---
type: "Article"
title: "AI Search Query Capture | search attribution and measurement"
description: "AI Search Query Capture: a Chrome extension that reveals, live and on-device, the search queries and cited sources ChatGPT and Claude use."
resource: "https://www.laurelinlabs.com/lab/ai-search-query-capture"
tags: ["ai-search-query-capture", "search-attribution-and-measurement", "query-fan-out-visibility"]
generated: { by: "laurelin-site-build/2026-08-25", at: "2026-08-25T12:00:00+00:00" }
status: "stable"
stale_after: "2027-02-25T00:00:00+00:00"
sources:
  - id: "stat-1"
    resource: "https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/"
    title: "TechCrunch"
  - id: "stat-2"
    resource: "https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/"
    title: "TechCrunch"
  - id: "stat-3"
    resource: "https://blog.google/products/search/ai-overview-expansion-may-2025-update/"
    title: "Google"
---
# AI Search Query Capture

A free Chrome extension that reveals the web-search queries ChatGPT and Claude run when they browse, and the source URLs they cite, captured live and on-device. It makes query fan-out visible, the bridge between a natural-language prompt and what actually gets retrieved and cited. Live on the Chrome Web Store.

## What does an AI assistant actually search for when it browses?

When ChatGPT or Claude runs a live web search it silently rewrites your prompt into its own search-engine queries. Those queries, and the sources the assistant then cites, are the bridge between what a user asks and what actually gets retrieved, but they are invisible without sitting in developer tools watching network traffic. This surfaces them live in a popup and an optional on-page ticker, de-duplicated and exportable to CSV.

## Why does seeing the fan-out queries matter?

Capturing the queries an assistant issues to the open web, plus the URLs it cites, reveals how prompts are decomposed and reformulated (query fan-out), and which phrasings and entities the model anchors on. That is first-party signal for content and entity optimisation that complements traditional keyword data.

The build itself, reverse-engineering how ChatGPT and Claude stream their responses, is as much the point as the data. Everything stays on the device; no data leaves the machine.

## Key figures

| Metric | Figure | Source |
|---|---|---|
| ChatGPT weekly active users (Oct 2025) | 800 million | TechCrunch[^stat-1] |
| Messages sent to ChatGPT per day | 2.5 billion | TechCrunch[^stat-2] |
| Monthly users of Google AI Overviews | 2 billion | Google[^stat-3] |



Hub: [Measurement & Attribution](/topics/measurement-attribution.md)

[^stat-1]: TechCrunch
[^stat-2]: TechCrunch
[^stat-3]: Google