Query Fan-Out: See the Searches ChatGPT Runs Before It Answers, and the Few Pages It Cites

A buyer types a question into ChatGPT. A few seconds later there's a neat answer with two or three sources linked under it. What they don't see is the work in between. ChatGPT turned their question into its own search, read a few dozen pages, and picked a tiny handful to cite.

This query fan-out tool shows you that hidden step. Type a question your buyers ask, and we run it through ChatGPT search. You get back the searches it ran, every page it read and the few it cited. Add your own site, and it tells you whether you were cited, read but passed over, or never found at all.

It's free, there's no signup, and it answers the question most teams can't, which is why ChatGPT picked those pages and not yours.

What query fan-out means

Query fan-out is the name for what an AI engine does with your question before it answers. It doesn't search for your exact words. It rewrites the question into one or more searches of its own, often longer and more specific than what you typed. Then it reads the pages those searches return and decides which ones to quote.

That process has three stages, and you can lose at any of them.

Most advice skips straight to the last stage and tells you to "write for AI". The tool shows you which stage you're actually losing at, and that changes what you should fix.

What the tool shows you

Every result has three parts, plus a fourth if you add your site.

The second and third lists are where the lessons are. Every page in "Read but not cited" made it through the search and still lost. Comparing them with the cited pages, side by side, shows what the winners did differently. Often it's something plain, like a price in the first paragraph or a date that shows the page is current.

The answer text itself isn't shown, only the sources behind it. That's on purpose. Answers change from run to run, but the searches and the sources tell you what to build.

What we saw when we ran 10 buyer questions

We ran 10 everyday buyer questions through the tool in October 2026, from the best CRM for a small business to how much an emergency plumber costs. We left the site field empty, so the runs show the full picture for each question.

Across the 10 questions, ChatGPT read 304 pages and cited 20 of them. That's about one in fifteen. Being read is nowhere near enough.

The cited pages mostly weren't Google's top results

Here's the part that should change how you think about this. We compared the sites ChatGPT cited with the sites ranking in Google's top 10 for the same everyday searches. Of the 18 sites it cited, only 3 were on Google's first page for that search.

A strong Google ranking helps a page get found. It clearly doesn't decide which page gets quoted. The other 15 cited sites got picked for what their pages said, since none of them held a first-page spot for that search.

The searches it ran looked nothing like the questions

ChatGPT ran one search per question in these runs, but it rewrote every one of them. "How much does a new roof cost?" became a search for a current-year average cost, with the roofing material and named cost sites in it. "What's the best CRM for a small business?" became a search for the current year's official pricing, with five vendor names added.

A few patterns showed up again and again across the 10 searches.

• The current year appeared in eight of the ten searches. ChatGPT went looking for fresh pages, and dated content had an edge.
• "Official pricing" or "pricing" showed up in five. It wanted the vendor's own price page, not a blog's guess.
• The software searches added a list of brand names. If you're not one of those names, you're not in that search.
• The cost questions asked for averages, typical fees or surveys. It wanted a number with a source, not a sales page.

Turn the searches into pages

The searches in each result are a content plan you didn't have to guess. Each pattern we saw points to a page you can build.

None of these are clever tricks. They're the pages a careful buyer wants anyway, which is exactly why the AI engine goes looking for them.

Read but not cited: how to move up a list

If your site shows up in "Read but not cited", you're close. ChatGPT found your page and decided another one answered better. That's usually fixable in an afternoon.

Look at the cited pages for the same question and compare their openings with yours. Check whether they answer in the first two sentences, give a real number or range, and show a recent date. If your page makes the reader dig, put the answer up top.

Then add the details the rewritten search asked for. If the search said "official pricing", put your current prices on the page in plain text. If it said the current year, update the page and say so. If it named rivals, add a fair comparison that names them too.

Not found: how to get into the reading list

If your site never shows up, the problem is earlier. ChatGPT's own search didn't return your page, so it never had the chance to read it.

The fix starts with the searches in the result. Each one is a search you can target. A page called "Payroll software pricing in 2026" with real numbers will match a search like the ones above far better than a generic product page.

Two other things to check before you write anything new.

1. Can AI crawlers reach your site at all? Our AI Crawler Checker tests that in about 20 seconds. A firewall block makes every other fix pointless.
2. Is your text in the page's HTML? Our What AI Sees tool shows whether AI crawlers can read your page or just an empty shell.

Five questions worth running first

You'll learn the most from the questions closest to a purchase. Start with these.

1. Your category question, like "best payroll software for a small business".
2. Your price question, like "how much does payroll software cost".
3. Your name against your main rival, like "you vs them".
4. The problem your product solves, phrased the way a buyer would say it.
5. Your brand name with "pricing" or "review", to see what sources describe you.

Add your site to each run. If you're not found on the first two, that's where your next pages should go.

Why one run isn't the answer

AI answers aren't fixed. Ask the same question twice and you can get different searches, different pages read and different sources cited. One run tells you what can happen. It doesn't tell you what usually happens.

That's why a single check is a starting point. Run the tool on your five most important questions now, fix what it shows, and come back in a month. Results are cached for 24 hours per question, so a run on the same day will show you the same result.

The real problem isn't one answer. It's knowing, week after week, which of your buyers' questions name you and which name someone else, and noticing when that changes.

Where RankControl fits

RankControl does that tracking for you. It checks 50 of your buyer questions every week across six AI engines, ChatGPT and Google AI Mode included. For each one, you see whether you were cited, which competitors were named instead, and which sources the answer quoted. A new citation sends you a notification, and lost ones show in your dashboard.

It also builds the pages that win those searches. It finds the questions buyers in your market search, then writes and publishes up to 100 articles a month as native posts on your own site, each scored for how easily AI engines can quote it. Traffic from Google and AI search shows up first.

Citations come more slowly, mostly as other sites link to and talk about you. Link Control helps there, finding sites to pitch for each article and sending the emails from your own inbox once you approve them.

You can do this by hand. Run your questions through each engine every week, note the sources in a spreadsheet, rewrite the pages that lose, publish new ones for the searches you're missing and pitch sites for links. That's 20 to 30 hours a month for a small team. Or RankControl can run it for $400 a month, with a 7-day free trial.