On July 23, 2026, Profound published the most detailed technical breakdown of ChatGPT Shopping produced to date. The research covers 201,137 unique prompt runs and 812,190 product cards from June 18 to June 25, 2026. The goal was to map the full architecture of the shopping experience from the user's initial query through to the offer list that drives actual clicks to merchant sites.

The report is the third in a series. The first, published in March 2026, tracked roughly two million prompts to establish how often Shopping mode activates. The second reverse-engineered the Shopping trigger itself, finding that category beats intent: naming a shippable product is a six-times lift over baseline, while purchase-oriented language without a product noun barely moves the needle. This third report goes deeper, into the mechanics of how product cards are built once Shopping mode fires.

The Architecture, Explained

The key finding is that ChatGPT Shopping operates through two distinct query fanout mechanisms. The first is the standard query fanout that AEO practitioners already know: ChatGPT rephrases the user's prompt into multiple internal sub-queries to build the text narrative. The second, which Profound calls product query fanout, is specific to Shopping mode. ChatGPT also translates the user's prompt into internal product sub-queries used to identify product card candidates. These are separate processes with separate retrieval logic.

Once product card candidates are selected, commercial information for each card is sourced from one of three places. Direct merchant sources account for just under 25% of retrievals. ChatGPT's own internal fallback search index handles a significant share. Google's commercial index covers the remainder. Bing does not appear in the retrieval stack. For the 12.7% of cards sourced from direct product feeds, the logic is simpler: commercial information always comes from the feed itself.

The most operationally important finding is how the offer list connects to the product card. When offers are successfully retrieved for a card, the card's commercial information, including product name, price, and merchant, is always pulled from the rank 1 offer. When no offer is retrieved, ChatGPT falls back to its own product search. Rank 1 in the offer list is not just a conversion lever. It determines what information the product card shows.

What Drives Product Card Rank

Profound compared the top-ranked product cards (rank 1) against the bottom-ranked (rank 4 and below) across 415,276 cards. Two factors showed the largest lift. GPT tag presence produced a 144% lift. Median review count produced a 124% lift, with top-ranked cards showing a median of 787 reviews versus 352 for bottom-ranked cards. Promotional pricing produced a modest 13% lift. Offer presence, URL length, and product name length showed no meaningful contribution to rank.

GPT tags are machine-assigned labels based on available product data. They are not set by the merchant directly. Tags in the value and premium clusters are the most prevalent, followed by style, everyday, and classic groupings. The practical implication is that product page and feed data that clearly signals a product's positioning, whether budget, premium, or performance-oriented, increases the probability of a tag being assigned and the card ranking higher.

The Citation Layer

Reddit is the most indexed domain in ChatGPT Shopping citations, accounting for roughly one third of all shopping citations. This is consistent with broader findings across non-shopping ChatGPT responses. For brands, it reinforces the case for maintaining a credible presence in Reddit communities relevant to their product categories, not as a direct commerce channel, but as a citation source that influences which products get surfaced and how they are described.

The 87.3% of product cards retrieved from web crawl, rather than product feeds, underscores that organic discoverability still dominates the Shopping surface. Feed integration matters for the 12.7% of cards it covers, and it matters significantly for offer list rank. But the majority of product card selection happens through the same web crawl and search index signals that determine general AI visibility.

What This Means for Retailers and AEO Practitioners

Profound draws four operational conclusions from the research. First, product query fanout requires its own optimization strategy. Brands need to align product listings and content with the specific query fanouts ChatGPT uses for product searches, not just the user-facing prompt. Second, ChatGPT fallback search matters: product listings must appear prominently when ChatGPT searches for the exact product name. Third, Google Shopping index signals feed directly into ChatGPT's retrieval stack, so Google Shopping optimization is not separate from AEO for retailers. Fourth, general citation share influences product card candidate selection, meaning content strategy and AEO are not separate workstreams for e-commerce brands.

The research also clarifies the relationship between the offer list and the product card in a way that changes how retailers should think about feed optimization. Getting into the offer list is not just about driving clicks from the sidebar. It determines the commercial information displayed on the card itself. A brand that ranks second in the offer list may still lose the product card display to the rank 1 merchant.

References

[1] Allen Wu, Profound, <a href="https://www.tryprofound.com/blog/chatgpt-shopping-end-to-end-breakdown" target="_blank" rel="noopener noreferrer" class="text-amber-700 underline">Breaking down how ChatGPT Shopping works behind the user experience</a>, July 23, 2026.

[2] Brandon Punturo, Profound, <a href="https://www.tryprofound.com/blog/chatgpt-shopping-prediction" target="_blank" rel="noopener noreferrer" class="text-amber-700 underline">We reverse-engineered ChatGPT's Shopping trigger. Here's how it works.</a>, March 17, 2026.

[3] Profound, <a href="https://www.tryprofound.com/blog/chatgpt-shopping-analysis" target="_blank" rel="noopener noreferrer" class="text-amber-700 underline">We tracked 2 million ChatGPT prompts. Shopping showed up less than 10% of the time.</a>, March 3, 2026.

AEO Updates is published by The Prompt Group. Editorial decisions sit with the AEO Updates team, and any commercial relationship that touches a story is labelled on the page.