ChatGPT Internal Query Extractor
Browser extension that extracts ChatGPT's internal fan-out search queries from any active conversation. Reuses the user's existing ChatGPT session, walks the conversation message tree to isolate the most recent user prompt and its assistant reply, and exports the fan-out queries to JSON, Markdown, or clipboard.
The ChatGPT Internal Query Extractor is a browser extension that pulls ChatGPT’s internal fan-out search queries from any active conversation. Single-conversation investigation tool that sits in the same family as the IG-Prediction and Fan-Out Coverage Study research pipelines, which do the same extraction at scale.
The gap it closes
ChatGPT does not show users the internal fan-out queries it generates while answering a prompt. Those queries are present in the conversation metadata but not surfaced in the UI. For anyone investigating how ChatGPT retrieves and chooses content, the fan-out queries are the most useful signal in the response, and the rendered UI throws them away.
Architecture
A toolbar-activated extension with a single popup UI that runs against the active ChatGPT tab. The extension reuses the user’s existing session to read conversation metadata. No external API key required; the extension uses the same calls the ChatGPT web app already makes for the active user.
Conversation tree traversal
The conversation metadata contains a tree of messages with parent / child references between them. The extractor finds the most recent user message, then finds the assistant reply whose parent is that user message and whose fan-out metadata is populated, and pulls the queries out. This handles regenerated answers and branched conversations correctly without picking up queries from earlier turns.
Output paths
Three export formats supported. Copy-to-clipboard (concatenated text), JSON download (structured { originalPrompt, queries }), and Markdown download. A theme toggle switches the popup between light and dark.
What shipped
Working browser extension that surfaces ChatGPT’s internal fan-out queries for any conversation the user has open, in three export formats. The single-conversation prototype behind the production-scale fan-out research pipelines.