias pricing” wants a short, direct comparison with numbers, not the phrase repeated.
- Track the sub-queries with volume. High-volume sub-queries are often prompts worth monitoring on their own.
- Generate an article for a cluster of uncovered questions. See AI articles.
Fan-out also shows up in organic search. The AI searches and Site operators reports under Reports read Search Console queries that contain a site: operator, which is how an assistant’s fan-out reaches Google. See search and traffic reports.