ChatGPT Query Fan-Out Bookmarklet v2 (GPT-5.6)

AI Summary

Since GPT-5.6, ChatGPT no longer saves its fan-out queries in the conversation, so the first version of Quentin Yacoub's bookmarklet only returned citations and search results. Version 2 adds a small recorder that captures the queries live from ChatGPT's answer stream, stores them locally in the browser, and exports them with the citations and web search results, still without an OpenAI API key.

Quentin Yacoub
Quentin Yacoub
8 min
Table of Contents
Blueprint-style graphic featuring the text CHATGPT QUERY FAN OUT BOOKMARKLET V2 (GPT-5.6) on a grid background.

Cet article est aussi disponible en français : Script d'extraction de query fan-out ChatGPT v2 (GPT-5.6)

Last week, I gave a demo to an Adobe intern studying at Harvard who wanted to talk about GEO (Generative Engine Optimization) with a specialist. He had built a Google Docs extension to optimize content for search and generative engines, and I wanted to show him a simple way to extract ChatGPT’s query fan-out with the free bookmarklet I shared in April. That’s when I realized it no longer worked: the citations and the web search results were still there, but the most interesting part, the fan-out queries, was empty.

The same week, I took part in a panel on AI in SEO and spent time with people from the SEO community in and around Montreal. I received great feedback on this site and on my posts, and honestly, that was the push I needed to get back to work after a summer break spent enjoying the good weather.

So I opened DevTools to understand what had changed. My first guess was that OpenAI had changed the format of the conversation data the bookmarklet reads. It turned out to be more interesting than that: OpenAI stopped saving the fan-out queries in the conversation. Here is version 2. Same idea, same export page, but a new way to catch the queries.

If you want to do this at scale, feel free to contact me on LinkedIn to discuss how Adobe Brand Visibility (formerly LLM Optimizer) could help you.

What changed with GPT-5.6

The first version never read the page itself. It loaded the conversation JSON (/backend-api/conversation/{id}), the same file the ChatGPT web app uses to display a chat, and looked for the search queries inside it.

I opened Chrome DevTools and compared two things: what ChatGPT receives while it writes the answer, and what it saves after. Here is what I found with GPT-5.6:

WhereFan-out queries
Live answer stream, while ChatGPT is answering✅ Present (search_model_queries)
Saved conversation, the file v1 reads❌ Removed: the web search step is still there, but its queries are stripped
ChatGPT interface, after the answer❌ Only “Searched 8 sites”

You can even see the query flash on screen for a second (Searching best SEO agencies Montreal…) while the answer is being generated, then it disappears.

Citations and web search results are still saved, which is why v1 kept showing them. The queries, however, now only exist for a few seconds, in the live stream.

How the Query Fan-Out Bookmarklet v2 Works

Since the queries only travel in the live stream, the bookmarklet now has to be there before ChatGPT answers. It does two jobs:

  1. Recorder: the first click turns on a small recorder in your ChatGPT tab. It reads the answer stream as ChatGPT receives it and keeps every fan-out query it sees. A blue badge in the top-right corner shows how many queries it has caught for the current conversation.
  2. Export: the next click (or a click on the badge) opens the same report as v1 in a new tab: fan-out queries, citations, and web search results.

The queries are stored in your own browser (localStorage on chatgpt.com), per conversation, so you can still export them after reloading the page.

What It Retrieves

  • Fan-out search queries: the actual strings ChatGPT typed into a search engine, captured live.
  • Detailed citations: every URL cited in the response.
  • Web search result snippets: the page titles and excerpts saved for each search hit.

One important limit: the recorder only catches queries from answers generated while it is on. It cannot recover the queries of an older conversation.

Why It’s Still Important to See the Fan-out

If you work in SEO/GEO, the reasons have not changed since v1:

  • Organic Ranking Visibility: check if your site appears in the results ChatGPT looked at.
  • Source Tracking: see which sites the AI relies on to form its opinion of your industry.
  • Content Gaps: if the AI searches for “Pricing for [Your Brand]” and finds nothing, you know what content to create.

And the fact that OpenAI now hides these queries after the answer makes a free way to see them even more useful.

