What is Query Fan Out?
Query fan-out is Google's own term. Google says AI Mode and AI Overviews can issue several related searches at once, across subtopics and data sources, and stitch what comes back into one answer. So the question a person types isn't the question that decides who gets cited. The sub-queries behind it are, and those never appear in a keyword tool or in Search Console.
That gap is measurable. Ahrefs looked at 863,000 keywords and 4 million AI Overview URLs and found only 38% of cited pages also ranked in Google's top 10 for the same query, down from 76% in its July 2025 run. You can hold position one and still be missing from the answer, because your page never addresses the sub-question the model actually went looking for.
How this tool helps
There are two layers here and they are deliberately not the same thing. The first is measured: the tool sends your query through OpenAI's Responses API with the web_search tool forced on, then reads the exact search strings ChatGPT issued straight out of the API response, along with the answer it wrote and the pages it cited. Nothing is inferred there.
The second layer is a simulation. A model decomposes your query into 8 to 12 sub-queries the way AI search engines tend to, and tags each with a type, an intent and a citability estimate. Read layer one as evidence and layer two as a hypothesis worth testing. For any sub-query from either layer you can pull the live Google SERP, the People Also Ask block, and the AI Overview when Google shows one.
Explore more tools
FAQ
How does this tool work?+
Two passes. First it calls OpenAI's Responses API with web_search forced on and extracts the real queries ChatGPT sent, its answer, and the pages it cited. Then a second model decomposes your query into 8–12 simulated sub-queries, grounded with the real result so it doesn't fall back on stale training data. For each sub-query you can load live Google SERP results, People Also Ask questions and the Google AI Overview, when one is shown, via DataForSEO.
What is the ChatGPT Simulation card?+
It is the fallback. When the Responses API call works, the top card is labelled Real ChatGPT Answer & Citations and holds exactly that: the answer the model wrote and the pages it actually cited. The ChatGPT Simulation card only appears when that call fails or comes back empty, and then it is an approximation of how a chatbot would answer, not a live response. Check the label before you quote the domains in it.
When does an AI Overview show up?+
Google only triggers AI Overviews for a subset of queries — mostly informational, research-style questions. Navigational, transactional, or very specific sub-queries rarely get one. When an AI Overview exists for a sub-query, the tool fetches it live via DataForSEO and shows the generated text plus the cited sources. US English queries trigger AI Overviews most often; coverage in other markets varies.
What is citability?+
Citability (0–100) estimates how likely an AI search engine is to cite external sources for a given sub-query. Factual queries with clear answers (e.g. product specifications) have high citability. Subjective or common knowledge queries have lower citability.
How accurate are the sub-queries?+
Depends which list you are reading. The queries under Real ChatGPT web queries are captured, not simulated: they come out of the web_search calls in the Responses API payload. The Simulated Fan-Out list is a model guessing how an AI search engine would decompose the query, so treat it as a plausible shape rather than a transcript. Nobody outside Google or OpenAI can read Gemini or AI Mode internal sub-queries, so any tool claiming to show them is guessing too. The Google SERP data, People Also Ask questions and AI Overviews are pulled live via DataForSEO.
What's the daily limit?+
5 free analyses per day. Each analysis uses 1 AI credit plus SERP credits for the sub-query lookups, which run on a separate and much larger bucket so one run never eats the next. For unlimited access, enter your own OpenAI and DataForSEO API keys in Settings.
How many web searches does ChatGPT actually run?+
However many it ran for your query, which is the number this tool reads back. OpenAI publishes no average, and the third-party figures that circulate shift with every model release, so a fixed number printed on a tool page would be stale within weeks. Run the same query twice and the count can differ. That variability is the finding, not a bug.
Can I analyze queries in different languages?+
Yes. Select your target country and language. The AI generates sub-queries and the AI Answer Simulation in the appropriate language, and the SERP results reflect local Google rankings for that market.
AI visibility in your browser
Lumina analyzes AI crawler access, content citability and more — automatically on every page.
Add Lumina to Chrome — Free