New Data Reveals Massive Shift in ChatGPT Shopping as Feed-Integrated Recommendations Overtake Web Search

The landscape of conversational commerce underwent a dramatic structural transformation this summer, according to observational data released by digital intelligence firm Profound. Tracking millions of prompt runs on OpenAI’s ChatGPT platform, the firm’s data reveals a decisive pivot away from traditional web-search retrieval toward direct, feed-integrated product recommendations. This shift, which accelerated sharply following the deployment of OpenAI’s GPT-5.6 model in July, has reshaped visibility dynamics for online retailers and underscored the critical importance of structured data pipelines in the generative artificial intelligence ecosystem.

According to Profound’s empirical analysis—which evaluated nearly 1.8 million tracked prompt runs throughout July—the share of ChatGPT Shopping product recommendations classified as feed-integrated surged from 8.26% to 61.54% on July 10 alone. This sudden inversion marked a watershed moment for how artificial intelligence models surface commercial products to consumers, moving away from dynamic, real-time web scraping and toward curated, structured product catalogs.

While the findings are derived from a proprietary sample of customer prompts rather than OpenAI’s exhaustive, global shopping traffic, the scale of the dataset offers unprecedented insight into the backend mechanics of AI-driven product discovery. The implications of this shift extend far beyond algorithmic technicalities, signaling a new era of digital shelf-space optimization where merchant integration pipelines dictate market visibility.

Chronology of a Transformation: The July 10 Inflection Point

The timeline of this algorithmic evolution points directly to early July as the crucible for change. On July 9, OpenAI released GPT-5.6, noting in its communications that the rollout would be executed progressively over a 24-hour window. Coincidentally or by direct design, July 10 marked the exact inflection point where Profound’s telemetry detected a massive spike in feed-sourced recommendations.

Before July 10, feed-sourced recommendations constituted a minority tier within Profound’s tracked prompt runs. After that date, they immediately established a dominant majority. In a broader longitudinal sample evaluated by the firm—spanning 97,725 prompt runs from July 1 through August 24—feed retrieval steadily consolidated its dominance, ultimately surpassing web search entirely by August. By September 3, Profound’s reporting indicated that feed retrieval accounted for approximately 65% of all tracked product recommendations.

Despite the tight correlation between the deployment of GPT-5.6 and the dramatic shift in shopping retrieval methods, OpenAI has maintained radio silence regarding the connection. Official ChatGPT release notes for July 9 and July 10 contained no explicit mentions of shopping-related algorithmic updates, leaving third-party analysts and merchants to reverse-engineer the technological adjustments based on output behavior.

Winners and Losers: Store Visibility Swings and Market Concentration

The sudden transition from web search to feed retrieval was not without casualties. For many e-commerce brands, the July 10 shift resulted in severe volatility in consumer visibility.

Profound’s sample included 687 active customer accounts that consistently triggered shopping-related prompts and maintained at least one product card pointing to their inventory between July 7 and July 12. A comparative analysis of visibility metrics across this cohort revealed stark bifurcation: out of the 687 merchants, 450 experienced a staggering reduction of at least one-third in their shopping visibility when comparing the periods of July 7–9 and July 10–12. Conversely, a mere 67 merchants enjoyed an increase of the same magnitude.

Statistical modeling applied to a subset of 517 affected merchants revealed that the degree of web-search retrieval lost, balanced against the amount of feed retrieval gained, could explain an overwhelming 83% of the variation in their visibility changes. This strong statistical correlation highlights how heavily brand presence became dependent on feed accessibility over traditional search engine optimization (SEO) tactics.

Furthermore, the data indicates that feed retrieval narrowed the scope of merchant representation. Before the update, a broader array of stores shared the spotlight. Following the shift, shopping retrieval became increasingly concentrated among a select group of elite vendors. The share of references directed to the top 10 most-referenced stores jumped from 22.5% to 41.8%. Concurrently, the total number of unique merchants referenced in the tracked prompts dropped by more than 20%, falling from 13,524 to 10,607 unique entities.

This contraction suggests that feed-based ecosystems inherently favor established platforms with robust, pre-existing data integrations, potentially squeezing out independent merchants who rely solely on open-web discovery.

The Architectural Backbone: Shopify, Etsy, and the Agentic Commerce Protocol

To understand why certain merchants reaped the benefits of the July update while others saw their visibility plummet, industry experts have looked closely at OpenAI’s official developer documentation and merchant guidelines.

According to OpenAI’s Help Center, ChatGPT sources product data directly from structured feeds provided by data aggregators and individual stores. The organization maintains that product results are selected independently by the artificial intelligence, asserting that they are neither advertisements nor influenced by commercial partnerships.

For millions of online merchants, integration depends heavily on pre-existing platform partnerships. Shopify stores, for instance, enjoy native integration via Shopify Catalog, allowing product data to flow seamlessly into the ChatGPT ecosystem without requiring manual merchant intervention. Similarly, Etsy catalogs are directly connected to the system.

Other independent retailers seeking direct feed access must navigate an application process through OpenAI’s dedicated merchant portal, which currently operates under a substantial waitlist. These data feeds are powered by the Agentic Commerce Protocol—OpenAI’s standardized framework for transmitting structured product data, pricing, and descriptions to conversational models. Originally extended to product discovery on March 24, the protocol supports data ingestion from major financial and commerce infrastructure providers, including Stripe and Salesforce.

Despite these architectural details, official OpenAI documentation remains silent on the exact algorithmic weighting dictating when the model relies on feed data versus open-web extraction.

Broader Implications for E-Commerce and Digital Marketing

The revelations from Profound’s tracking data carry profound implications for digital marketers, brand strategists, and e-commerce operators. Traditional search engine optimization—long the cornerstone of digital visibility—is increasingly sharing space with generative engine optimization (GEO) and data-feed management.

For brands operating outside of native ecosystem giants like Shopify or Etsy, securing visibility in AI-driven shopping environments requires proactive engagement with structured data protocols. The data clearly demonstrates that merchants lacking feed integration are increasingly disadvantaged as conversational AI platforms prioritize structured, verified, and easily parsable product feeds over ambiguous web-search results.

However, analysts note important limitations in current observational studies. Datasets derived from third-party prompt tracking reflect specific simulated customer journeys rather than aggregate consumer traffic across the entire global user base. Furthermore, while these metrics illustrate where data is being pulled from, they do not yet provide absolute clarity on how individual products are ranked once a feed is successfully integrated.

Looking Ahead: Self-Serve Feeds and the Next Phase of AI Competition

As the retail technology sector looks toward the remainder of the year, further evolution in AI-driven commerce is virtually guaranteed. OpenAI has indicated plans to expand the regional availability of ChatGPT Shopping and intends to launch a new self-serve platform later this year. This upcoming tool is expected to democratize the process, allowing individual stores to connect and manage their product feeds independently without relying entirely on major platform intermediaries or protracted waitlists.

Industry experts anticipate that as self-serve options proliferate, the nature of competition will shift. Rather than merely possessing a feed, brands will likely compete on the granularity, richness, and contextual depth of the data fields contained within those feeds—optimizing attributes, specifications, and real-time inventory metrics to satisfy complex generative queries.

Adding another layer of technical evolution, OpenAI began rolling out its advanced GPT-6 Astra model to a limited cohort of organizations on September 3. As newer, more capable models continue to supersede older iterations like GPT-5.6, the underlying mechanics of product discovery will undoubtedly continue to evolve. For merchants navigating this rapidly shifting digital frontier, adaptability, structured data readiness, and a deep understanding of AI retrieval pipelines have officially become mandatory for survival.