In the rapidly evolving landscape of artificial intelligence and digital commerce, the question of where AI models derive their data has become a central focus for marketers, SEO professionals, and tech enthusiasts. For a long time, the industry assumption was that ChatGPT, through OpenAI’s close partnership with Microsoft, relied almost exclusively on Bing for its real-world data retrieval. However, a groundbreaking new study has revealed a startling shift in this dynamic, specifically regarding how ChatGPT handles e-commerce and product recommendations.
Recent forensic analysis of ChatGPT’s source code and output behavior has uncovered a significant trend: OpenAI’s flagship chatbot is now sourcing approximately 83% of its carousel products directly from Google Shopping. This discovery, centered around a process known as “shopping query fan-outs,” suggests that despite its corporate ties to Microsoft, ChatGPT is increasingly leaning on Google’s massive shopping index to power its consumer-facing product recommendations.
For brands and retailers, this finding is more than just a technical curiosity; it represents a fundamental shift in how “AI SEO” works. If your products aren’t ranking on Google Shopping, the chances of them appearing in a ChatGPT product carousel are now statistically slim. Let’s dive deep into the mechanics of this study, the data behind the findings, and what it means for the future of digital publishing and e-commerce.
The Technical Smoking Gun: Decoding id_to_token_map
The investigation into ChatGPT’s sourcing began in late 2025, when researchers identified a mysterious field within the platform’s source code labeled id_to_token_map. While this field initially appeared to be a string of nonsensical characters, it was actually base64 encoded. Upon decoding this data, researchers found a treasure trove of information that pointed directly to Google’s infrastructure.
The decoded fields contained specific Google Shopping parameters, including productid, offerid, and various language or locale identifiers. Most importantly, the data revealed the exact query used by the AI to look up a specific product—a process known as a “shopping query fan-out” (QFO).
To verify this connection, researchers attempted to reconstruct full Google Shopping URLs using only the parameters extracted from ChatGPT’s code. For example, when a user asked for the “best smartphones under $500,” ChatGPT generated a product carousel. By extracting the hidden parameters from that carousel, researchers were able to generate a link that led directly to the exact product page on Google Shopping. The match was not just a similarity; it was a precise architectural link, proving that ChatGPT wasn’t just “finding” these products on the web—it was actively querying Google’s specialized shopping index.
Understanding Query Fan-Outs: Search vs. Shopping
To understand why this is happening, we must first understand the concept of a “Query Fan-Out” (QFO). When you submit a prompt to an AI like ChatGPT, the model doesn’t just “know” the answer if the information is recent or specific. Instead, it generates several internal search queries—fan-outs—to gather data from the web before synthesizing a response. This is the core of Retrieval-Augmented Generation (RAG).
The study analyzed over 1.1 million shopping QFOs to determine if they differed from standard search QFOs. The results were telling. Shopping QFOs were unique to the user prompt 99.7% of the time, meaning the AI creates a very specific “shopping-only” search path that is distinct from its general knowledge retrieval.
Word Count and Intent
There is a distinct difference in the complexity of these queries. General search QFOs, used to gather context for a written answer, averaged about 12 words in length. This makes sense, as contextual retrieval benefits from the nuances of vector search, which requires more linguistic detail to find relevant web pages.
In contrast, shopping QFOs averaged only seven words. This brevity indicates a different objective. Rather than seeking a broad narrative or an article, the AI is targeting a structured index. It essentially acts as a “searcher” on Google Shopping, using concise keywords to trigger the most relevant product listings. The study suggests that for ChatGPT to populate an eight-product carousel, a single page of Google Shopping results is usually sufficient.
Frequency of Queries
The study also found that ChatGPT uses fewer queries for shopping than for general information. On average, a prompt triggers 2.4 search fan-outs but only 1.16 shopping fan-outs. This efficiency further supports the theory that ChatGPT is relying on the heavy lifting already performed by Google’s ranking algorithms. Instead of “shopping around” across multiple search engines, it goes to the most comprehensive source, retrieves the top results, and displays them.
Google vs. Bing: The Battle for the Carousel
The most striking aspect of this research is the disparity between Google and Bing. Given the multi-billion dollar partnership between OpenAI and Microsoft, one would expect Bing Shopping to be the primary source for these carousels. The data, however, tells a different story.
Researchers analyzed 43,000 products across 5,000 ChatGPT carousels, comparing them against the top 40 organic results from both Google and Bing. The methodology involved a multi-stage matching algorithm to account for minor differences in product titles or formatting.
The Findings
- Google Shopping Overlap: Over 83% of the products featured in ChatGPT carousels were found within the top 40 organic results on Google Shopping.
- Bing Shopping Overlap: Only 11% of the products appeared in Bing’s top 40 results.
- Exact Matches: 45.8% of ChatGPT’s product titles were an exact string match for Google Shopping titles. For Bing, the exact match rate was a negligible 0.48%.
- Exclusive Sourcing: Out of 43,000 products, only 70 (0.16%) were found exclusively on Bing. In almost every instance where a product appeared on Bing, it was also present—and usually ranked higher—on Google.
These numbers indicate that ChatGPT’s product retrieval system is almost entirely dependent on Google’s organic shopping index. While it may still use Bing for general web context (the text-based portions of the answer), the “visual” commerce portion of the experience is powered by Google.
