How to see what ChatGPT searched before it answered

Published 30 October 2026. Updated 30 October 2026.

ChatGPT lists the pages it cited and keeps the search queries it ran out of view. To see the queries, catch them in the live answer stream as the reply arrives, either with a Chrome extension such as Only Fan Out or with your browser's developer tools open before you send the prompt.

The queries decide which pages ChatGPT reads at all, as our guide to how ChatGPT searches the web explains. This post covers each way to see them, what it shows and where it falls short, as of October 2026.

What does ChatGPT show you about its own searches?

ChatGPT shows citations in the answer and a Sources button, and nothing more. OpenAI's help article on searching the web with ChatGPT says to "Select Sources, when available, to view cited sources and other relevant links." The same article says ChatGPT "typically rewrites your query into one or more targeted queries" and may send more specific follow-up queries after reading the first results. It describes no way to see those queries in the app.

Why can't you find the queries in older ChatGPT conversations?

Since 14 September 2026, ChatGPT sends the text of its search queries only in the live answer stream, then strips it from the saved conversation. The sources it fetched and cited are still saved.

How-to guides written earlier in 2026 tell you to reload a finished conversation with Chrome DevTools open and search its data for "queries". On a conversation from after the change, that search finds sources and no query text. Field names have also come and gone between ChatGPT model versions this year, according to Little Green Agency's notes on ChatGPT's hidden fan-out queries, so a method that depends on one field can stop working without warning.

In practice, you can only see the queries for a prompt you send while something is reading the stream. A conversation from before you set that up shows its sources and nothing else.

Which ways of seeing ChatGPT's searches still work?

Two of these read the live stream as the answer arrives, and only those two show the queries ChatGPT ran in your own chat.

Method	What it shows	The catch
Sources button in ChatGPT	Cited sources and other relevant links	No queries at all
Only Fan Out side panel	Queries in the order ChatGPT ran them, plus fetched and cited sources, per turn	Chrome only, and it records prompts sent after you install it
Chrome DevTools, Network tab	The raw answer stream, query text included	Must be open before you send; raw JSON whose field names change
OpenAI API web search tool	Queries and sources for a request you send with code	A separate request from your chat, so its searches can differ
Search Console query filters	Long queries that may come from AI assistants	Search Console never says where a query came from

How do you see ChatGPT's searches with Only Fan Out?

Install the extension, open its side panel on a ChatGPT tab and send your prompt.

1. Add Only Fan Out to Chrome from the Chrome Web Store.
2. Open chatgpt.com and click the Only Fan Out icon in the toolbar to open the side panel. There is no popup and no options page.
3. Send your prompt as you normally would.
4. Read the turn in the side panel: the queries in the order ChatGPT ran them, then the pages it fetched and the pages it cited.

ChatGPT writes four kinds of query (web, business, image and find), and one turn often mixes them. Shopping prompts also bring back product cards with price, rating and merchant, which the panel keeps apart from the cited sources. To keep a record, export the turn or the whole conversation as CSV, Markdown or JSON. If queries stop showing, Diagnostics gives the parser status for each engine and flags when ChatGPT changes a field. After the extension updates, reload the ChatGPT tab.

How do you find ChatGPT's search queries in Chrome DevTools?

You can read the raw stream yourself, as long as DevTools is open before you send the prompt.

1. Open a ChatGPT conversation, press Ctrl+Shift+I (Cmd+Option+I on a Mac) and pick the Network tab.
2. Send a prompt that needs current information, such as a price or a recent event.
3. Find the request that started when you pressed send and kept loading until the answer finished.
4. Open it, go to its Response or EventStream view, and search for "queries".

Expect raw JSON split across many small events, with field names that change from model to model. It works for one prompt and gets slow across twenty.

Does the OpenAI API show the same searches as ChatGPT?

The API shows the searches behind requests you send through it, and those are separate from your ChatGPT chats. OpenAI's web search guide for developers says each search step comes back as a web_search_call item recording the action taken, and that a search action "usually (but not always)" includes the queries it ran. The full list of pages sits in a sources field, which the guide's examples request explicitly.

That makes the API handy for running many prompts with a script. The searches still belong to your API request and model settings, so they can differ from what ChatGPT ran in a chat.

Can Google Search Console show what ChatGPT searched?

Only as a guess. Ann Smarty's write-up in Practical Ecommerce suggests filtering Search Console queries with a regular expression for queries of ten words or more, since AI assistants tend to write longer queries than people type. Search Console does not record where a query came from, and OpenAI does not name the search providers ChatGPT uses, so a long query could come from anyone. Treat these as leads to check against queries you have actually seen.

What should you do with ChatGPT's search queries?

Check whether your pages rank for them. Each query is a search ChatGPT ran while building its answer, and a page that ranks for none of them never gets opened, however well it answers the question the person typed.

Compare the queries with the keywords you already track. A query that names a brand, a year or a site shows what ChatGPT went looking for. Then compare the fetched pages with the cited ones: our guide to query fan-out explains why the fetched list is nearly always the longer of the two.

Related questions