> ## Documentation Index
> Fetch the complete documentation index at: https://docs.fixaeo.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Query fanouts
> The sub-questions an AI engine answers behind your prompt.
A fanout is a sub-question an AI engine answered on its way to answering yours.
Ask an engine "best CRM for a 10-person SaaS startup" and it doesn't answer one question. It quietly answers four or five. Which CRMs target startups. What features matter at ten seats. What the price ceiling is. How they fit a dev stack. Each of those pulls its own evidence, and the final answer is stitched together from all of them. Query Fanouts shows you that breakdown, one prompt and one engine at a time.
## Why do fanouts matter?
[Visibility Score](/metrics/visibility-score) tells you a prompt didn't name you. Fanouts tell you what the engine was actually thinking about instead.
Say you track "best CRM for startups" and never appear. The fanouts show the engine spent that answer on pricing tiers, migration effort, and free-plan limits. If your site talks about pipeline automation and nothing else, you're not losing the race. You're in a different one. The fix is a page that answers the sub-question, not a louder version of what you already published.
That makes each sub-question a small content brief. It's the closest thing in the product to the engine telling you what it wanted to read.
## How does FixAEO find the sub-questions?
Engines don't publish their internal steps, so FixAEO reads the finished answer back.
A second model receives three things: your prompt, the answer text, and the list of URLs that answer cited. It's asked to name the sub-questions the answer covered. The result is stored against the answer it came from, and refreshed whenever that answer changes.
This is a reconstruction, not a transcript. It's one model's best reading of another model's answer. Treat a fanout as a strong hint about what the engine covered, not as proof of what it did internally.
Two kinds of answer are skipped. Anything very short is too thin to break apart, and so is an answer an engine never returned — a timeout or a quota error has no sub-questions in it.
## How fresh is the data
Fanouts are prepared in the background, so opening the page never waits on a model call.
New scans are picked up on their own, and an answer that hasn't changed isn't re-read. A large workspace catches up over a few passes rather than all at once, so give a busy account a little time after a scan lands.
## What's on the page
Three counters sit at the top. **Prompts** is how many of your tracked prompts have fanouts in the window, and matches the number of cards below. **Engines** is how many distinct engines decomposed at least one of them. **Sub-questions** is the total across every prompt and engine pair.
Below that, one card per prompt, newest scan first, ten to a page. A collapsed card shows the prompt text and the brand it belongs to. Beside those sit the engine count, the sub-question count, the date of the latest scan, and a small rail of engine logos.
Open a card and, if two or more engines decomposed that prompt, you get a tab per engine tagged with its own sub-question count. A single-engine card skips the tabs and just names the engine. The list underneath is numbered in the order the reading model produced it, so number 1 is what it judged the answer led with. Switching tabs is the useful move here. Two engines rarely split the same prompt the same way, and the gap between them is often the story.
## Filters and scope
The range tabs cover 7d, 30d, 90d and 1y, and default to 30 days. The window filters on scan date.
The engine filter is a multi-select, and only appears once two or more engines have data in the window. Ticked engines are shown combined, not one at a time. Untick everything and the page honestly shows nothing.
The cards are scoped to the brand you have selected in the app. See [Managing brands](/managing-brands) for how the selection works.
A very wide range on a busy account can return more prompts than one response holds. If that happens a banner tells you to narrow the range, and what you can see is always whole prompts rather than half a card.
## Why is the page empty?
The empty state names both likely causes instead of picking one. It reads "either there are no tracked prompts yet, or they haven't been scanned since you added them."
What actually populates the page is a chain of three things:
A real buyer question, not just your brand name. Brand-name scans give an engine nothing to decompose. See [Setting up your prompts](/setting-up-prompts).
The prompt must have been answered, and the answer must have enough substance to break apart.
Usually minutes after the scan is analyzed, longer on a big workspace.
There's a second, narrower message. Once some brand in your workspace has fanouts but the selected one doesn't, you get "no fanouts for this brand in this window" and a nudge to widen the range or switch brand. If you've set an engine filter, a **Show all engines** button appears beside that message.
## What to do with a fanout
Pick a prompt where you're not named. Open it, switch to the engine you care about most, and read the numbered list top to bottom. Any sub-question you have no page for is your next brief.
Read sub-questions as topics to cover. For link-level evidence — which URLs an engine actually cited and which pages it read — use [Citations](/citations) and [Sources](/sources).
Group those briefs before you write. [Topics and labels](/topics-and-labels) shows whether a gap is one prompt or a whole theme, and [Improve](/improve) turns the theme into specific work.
The same prompts split by engine, scored rather than decomposed.
How an answer becomes a mention, a citation, and a score.
The pages an engine read while drafting, including ones it never showed.
How many prompts and engines your plan scans, which sets how much lands here.