ChatGPT Search Query Extractor

When OpenAI quietly added a live-web search to ChatGPT, many marketers wrote it off as "just another AI feature". Fast-forward to 2025, ChatGPT Search is now one of the fastest growing referral channels on the internet.

If your brand is already popping up in answers, you've probably noticed a welcome bump in traffic. If you're not showing up yet, keep reading - this is the playbook.

A breakdown of ChatGPT search... 👇

Every time you send a message to ChatGPT, the 1️⃣ first thing it does is decide whether or not to use the web search tool.

To search or not to search?

It makes this decision using the following rules. These are the exact instructions (extracted June 2025) from ChatGPT's system prompt set by OpenAI:

## web

Use the web tool to access up-to-date information from the web or when responding to the user requires information about their location. Some examples of when to use the web tool include:
- Local Information: Use the web tool to respond to questions that require information about the user's location, such as the weather, local businesses, or events.
- Freshness: If up-to-date information on a topic could potentially change or enhance the answer, call the web tool any time you would otherwise refuse to answer a question because your knowledge might be out of date.
- Niche Information: If the answer would benefit from detailed information not widely known or understood (which might be found on the internet), such as details about a small neighborhood, a less well-known company, or arcane regulations, use web sources directly rather than relying on the distilled knowledge from pretraining.
- Accuracy: If the cost of a small mistake or outdated information is high (e.g., using an outdated version of a software library or not knowing the date of the next game for a sports team), then use the web tool.

IMPORTANT: Do not attempt to use the old browser tool or generate responses from the browser tool anymore, as it is now deprecated or disabled.
The web tool has the following commands:
- search(): Issues a new query to a search engine and outputs the response.
- open_url(url: str) Opens the given URL and displays it.

You'll notice here there are four reasons which would cause it to use web search; local context, very fresh data, niche knowledge, or extra verification for accuracy.

Example:

• "What's the capital of France?" is well known and will never trigger search.
• "What's the weather like in Newquay UK today?" is unknown and will always trigger web search.

If chatGPT says "no" to search, it will skip the next steps and the model will immediately generate a response from it's training data. You can learn how LLMs work here.

Normalizing the prompt

Next, 2️⃣ ChatGPT breaks down your large detailed prompt into multiple small search queries.

1. Splits your big prompt into several focused queries.
2. Rephrases each using the keywords and entities it thinks matter most.

For example my question to ChatGPT, "What's a good first surfboard for a beginner in 3-4 ft waves?", was transformed into:

• "best surfboard for beginner in three to four foot waves"
• "soft-top surfboard beginner recommendations"

Real-time data with Bing

Once the original prompt is destrucutred, ChatGPT sends each of these search queries to the web() tool AI agent, which first, 3️⃣ searches Microsoft Bing.

Yes! Bing is the search engine used by ChatGPT 🤯.

Bing returns a full set of search results each with the following metadata:

• Full URL - Descriptive slugs are really important!
• Title - Page <title> tag
• Description - Usually based on the meta description
• Ranking - Position in the SERP
• Published Date - Structured data

Next, from this list of search results, 4️⃣ the web tool agent will select which pages it thinks are most relevant to the original prompt.

And then, will 5️⃣ crawl the full page content.... if not blocked:

• blocked by robots.txt
• non 2xx status code
• request times out

Important! Our investigations show that ChatGPT-user bot does not render JavaScript, so ensure your pages are serving pre-rendered HTML.

Once crawled, 6️⃣ the agent is able generate a summary of the content or find the best 200 character snippet for each page.

This web search tool is actually a black box. The only people who know how it works is OpenAI themselves. But, what we DO KNOW is that queries are input and sources are output.

We built a specialist Chrome extension for SEOs to reveal the underlying ChatGPT search queries and sources from the web search tool.

Generating your response

The output sources are then handed back to the ChatGPT model that you are personally chatting to, the one that has context of memories and conversation history.

The 7️⃣ final output response is then generated, which contains brands mentions, external hyperlinks to inline citations as well as recommendations, preferences & biases completely embedded throughout the content.

Note: ChatGPT is unable to recall any page content it has just crawled outside of the returned snippet in follow-up messages. You should be periodically checking your AI visibility, and optimizing your content for AI search .

Where the Search Queries Are Hiding?

By design, ChatGPT doesn't show you the exact queries it sent to Bing, it only shows the sources. But they're not truly hidden-just buried in your browser's network logs.

DevTools Hack

1. Right-click → "Inspect" to open DevTools.
2. Go to the "Network" tab and filter by "/conversation" requests.
3. Look for the payload that includes "search_query" - there are your Bing queries in plain text.

An Easier Way?

We built a free Chrome extension that extracts those hidden ChatGPT queries with a single click of a button. Install ChatGPT Search Query Extractor

James Berry
Founder & CEO at LLMrefs
llmrefs.com