Already Using v1? Update It in 30 Seconds

If you already added my previous bookmarklet, you don’t need to create a new one. Just replace its code:

  1. Copy the new code: Use the copy button on the code block below.
  2. Edit your bookmark: In your bookmarks bar, right-click your existing ChatGPT Fan-out bookmark and choose Edit….
  3. Replace the URL: Click in the URL field, select everything with Ctrl + A (Windows) or ⌘ + A (macOS), then paste the new code. It must still start with javascript:.
  4. Save: Click Save. You can keep the same name.
  5. Reload your ChatGPT tabs: Tabs that were already open still run the old version until you reload them.

That’s it: from now on, click the bookmark once before asking your question, as explained below.

How to Install and Use the Bookmarklet on Chrome

This works for Google Chrome on desktop. Already using v1? See the update steps above.

  1. Show your bookmarks bar: Use Ctrl + Shift + B (Windows) or ⌘ + Shift + B (macOS).
  2. Create a new bookmarklet: Right-click an empty area of the bar and choose Add page….
  3. Name it: Call it something like ChatGPT Fan-out.
  4. Paste the script: In the URL field, paste the entire code block below. It must start with javascript:. Chrome bookmarklet dialog: name “ChatGPT Fan Out”, URL field with the javascript: bookmarklet
  5. Turn the recorder on, before asking: Open ChatGPT (you must be logged in) and click the bookmark. A blue “Fan-out recorder” badge appears in the top-right corner.
  6. Ask your question: Use a prompt that triggers a web search. The badge counts the fan-out queries while ChatGPT answers. ChatGPT answer with the blue Fan-out recorder badge showing 1 query in the top-right corner
  7. Export: Click the bookmark again, or click the badge. A new tab opens with the queries captured live, the citations, and the web search results. Export tab showing the fan-out query captured live by the recorder, followed by the citations

Tip: the badge is your safety check. No badge means no recording. After a page reload, click the bookmark once before asking your next question.

The Bookmarklet Code

I know we probably haven’t met yet, and I’ll be the first to admit that pasting a block of code into your browser can feel a little sketchy. It definitely did for me the first time.

This version does one more thing than v1, and I want to be upfront about it: to catch the queries, it wraps the fetch function of the ChatGPT page so it can read the answer stream as it arrives. It never modifies what ChatGPT receives. The queries are saved in your browser’s localStorage on chatgpt.com, and nothing is sent to an external server.

If you want a second opinion, paste the code into your favorite LLM and ask: Does this code contain anything malicious, does it send data anywhere, and can the author access my ChatGPT conversations? It’s a great way to verify everything for yourself.

I also follow a strict privacy-first approach and do not use tracking cookies, which is why you won’t see a cookie banner on my site. You can read my privacy policy for full details.

Limitations and Potential Errors

  1. No fan-out query was recorded for this conversation. (yellow note in the export) The answer was generated before the recorder was on, or the page was reloaded in between.
  • Solution: keep the tab open, check that the badge is visible, ask your question again, then export.
  1. Nothing to export from this conversation yet... The conversation has no web search, no citations, and no recorded query yet.
  • Solution: ask a question that forces ChatGPT to look for external data, for example: What are the best SEO agencies in Montreal?
  1. Your browser blocked the new tab. Chrome blocked the export tab.
  • Solution: allow pop-ups for chatgpt.com, then click the bookmark again.
  1. The badge stays at 0 after an answer that clearly searched the web. OpenAI may have changed the format of the stream again.

As with v1, my time for troubleshooting is a bit limited with a baby at home, but I’ll do my best to keep this tool working as ChatGPT evolves.

Please note: All opinions and blog posts shared on this website are strictly my own personal projects. They do not represent the views, strategies, or official products of my employer, Adobe.

Quentin Yacoub
Quentin Yacoub
SEO & GEO Specialist

Hi, I'm Quentin, an SEO/GEO Specialist at Adobe building AI powered tools for Site Optimizer and Brand Visibility. I use this site to document my thoughts on where search is heading in the age of AI, and to share the strategies and shortcuts I rely on.

Connect on LinkedIn

Related Articles

What is AEM Sites Optimizer?

Go behind the scenes of AEM Sites Optimizer to see how we are building an autonomous site optimization system at Adobe.

6 min read

Hello World: My First Website

Welcome to my personal blog. Discover why I built this sandbox to document my experiments in SEO, GEO, and AI-driven search. Here's what to expect.

9 min read