Discovery Science · established evidence
Why AI Shopping Assistants Skip Stores With Great Design and Bad Product Data
When a shopper asks ChatGPT, Google's AI Mode or Perplexity for the best product under a budget, the engine does not browse a website the way a person does. It reads structured product data: the Merchant Center feed, the Product and Offer schema on the page, the GTIN, the price, the availability, the review fields, and it builds an answer from what that data says. A store with beautiful photography and a thin or contradictory feed loses to a plainer competitor whose data is complete and accurate, because the engine never sees the photography at all. The clearest evidence this is the real bottleneck is not a theory. It is OpenAI's own admission: when it relaunched shopping inside ChatGPT in 2026, it pulled back from letting people buy inside the chat, citing inaccurate product data and difficult merchant onboarding, not weak shopper demand. The fix is not a redesign. It is the data underneath the design.
A shopping assistant does not browse your store the way a person does
A human visitor scrolls, looks at photos, reads a headline, and forms an impression before they read a single spec. An AI shopping assistant does none of that. It ingests structured data, your Google Merchant Center feed if you run one, the Product and Offer schema markup on your pages, the GTIN and brand identifiers, price and availability, and whatever review data is attached to the item, and it uses that data to decide what to name in its answer.
That is a fundamentally different audience than the one most stores are built for. A store can look genuinely excellent, fast, well-photographed, well-written, and still be functionally invisible to an assistant if the feed behind it is incomplete, the schema disagrees with the page, or a required identifier like the GTIN is missing. The design is doing its job for the human who arrives. It is doing nothing for the system that decides whether they arrive at all.
The proof is in OpenAI's own retreat from in-chat checkout
On February 16, 2026, OpenAI relaunched its "Buy it in ChatGPT" shopping feature, extending eligibility to more than 1 million Shopify merchants against a base of roughly 900 million weekly active ChatGPT users, a genuinely large distribution surface for anyone selling online. What happened next is the more important fact for this article.
Within weeks, OpenAI pulled back from its original goal of letting shoppers complete a purchase inside the chat itself, Instant Checkout. Its own developer communications and coverage in CNBC and Digital Commerce 360 cited specific, named reasons: inaccurate product data, difficulty onboarding merchants, and no support for multi-item carts. OpenAI pivoted to prioritizing product discovery over in-chat checkout completion. This is a company with enormous engineering resources and a direct financial incentive to make agentic checkout work, saying in public that the thing standing in the way was the quality of the product data it was being fed, not the technology of the checkout itself or a lack of shopper interest.
The AI-shopping channel is already real revenue, for the stores that show up
The revenue is already here. Adobe Digital Insights, drawing on its first-party analytics panel across thousands of US retail sites, reported AI-referred traffic to US retail sites up 393 percent year over year in the first quarter of 2026, and 693 percent year over year over the 2025 holiday season.
The more striking number is what that traffic does once it arrives. In March 2026, AI-referred traffic converted 42 percent better than non-AI traffic on the same sites, a reversal from March 2025, when AI-referred traffic converted 38 percent worse. AI-referred shoppers also generated 37 percent more revenue per visit and spent 48 percent longer on-site than other visitors. Morgan Stanley's own market model projects agentic commerce could capture 10 to 20 percent of US e-commerce spend, 190 to 385 billion dollars, by 2030. The channel converted worse than average a year ago and now converts better. That is the shape of an emerging channel crossing into a real one, and it rewards the merchants whose data already qualifies them for it.
Where the data actually breaks
Two specific failure points show up repeatedly, and both are things Google itself documents rather than a third party estimating.
A missing GTIN is a documented suppression, not a guess
Google Merchant Center's own help documentation states plainly that products missing a required GTIN are demoted to "limited performance," meaning they compete in a materially smaller set of auctions and surfaces. Google does not publish an exact percentage of impressions lost, so any specific uplift figure circulating online is a vendor estimate, not a Google-disclosed number. What is not in dispute is the direction: an incomplete identifier field suppresses a product Google itself would otherwise be willing to show.
When the feed and the on-page schema disagree, engines trust neither
The same product frequently lives in three places at once: the Merchant Center feed, the Product and Offer schema on the store's own page, and sometimes a marketplace listing, each with its own price, availability and title. When those three disagree, an engine has no reliable way to know which one is true, and the practical result is that it deprioritizes the SKU rather than guessing which source to trust. The fix is not adding more data. It is making the data that already exists agree with itself everywhere it appears.
