Ever wonder what ChatGPT searches for when it finds sources? Learn how to identify its fan-out queries, the exact search terms it sends to Google or Bing, and how that insight can shape your SEO and Answer Engine Optimization strategy.
If you've ever asked ChatGPT a question and noticed that it shows you a list of sources, that means it went out to search the web. It probably used Google or Bing, found a few pages that matched your question, and then brought that information back to give you an answer.
So what's actually going on here? ChatGPT takes your prompt, turns it into a search term, and sends that query to a search engine before writing its response.
Now, if you work in marketing, SEO, or AEO, this is something worth paying attention to. Because if you can understand which queries ChatGPT is sending to Google or Bing, you can start optimizing your content around those exact search terms. That's how you improve your chances of being cited or showing up in ChatGPT answers.
Let's go through it step by step.
Start by opening ChatGPT and typing something that will make it search the web. In this example, the query was: "I'm a healthcare executive looking for a solution that can help me better understand my brand's visibility in AI engines."
When ChatGPT gives you an answer that includes a list of sources, that means it's using the web search mode. You'll often see it reference a few websites, and that's your clue that it's pulling data from Google or Bing.
Not every prompt sends ChatGPT to the web. Research shows that only around 31% of all ChatGPT prompts trigger at least one background search. Knowing what triggers that behavior is useful, because it tells you which types of queries you most need to optimize for.
The most common trigger words are "reviews," "2026," "free," "features," and "comparison." Local intent prompts trigger a web search 59% of the time, making them the highest trigger category. Commercial intent prompts follow at 53.5%, compared to just 18.7% for purely informational queries.
Industries like software and jobs and careers tend to average close to three fan-out searches per prompt, more than almost any other category.
If you are creating content, this data tells you something useful: writing content structured around comparisons, reviews, and features, and keeping your page dates current, gives you a better chance of ending up in the pool of sources ChatGPT draws from.
Once you know it's using the web search mode, the next step is to see exactly what it's sending out.
Here's how to do that:
You'll then be able to see the exact queries ChatGPT sent to Google or Bing.
This method still works. I've tested it recently and it holds up. One thing I have noticed since first writing this article, though, is that the queries themselves have gotten noticeably longer and more specific over time. The fan-out queries ChatGPT generates have evolved from short, broad terms to longer and more targeted phrases. This has been confirmed by other teams working in this space as well. Peec AI, for example, published a study analyzing over 20 million fan-out queries and observed the same trend in their data.
Inside the JSON file, look for something that says "search_model_queries." Under that, you'll see the actual search terms that ChatGPT used.
In the example, ChatGPT searched for: "brand visibility in AI engines SaaS" and "AI search visibility analytics platform"
It used those search terms to pull information from the web, and then built the answer visible in the chat window.
One of the more striking findings to come out of recent research is that ChatGPT doesn't just fan out into multiple queries. It often switches to English to do it, even when the original prompt was written in a completely different language.
Tomek Rudzki, GEO expert at Peec AI and founder of ZipTie.dev, published research analyzing over 20 million fan-out queries across more than 10 million user prompts. The findings were notable: across all non-English prompts in their dataset, 43% of the background searches ran in English. In nearly 78% of non-English prompt runs, at least one fan-out query was conducted in English.
This has real consequences for non-English-speaking markets. When Polish users asked about local auction portals, ChatGPT's English-language fan-outs surfaced global platforms like eBay instead of Allegro, Poland's dominant platform. When German users asked about German software companies in German, no German companies appeared in the results. When Spanish users asked about cosmetics brands, no Spanish brands came up either.
The reason behind this is fairly straightforward. Around 50% of internet content is written in English, and ChatGPT tends to prioritize sources it can more reliably evaluate for authority. By switching to English, it taps into a broader and more cited pool of content.
For marketers and brands operating in non-English-speaking regions, this is worth understanding. It means that even if your audience is local, having well-structured, authoritative English-language content can directly affect whether you show up in AI answers, regardless of the language your potential customers are prompting in.
Once you can see what ChatGPT is searching for, you'll know exactly which keywords to focus on when optimizing your content. This gives you a clearer understanding of how your brand can show up in AI-generated answers, not just traditional search results.
It's one of those small but powerful insights that a lot of people overlook. Knowing what prompts and queries AI systems are using helps you position your brand where it matters most, inside the answers themselves.
This approach connects directly with AEO, or Answer Engine Optimization. The idea is that as more people turn to AI chatbots instead of Google searches, understanding how these systems find and reference your content becomes just as important as ranking on page one of search results.
Finding the fan-out queries is only the first step. The real value comes from knowing what to do with them.
Use them as keyword targets. The fan-out queries ChatGPT sends to Google are essentially the exact phrases you need to rank for if you want to be cited in AI answers. Treat them the way you'd treat keyword research, but instead of guessing what someone might search for, you're reading what the AI actually sent.
Identify your content gaps. If you see ChatGPT consistently searching for specific angles or subtopics you haven't covered, that's a direct signal of where to build new content. If your site doesn't rank for those fan-out queries, you won't be in the pool of sources the model pulls from.
Group queries into clusters. ChatGPT often fans out in multiple directions from a single prompt. If you can map the full cluster of fan-out queries around a topic, you can build content that covers multiple branches of it. Brands showing up across multiple fan-out branches tend to see significantly higher citation rates than those visible for only one.
Prioritize structure and clarity early in your content. A meaningful share of AI citations come from the first section of a page. How you open an article, and how clearly you state your main point up front, directly affects whether the model picks you as a source.
Track changes over time. Fan-out query patterns shift as the AI evolves. What ChatGPT searched for three months ago may not be what it searches for today. Monitoring these queries regularly gives you a way to stay ahead of those changes rather than catching up to them.
The DevTools method described in this article is a great starting point for understanding what's happening. But if you want to track this at scale, across many prompts and topics over time, doing it manually isn't practical.
A few tools have been built specifically for this:
AI Peekaboo tracks fan-out queries automatically and gives you a clear view of which queries the AI is sending for your target prompts. It's designed for marketers who want to monitor this systematically without having to dig through network tabs every time. Full transparency: I'm one of the co-founders. You can find it at aipeekaboo.com.
Peec.ai is another strong option, with solid research credentials and a platform that covers AI search visibility across multiple models. They've done some of the most thorough public research on fan-out query behavior to date.
Profound is the more enterprise-grade option and significantly more expensive, but it offers deep fan-out analysis and is worth knowing about if you're working at that scale.
Otterly.ai is also worth looking at, particularly for teams focused on tracking how often prompts trigger web searches and how AI platforms cite content.
ChatGPT searches the internet by sending out fan-out queries to either Google's or Bing's search engines.
The method discussed explains how you can extract and identify these fan-out queries, giving you a significant advantage for your SEO and AI SEO strategy. When you know what queries ChatGPT sends to Google or Bing, you can adjust your content to align with them. It's one of the easiest ways to start optimizing for both SEO and AI visibility simultaneously. The approach has proven effective for understanding how AI engines source and prioritize information.
P.S. A 4-minute video walkthrough of this entire process is available for reference.
I'm Filipe, the CEO & Co-Founder of Peekaboo. I lead all commercial and customer facing functions here at the company. I am obsessed about making sure our customers are heard and have a great experience with us!