The World's First Visualization Tool for ChatGPT's Internal Search Queries Unveils Its Process

Introduction

In a groundbreaking move, Queue Inc., headquartered in Chuo-ku, Tokyo, has introduced a pioneering tool that allows users to visualize ChatGPT's internal search queries. This innovative free tool, part of their AI search optimization service known as umoren.ai, sheds light on the processes employed by ChatGPT to generate responses.

Visualizing Query Fan-out

The newly launched tool empowers users to simply enter prompts into ChatGPT and observe the actual search queries it executes in real-time. This brings unprecedented transparency to the information gathering process of generative AI, previously regarded as a black box. Essentially, this tool provides users with a firsthand view of how ChatGPT conducts its research for generating answers.

For instance, one user tested the tool by entering the following prompt: "I am looking for a coffee maker suitable for living alone. Ideally, it should brew from beans automatically and be easy to clean. Please recommend 3 to 5 models, along with their features (extraction method, size, maintenance, and price)."

Understanding the Search Process

During the demo, it became clear that ChatGPT does not generate responses based on a single search. Instead, the AI engages in a multi-round inquiry process. Observations from the tool revealed distinct structures within ChatGPT's search queries.

Search Structure:

The observation of prompts through the tool highlighted that ChatGPT conducts multiple rounds of searches as it forms answers to user queries. Below is the example observed:

1. Broad Exploration Phase: The first round involves scouting for a wide range of information. For example, in a query like "Fully automatic coffee maker for living alone, easy to clean, 2026 official", ChatGPT scans various product categories, existing market offerings, and related review articles.

To summarize, this phase serves as an initial exploration to identify potential candidates.

2. Specifically Targeted Searches: The subsequent phase showcases a marked shift in the nature of the searches. Queries at this stage include specific brand names like Delonghi, Panasonic, and siroca. This indicates that ChatGPT, after its broad exploration, delves deeper into the individual candidates identified in the initial phase.

Additionally, the presence of site-specific searches, such as 'site:delonghi.com', signifies that ChatGPT explicitly targets particular sites for gathering information.
Essentially, the overall process can be viewed as a two-step research workflow consisting of:
- Broad exploration → Candidate extraction → Targeted searches and deep dives into specified sites.

Integrating Multiple Information Sources

The tool not only reveals the executed search queries but also allows users to see the sources of information that ChatGPT consolidates to generate its final answer. This finding underscores that ChatGPT does not merely summarize a single page, but rather draws from multiple sources to create comprehensive responses.

Implications for AI Search Optimization

The structure of the searches presents crucial insights for the future of AI search optimization. Traditional search engine optimization (SEO) has primarily focused on achieving high rankings in search results, which is no longer sufficient in the context of AI. The necessity now lies in being recognized as a candidate in ChatGPT's exploratory phase, being targeted through specific searches, and ultimately being cited as a source of official information.

This evolution leads to the concept of Large Language Model Optimization (LLMO), defined by Queue Inc. as a method to enhance the visibility and relevance of content for AI-driven searches.

How to Use the ChatGPT Query Fan-out Visualization Tool

This innovative tool is straightforward to use. By simply entering a prompt and clicking on “Retrieve ChatGPT’s Fan-out Queries,” users can gain insights into: