New finding: ChatGPT sources 83% of its carousel products from Google Shopping via shopping query fan-outs
The landscape of artificial intelligence and search engine technology is shifting at a breakneck pace. For years, the industry assumption was that OpenAI’s partnership with Microsoft meant that ChatGPT would naturally lean on Bing for its real-time data needs. However, as OpenAI pursues greater independence and refines its search capabilities, a surprising new reality has emerged. A comprehensive study into ChatGPT’s product recommendation engine has revealed a staggering reliance on Google Shopping, rather than Microsoft’s own search infrastructure.
New research indicates that approximately 83% of the products appearing in ChatGPT’s interactive shopping carousels are sourced directly from Google Shopping. This discovery was made by analyzing “query fan-outs” (QFOs)—the behind-the-scenes search queries the AI generates to fetch live data. The findings suggest that despite its corporate ties to Redmond, OpenAI’s “Search” functionality is deeply intertwined with the Mountain View ecosystem when it comes to e-commerce and product discovery.
Understanding the Technical Framework: What is a Query Fan-Out?
To understand how ChatGPT chooses which products to show you, we must first look at the mechanics of its retrieval-augmented generation (RAG) process. When you ask ChatGPT for the “best running shoes for flat feet,” the model doesn’t just rely on its training data. It generates specific sub-queries to browse the web for current pricing, availability, and reviews. These sub-queries are known in the research community as Query Fan-Outs (QFOs).
In late 2025, researchers identified a hidden field within ChatGPT’s source code labeled id_to_token_map. When this field is decoded from its Base64 format, it reveals the specific parameters the AI uses to identify products. These parameters include specific identifiers such as productid and offerid, as well as locale and language settings. Most importantly, these parameters are identical to those used by Google Shopping’s internal indexing system.
The Shopping QFO vs. The Search QFO
The study found that ChatGPT treats product discovery as a fundamentally different task than general information gathering. There are two distinct types of fan-outs occurring simultaneously:
- Search Query Fan-Outs: These are longer, more descriptive queries (averaging 12 words) used to find blog posts, reviews, and articles. They are designed for vector search—comparing “chunks” of text to find the most relevant context for a written response.
- Shopping Query Fan-Outs: These are shorter (averaging 7 words) and highly targeted. Their sole purpose is to hit a structured shopping index to populate the visual carousel.
The data shows that while a single prompt might trigger multiple search fan-outs to gather information, it usually triggers only one or two shopping fan-outs. This suggests that ChatGPT relies on a single authoritative source—Google Shopping—to fill its eight-product carousel in one go.
Inside the Study: Measuring the Google vs. Bing Divide
To prove that this wasn’t an anecdotal fluke, researchers utilized data from Peec AI to conduct a large-scale analysis. The study scrutinized over 43,000 products appearing in ChatGPT carousels across 10 major industry verticals. These included highly competitive categories like Electronics, Beauty & Personal Care, Home & Kitchen, and Apparel.
The researchers then cross-referenced these ChatGPT results against the top 40 organic shopping results from both Google and Bing. To ensure accuracy, they excluded paid advertisements and sponsored listings, focusing entirely on organic rankings.
The Matching Methodology
Matching products across different platforms is notoriously difficult because titles are often rewritten or truncated. To solve this, a three-stage matching algorithm was used:
- Stage 1: Exact Match. A strict comparison of strings, ignoring case and whitespace.
- Stage 2: Near-Exact Match. Using a sequence matcher to account for minor differences in punctuation or special characters (like different types of dashes).
- Stage 3: Hybrid Match. A weighted average of character-level similarity (40%) and word overlap (60%).
A “strong match” was defined as any product reaching a similarity score of 0.8 or higher. This threshold typically ensures that the brand and the specific product model are identical, even if the descriptive text varies slightly.
The Findings: A Near Total Dominance for Google
The results of the comparison were conclusive. Across the 43,000 products analyzed, 45.8% were an exact string match with Google’s organic shopping results. For Bing, that number plummeted to just 0.48%.
When looking at “strong matches” (the 0.8 threshold), 83.3% of ChatGPT’s carousel products were found within the top 40 Google Shopping results. In contrast, Bing only shared 10.9% of the products featured in ChatGPT. More tellingly, of the few products Bing did match, nearly all of them were also present in the Google results. Only 0.16% of the products—a mere 70 items out of 43,000—were exclusive to Bing. This confirms that ChatGPT is almost certainly not using Bing as a primary or even secondary source for shopping data.
The Influence of Rank: Positional Bias in the Carousel
One of the most critical takeaways for e-commerce brands is the correlation between Google Shopping rank and ChatGPT carousel placement. The study found a clear “sloping trendline” that links the two.
If a product ranks in the top five on Google Shopping, it is significantly more likely to appear in the first or second position of the ChatGPT carousel. The data revealed that 60% of all strong matches in the ChatGPT carousel were pulled from the top 10 results on Google. When expanding that to the top 20 Google results, the match rate rises to nearly 84%.
This suggests that ChatGPT isn’t just picking random products from the web; it is effectively “cloning” the top of the Google Shopping organic index. If your product doesn’t rank on the first page of Google Shopping for a specific query, the chances of it appearing in a ChatGPT recommendation are statistically slim.
Analyzing Performance Across Industry Verticals
The study was designed to be robust, covering 10 different industries to ensure the behavior wasn’t limited to a specific niche. The findings remained consistent across the board, proving that this is a systemic architectural choice by OpenAI.
Branded vs. Non-Branded Queries
Researchers also looked at whether the type of prompt changed the sourcing behavior. They compared branded prompts (e.g., “best Nike running shoes”) with non-branded prompts (e.g., “best running shoes under $100”).
Surprisingly, the match rate for Google Shopping remained high for both. Non-branded queries actually showed a slightly higher match rate, indicating that when ChatGPT has to “decide” which products are best without a specific brand mentioned, it leans even more heavily on Google’s ranking signals to determine what is relevant and high-quality.
What This Means for SEO and E-commerce Strategy
For digital marketers, SEO specialists, and retail brands, this “new finding” is a game-changer. For the last year, much of the conversation around “AI Optimization” (AIO) or “Generative Engine Optimization” (GEO) has focused on being mentioned in articles and citations. While that remains important for the written portion of ChatGPT’s responses, the shopping carousel operates by a different set of rules.
The Google Shopping Power Play
If you want your products to appear in ChatGPT, your primary focus should be on traditional Google Shopping optimization. This includes:
- Feed Accuracy: Ensuring your product titles, descriptions, and GTINs are perfectly aligned with Google Merchant Center requirements.
- Organic Ranking Factors: Improving your product’s organic visibility on Google through high-quality images, competitive pricing, and positive reviews.
- Keyword Alignment: Since ChatGPT’s shopping fan-outs are short and direct, ensuring your product titles match common “shopping” queries is essential.
The Sentiment Factor
While Google Shopping rank is the primary “selection set” for the carousel, the study suggests that the final ranking within the ChatGPT carousel might be influenced by secondary factors. Because ChatGPT also runs “Search QFOs” to find web context, it likely parses the sentiment of the products it finds on the Google index. If a product ranks #1 on Google Shopping but has poor reviews or negative sentiment in the articles ChatGPT browses, it may be demoted in the carousel in favor of the #2 or #3 product that has glowing reviews.
Why Does OpenAI Use Google Instead of Bing?
The industry is left wondering why OpenAI would favor Google over its primary investor, Microsoft. While neither company has commented on the specific data flow, several architectural reasons likely play a role:
1. Data Maturity and Index Size
Google Shopping has historically maintained a larger and more frequently updated index of e-commerce products globally. For an AI that prides itself on accuracy and real-time information, Google provides a more reliable data stream for stock status and pricing.
2. Structured Data Integration
Google’s “Merchant Center” infrastructure provides highly structured data that is easy for an LLM to parse. The productid and offerid parameters found in ChatGPT’s code suggest a deep technical integration that might simply be more efficient than Bing’s current shopping API.
3. User Intent Matching
Google’s ability to map user intent to specific products is widely considered the gold standard in search. By “piggybacking” on Google’s ranking algorithm, OpenAI ensures that the products its AI recommends are already vetted by one of the most sophisticated recommendation engines in the world.
The Future of AI Shopping Retrieval
This study represents a snapshot of ChatGPT’s behavior over a four-month period. In the world of AI, four months is an eternity, and OpenAI is known for constant iteration. It is entirely possible that OpenAI is developing its own proprietary shopping index or that a future update could shift the weight back toward Bing or even a partnership with Amazon.
However, for the foreseeable future, the “Google-to-ChatGPT” pipeline is the dominant force in AI-driven e-commerce. As ChatGPT continues to evolve into a full-fledged search engine (SearchGPT), the line between traditional search and AI assistance will continue to blur.
Final Thoughts for Retailers
The revelation that 83% of ChatGPT’s carousel products come from Google Shopping simplifies the roadmap for retailers. You do not need a separate “AI Strategy” for ChatGPT carousels; you need a world-class Google Shopping strategy.
By focusing on your organic performance within the Google ecosystem, you are effectively optimizing for the most popular AI assistant in the world. The connection between these two tech giants—one the king of search, the other the king of AI—is currently the most important bridge in digital commerce. Understanding and exploiting that bridge will be the key to winning the next era of online retail.