Query Fan-Out Simulator & Coverage Checker

Google describes AI Mode and AI Overviews as using a query fan-out technique: breaking your question into subtopics and running many related searches at once. The passages an AI answer draws on can be retrieved for any of those sub-queries, so a page that only answers the main query leaves gaps. This free Google Colab notebook simulates the fan-out for a seed query and checks, passage by passage, how well your page covers each sub-query.

An honest note before you use it: Google doesn’t publish its sub-queries, so nothing here shows you the real fan-out. The notebook generates plausible synthetic queries as a structured brainstorm and a content-gap check.

Who it’s for

Content strategists, SEOs and writers who are optimising pages for AI Mode and AI Overviews and want a practical way to find the questions a page leaves unanswered.

How to use it

1. Open the notebook in Colab and run it once on the sample pages.
2. Enter your seed query and the URL of your page (add its supporting pages too, if you have them).
3. Pick a fan-out method: the free rule-based expansion, an LLM (OpenAI, Anthropic or Gemini, with your own key and a documented prompt you can edit), or both.
4. Run the passage step: the page text is extracted and split into heading-aware passages (chunking).
5. Run the coverage step: embeddings shortlist the closest passages by cosine similarity, then a small reranker judges whether each passage actually answers the sub-query.
6. Review the gap list and export the results.

What you get

• A list of simulated sub-queries for your seed query, tagged by facet (comparison, implicit need, follow-up and so on)
• A coverage table: every sub-query labelled covered, partial or gap, with the best-matching page, heading and passage
• A gap list with a suggested action for each one, gaps first
• A per-page view showing which page owns which sub-queries, useful when you check a page with its supporting content
• CSV files and an Excel file with Coverage, Gaps and Passages tabs to import into Google Sheets

Limits and costs

The sample run and the rule-based mode are free and need no key. LLM fan-out is one short call per seed query to the provider you choose, typically a fraction of a cent with the small default models. The scores measure relevance rather than quality, so a covered passage can still be vague or out of date: read the matches. The thresholds were set on the sample pages and the models are English. If you want the background on passage ranking, it explains why passage-level coverage matters.

Read how entities power query fan-out in AI search for the theory, and build the query research skills behind it in the Semantic AI-powered SEO Keyword Research course.