How Only Fan Out works, and why your data stays in your browser

Published 1 November 2026. Updated 1 November 2026.

Only Fan Out shows you the searches ChatGPT, Claude and Grok ran by reading the data those sites already send to your browser to display each answer. It has no servers and no accounts, and everything it records stays in Chrome's local storage on your own machine.

The searches themselves are run by OpenAI, Anthropic and xAI, away from your machine, as our guide to how AI assistants search the web explains. This post follows the data from there to your screen: where the queries come from, what reaches your browser, what the extension reads, and what it never touches.

Where do ChatGPT, Claude and Grok get their search queries?

The model writes them. When it decides a question needs current information, it rewrites your prompt into its own search queries, runs them, reads some of the pages that come back and cites a few in the answer.

OpenAI's help article on searching the web with ChatGPT says ChatGPT "typically rewrites your query into one or more targeted queries" and sends them to search providers it partners with. Anthropic's web search tool documentation says "Claude determines when to search based on the prompt" and that the API runs the searches and hands Claude the results, which "can repeat multiple times throughout a single request." Grok's tools overview says its built-in tools, including Web Search and X Search, run on the company's own servers, and that the model keeps calling them "until sufficient information is gathered."

So the assistant writes and runs the queries on its own side, and your browser receives the results through the chat page. Our guide to query fan-out explains why one prompt turns into several searches.

What does each assistant send back to your browser?

Each one sends the answer along with a record of the search behind it, because the chat page needs that record to draw sources and citation links. Some of it the page never shows you.

Engine	Where the search runs	What reaches your browser, as Only Fan Out shows it
ChatGPT	OpenAI's servers and its search partners	Web, business, image and find queries, sources fetched and cited, product cards
Claude	Anthropic's servers	Web and places searches, citations attached to the answer text, places on the map
Grok	xAI's servers	Web searches, page browses, X searches, numbered citations resolved to their sources

ChatGPT is the clearest case. Since 14 September 2026 it sends the text of its search queries only in the live answer stream, then leaves it out of the saved conversation. The ChatGPT app shows the sources and never the queries, so that stream is the one place the query text reaches you. Our post on how to see what ChatGPT searched compares the ways to catch it. To read that stream yourself, our guide to finding ChatGPT's fan-out queries in Chrome DevTools shows which request to open.

What does Only Fan Out actually read?

It reads the responses chatgpt.com, chat.openai.com, claude.ai and grok.com send to your browser while you use them, and nothing else. The privacy policy names two kinds: the streamed response when you send a message, and the conversation data when you open a saved conversation. It reads no other page, no other network request and no data from any other site.

Here is what happens on a normal turn:

1. You send a prompt in a ChatGPT, Claude or Grok tab.
2. The assistant searches on its own servers, reads pages and streams the answer back to the tab.
3. As the answer arrives, Only Fan Out reads that same response and picks out the queries, the pages fetched and the pages cited, plus product cards, places and X posts where the engine sends them.
4. It saves the turn in the extension's local storage and shows it in the side panel next to the chat.

Under the hood, a small script runs inside those four pages from the moment they start loading. It wraps the browser's fetch function, which the chat page uses to talk to its own servers. When a response is one of the chat responses above, the script reads a copy of it and hands the original back to the page untouched; every other response passes straight through. It watches the older XMLHttpRequest route the same way, as a fallback, and a second script passes what it found to the side panel.

On a normal turn the extension sends no requests of its own. It only reads what the tab has already received.

Does Only Fan Out send your data anywhere?

No. The extension has no servers, no analytics, no accounts and no telemetry, and what it captures goes to nobody: not OpenAI, Anthropic or xAI, not the developer, and not any third party.

Captured conversations, your watchlist and your settings live in chrome.storage.local. Chrome's storage API documentation says items in this area "are local to each machine" and that the data is "cleared when the extension is removed." You can also delete one conversation from History, or use "Clear all data" in Settings. If you switch off capture in Settings, the extension records nothing.

The website is a separate thing. onlyfanout.com uses Google Analytics 4 to count page views and Add to Chrome clicks, as the privacy policy says. The extension itself runs no analytics of any kind.

Does Only Fan Out ever make requests of its own?

Only to re-read a conversation, and only to the AI site itself. When you open the side panel on a conversation it has not captured yet, or click Re-read, it fetches that conversation from the engine's own endpoint, the same request the site's own web app makes.

• ChatGPT: it reads your session token from ChatGPT's own session endpoint and uses it for one request to chatgpt.com. The extension holds the token in memory for that one request and never stores, logs or sends it anywhere else.
• Claude: it reads no session token. It finds your workspace from a cookie claude.ai already set, or from claude.ai's own organisations endpoint, then makes one cookie-authenticated request to claude.ai's own conversation endpoint.
• Grok: it reads no session token and no cookie value. It makes two or three requests to grok.com's own conversation endpoints using the session your browser already has.

Every one of these requests goes to the AI site you are signed in to. None goes to a server run by Only Fan Out, because there is no such server.

What permissions does Only Fan Out ask Chrome for?

The privacy policy lists three: a side panel, local storage, and access to the four AI sites.

Permission	Why it is there
sidePanel	The whole interface is a Chrome side panel
storage	Keeps captured conversations, the watchlist and settings on your device after a restart
chatgpt.com, chat.openai.com, claude.ai, grok.com	Lets it read the responses those pages already receive

The policy lists what it does not ask for as well: no tabs or scripting permission, no downloads, no access to other sites and no remote code.

How can you check any fan-out extension before you install it?

Look at what it asks Chrome for and where your data would go. Several extensions now show ChatGPT's searches, and they differ in what they send away from your machine.

1. Read the permissions Chrome shows when you install it, and later under Details on the chrome://extensions page. Access to "all sites" is far more than a fan-out tool needs.
2. Read the Privacy practices section of its Chrome Web Store listing, which lists the kinds of data the developer says it handles. Compare that with the claims in the description.
3. Check whether it wants an account or an API key. Either means some of your data goes to a server, for example to have an AI model write suggestions from your queries.
4. Look in its settings for optional lookups to outside services, such as domain ratings, and switch off any you do not need.
5. Check how it re-reads a conversation: whether it reads your session token, and whether it keeps it.

Only Fan Out's answers are above: four sites, no account or API key, no outside lookups, and a ChatGPT session token held in memory for one request and never stored.

What about Grok's searches of Gmail, Drive and Calendar?

Only Fan Out counts them and never stores what they searched for. Grok can search connected Gmail, Drive or Calendar accounts while it answers. The extension records that one of these searches ran and which tool ran it, and it never reads, stores, exports or watchlist-matches the query text, since that text can reach into private mail and files.

How do you try it?

Add Only Fan Out to Chrome, then use the assistant as you normally do.

1. Install it from the Chrome Web Store.
2. Open ChatGPT, Claude or Grok 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 a message as normal.
4. Read the queries, the pages fetched and the pages cited for that turn in the side panel.

The side panel follows the active tab. After the extension updates, reload the AI tab so it can talk to the new version.

Related questions