Positional Bias: Why Ranking Still Matters
For years, SEOs have lived by the mantra that “the best place to hide a dead body is the second page of Google.” It appears this rule applies to AI as well. The study found a clear correlation between a product’s rank on Google Shopping and its likelihood of appearing in a ChatGPT carousel.
The data revealed that 60% of the products ChatGPT selects as “strong matches” come from the top 10 results on Google Shopping. When expanding that to the top 20, the match rate jumps to nearly 84%. There is a visible “sloping trendline” where the first position in a ChatGPT carousel often corresponds to a top-5 ranking on Google Shopping.
This suggests that ChatGPT isn’t just picking products at random from the index; it is respecting the authority and relevance already established by Google’s ranking engine. If your product is sitting at position 35 on Google Shopping, your chances of being “hand-picked” by an AI for a recommendation carousel are drastically lower than if you were in the top three.
Branded vs. Non-Branded Queries
Interestingly, this behavior remains consistent regardless of whether the user searches for a specific brand or a general category. Whether a user asks for “best Nike running shoes” or just “affordable running shoes,” the reliance on Google Shopping’s top-tier results remains the same. This reinforces the idea that this is a systemic architectural choice by OpenAI rather than a quirk of specific types of queries.
Implications for E-commerce SEO and Digital Marketing
The discovery that ChatGPT is effectively a “Google Shopping wrapper” for its commerce features has massive implications for the industry. It changes the priority list for digital marketers who want to stay relevant in the age of generative AI.
The Convergence of SEO and Feed Management
Historically, Google Shopping was managed by PPC (Pay-Per-Click) teams, while organic search was the domain of SEOs. This data suggests those walls must come down. Because ChatGPT is pulling from the organic shopping results (not the paid ads), optimizing your product feed for organic visibility is now a prerequisite for AI discovery. This includes ensuring high-quality product titles, detailed descriptions, and accurate metadata—all factors that Google’s algorithm uses to rank organic shopping listings.
The Role of Sentiment and Context
While Google Shopping provides the “selection set,” it is likely that ChatGPT uses other factors for the final ranking within its own carousel. The study notes that while 83% of products match Google, the order isn’t always identical. This implies that ChatGPT may be cross-referencing the products it finds on Google Shopping with sentiment data it finds through its general search QFOs. If a product ranks #1 on Google Shopping but has poor recent reviews or negative sentiment in the articles ChatGPT “reads” during its search fan-out, the AI might move it further down the carousel or replace it entirely.
OpenAI’s Strategic Independence
This finding also sheds light on OpenAI’s broader strategy. By sourcing shopping data from Google despite its partnership with Microsoft, OpenAI is demonstrating a commitment to “utility over loyalty.” Google Shopping is widely considered the most robust and comprehensive product index in the world. For OpenAI to provide the best user experience, it must use the best data source, even if that source belongs to its partner’s biggest rival.
Methodology: How the Study Was Conducted
To ensure the findings were robust, researchers utilized Peec AI data to analyze a massive dataset. The goal was to eliminate any “flukes” and prove that this behavior was consistent across different industries.
Verticals Analyzed
The study didn’t just look at one niche. It covered 10 major industry verticals to ensure the findings were universal:
- Apparel & Footwear
- Electronics
- Home & Kitchen
- Beauty & Personal Care
- Toys & Games
- Pet Supplies
- Sports & Outdoors
- Baby & Kids
- Home Improvement
- Office Supplies
The Matching Algorithm
Matching products between two different platforms is difficult because titles are often formatted differently. To solve this, the study used a three-stage “cascade” approach:
- Exact Match: Simple case-insensitive string equality.
- Near-Exact Match: Used the Python SequenceMatcher to find products with a 0.95 similarity score or higher (catching minor differences like punctuation or dashes).
- Hybrid Match: A weighted average (40% character similarity, 60% word overlap) to identify the same brand and product name even if the word order was slightly different.
A score of 0.8 was set as the threshold for a “strong match,” which researchers determined was a reliable indicator that the AI was looking at the exact same brand and specific product model.
Conclusion: The New Roadmap for AI Visibility
The “New Finding” that ChatGPT sources the vast majority of its carousel products from Google Shopping is a wake-up call for the e-commerce industry. It clarifies a previously opaque part of the AI’s “brain” and provides a clear roadmap for brands that want to be recommended by the world’s most popular AI.
We are entering an era where e-commerce success is defined by a “triple-threat” strategy:
- Google Shopping Optimization: Maintaining a top-tier organic ranking on Google Shopping is now the primary gateway to being featured in ChatGPT.
- Traditional SEO: Building authority and positive sentiment in web articles and reviews provides the “contextual proof” that AI models look for when validating their product selections.
- Technical Feed Accuracy: Ensuring that product IDs and offer data are structured correctly, as these are the “tokens” that the AI uses to identify and link products in its internal code.
As OpenAI continues to refine its search and shopping capabilities, we may see these sources shift. However, for the last several months, the trend has been undeniable. If you want to sell products through ChatGPT, your journey doesn’t start in a chat box—it starts in the Google Shopping index.