ChatGPT fan-out queries

When ChatGPT answers a question that needs current information, it does not answer from memory alone. It first turns your prompt into one or more web searches, reads the results, and then writes its answer from the pages it found.

Those searches are called fan-out queries. One prompt "fans out" into the searches ChatGPT actually ran.

Bloomiro now stores them for every ChatGPT check.

Why fan-out queries matter

Your prompt is what a person asks. The fan-out queries are what ChatGPT searched for to answer it. They are often longer and more specific than the prompt itself.

For example, a running shoe store tracks the prompt "best running shoes for beginners". ChatGPT turned it into this search:

best running shoes for beginners 2026 Nike Adidas Puma comparison reviews

That one line tells you a lot:

• ChatGPT looked for a comparison and reviews, not just a list of shoes.
• It added the current year, so fresh pages have an edge.
• It searched for named brands directly: Nike, Adidas and Puma. If your brand or store is not in that search, your pages have to earn their way in through the results it reads.

Fan-out queries show you the exact wording to rank for if you want to be one of the pages ChatGPT reads before it answers. They are observed searches, not search volume, and not a promise that a page ranking for them will be cited.

Where to see them

Prompts tab

Open a prompt and expand a ChatGPT result. Right above the answer you will see Fan-out queries, followed by each search ChatGPT ran for that answer. Hover the label for a short explanation.

The line only appears when ChatGPT searched the web. Some answers come from the model's own knowledge without a search, and those have no fan-out queries.

Actions

Ask in your own words. You do not need any IDs:

• "Tell me the ChatGPT fan-out I have"
• "For the prompt best running shoes for beginners, what fan-out queries were used?"
• "What did ChatGPT search for the prompts in my Running shoes group?"

Actions lists each prompt with its searches under it, grouped by prompt group, and tells you how many prompts had none and why.

MCP

Assistants connected over MCP, such as Claude or ChatGPT, read the same data with get_ai_search_trace:

• No arguments: every tracked prompt that has fan-out queries, newest ChatGPT answer per prompt.
• prompt_text: one prompt, matched by its wording.
• prompt_group_name: one prompt group, matched by its name.
• prompt_id, run_id or result_id when you already have an ID.

Reading fan-out queries is free in the Prompts tab, in Actions, and over MCP.

Good to know

• ChatGPT only, for now. It is the only engine that exposes the searches it ran. Other engines show their answers and cited sources as before.
• Older checks have none. ChatGPT answers checked before this release were stored without fan-out queries. Your next scheduled run, or a rerun, captures them.
• Not the same as Google search ideas. People Also Ask and related searches from Google AI Overview checks stay in Opportunities as Search ideas. Fan-out queries are what ChatGPT itself searched.