What "clean data" actually requires
Getting a catalog to the state where an AI shopping surface can confidently recommend it is not one task, it is the reconciliation of several: complete GTIN and brand identifiers across every listed product, Product and Offer schema on the page that states the same price and availability as the feed, titles that lead with what a shopper actually types rather than an internal SKU or name, and a review data layer that is real, current, and compliant with the FTC's 2024 rule against fake or incentivized reviews. None of it is glamorous work, and that is precisely why it is the part most stores, run by an owner without an in-house data team, have never had anyone actively maintain.
The evidence
Key findings, with their sources
-
OpenAI pulled back from letting shoppers complete purchases inside ChatGPT (Instant Checkout), citing inaccurate product data, difficult merchant onboarding, and no multi-item cart support, not weak shopper demand.
established OpenAI developer announcement, 2026-03; CNBC, "OpenAI revamps shopping experience in ChatGPT after struggling with Instant Checkout," 2026-03-24.
-
ChatGPT shopping eligibility extended to over 1 million Shopify merchants as of its February 2026 relaunch, against a base of roughly 900 million weekly active ChatGPT users.
established OpenAI developer announcements, 2026-02.
-
AI-referred traffic to US retail sites was up 393% year over year in Q1 2026, and 693% year over year over the 2025 holiday season.
established Adobe Digital Insights, Quarterly AI Traffic Report, 2026.
-
In March 2026, AI-referred traffic converted 42% better than non-AI traffic on US retail sites, a reversal from converting 38% worse in March 2025.
established Adobe Digital Insights, Quarterly AI Traffic Report, 2026.
-
Agentic commerce could represent 10 to 20% of US e-commerce spend, 190 to 385 billion dollars, by 2030.
established Morgan Stanley, Agentic Commerce Market Impact Outlook, 2026.
-
Products missing a required GTIN are demoted to "limited performance" in Google Merchant Center. Google does not itself publish an exact impression or click uplift figure for fixing this.
established Google Merchant Center Help, GTIN [gtin] and Limited performance due to missing value: GTIN, support.google.com/merchants.
Reference
Glossary
- GTIN
- Global Trade Item Number, the standard product identifier (UPC, EAN, ISBN) Google and other engines use to match a listing to a known product. A missing GTIN is a documented cause of feed suppression.
- Product and Offer schema
- Structured data markup on a product page, following schema.org vocabulary, that states machine-readable price, availability, brand and review information for that product.
- Merchant Center feed
- The structured product data file a store submits to Google to make its catalog eligible for Shopping, Performance Max and, increasingly, AI shopping surfaces.
- Agentic commerce
- Purchases discovered, researched or completed with the help of an AI agent acting on a shopper's behalf, rather than the shopper browsing a site or search results directly.
Straight answers
Frequently asked questions
Why does my store not show up when I ask ChatGPT or an AI Overview about my product category?
Most often because the data those engines read, your Merchant Center feed, your on-page Product schema, your GTIN and review fields, is incomplete, stale, or disagrees with itself across the places it appears. The engine is not judging your homepage. It is reading structured data your homepage was never designed to expose.
Is this the same problem as regular SEO?
Related but distinct. Classic SEO optimizes page content for a ranked list of links. Being named inside an AI shopping answer depends on structured product data an engine can parse with confidence: identifiers, schema, feed accuracy and real reviews. A page can be well-written and still fail on the data layer underneath it.
Did OpenAI really say its own agentic-commerce launch struggled because of merchant product data?
Yes. OpenAI's own developer communications, corroborated by CNBC and Digital Commerce 360 reporting from March 2026, named inaccurate product data and onboarding difficulty as reasons it pulled back from in-chat purchase completion. It is first-party evidence from the company building the surface, not a third-party guess.
Can you guarantee my products will be cited in AI shopping answers?
No. AI-answer selection is undocumented and changes constantly. What can be engineered is the underlying data quality that gives you a real chance to qualify, measured over time as a share-of-answer rate, not asserted as a guarantee.
Provenance
Sources
- OpenAI, developer announcements on ChatGPT shopping and the Instant Checkout pullback, 2026-02 and 2026-03 (established)
- CNBC, "OpenAI revamps shopping experience in ChatGPT after struggling with Instant Checkout," 2026-03-24 (established)cnbc.com
- Digital Commerce 360, coverage of the OpenAI shopping pivot, 2026-03-24 (established)
- Adobe Digital Insights, Quarterly AI Traffic Report, January and June 2026 releases, business.adobe.com (established)
- Morgan Stanley, Agentic Commerce Market Impact Outlook, 2026 (established, forward-looking estimate)morganstanley.com
- Google Merchant Center Help, GTIN [gtin] and Limited performance due to missing value: GTIN, support.google.com/merchants (established, primary source, non-quantified)
Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.