Start with the number, because it is the whole argument. An audit of product pages from leading brands found 70% currently don’t meet Google’s Universal Commerce Protocol for AI selling. That is not a long-tail problem or a small-merchant problem. That is the top of the market, functionally invisible at the exact moment the buyer changed.
Here is the shift most boards have not internalized. For two decades we optimized product data for a human who browses. Marketing copy, hero imagery, emotional hooks, keywordy titles built to win a search box. That human is now being intermediated at the point of purchase by an agent, and the agent does not read like a person. In traditional ecommerce, poor product data leads to lower conversion. In agentic commerce, poor product data means the product is never selected at all. An AI agent doesn’t browse. It decides. Based on structured inputs. If data is incomplete, contradictory, or outdated, the agent simply moves on.
No notification. No abandoned cart to retarget. No second chance. The product page is no longer the interface. The data is.
Your PIM is not the thing being graded
This is where senior teams comfort themselves with the wrong evidence. Somebody says, we invested in a PIM, our data is clean. Clean for whom?
The distinction that matters: internal PIM workflows and external agentic readiness are different problems. Structured catalog quality at the protocol layer is what AI agents evaluate. A retailer with a sophisticated PIM and a poor protocol-compliant feed gets skipped. A retailer with a basic PIM and a clean feed gets recommended.
Read that as an operator. Your internal tidiness does not earn revenue. Protocol-legible output does. The agent evaluates every interaction with a merchant’s systems as a test of data quality, response speed, and resolution capability. It evaluates your catalog data, your checkout process, and your post-purchase support to decide whether to recommend your store.
And what wins is unglamorous. Not adjectives. Attributes. Vague descriptions, missing variants, and stale inventory mean your products never surface in agent-driven discovery. Precise attributes like material composition, exact dimensions, and specific use cases consistently outperform generic marketing copy. The copywriting that won the human loses to the spec sheet that wins the machine.
This is a revenue line, not an SEO ticket
Stop filing structured data under technical debt. The signal is already in the numbers. Pages with structured data are cited 3.1x more frequently in Google AI Overviews. On the demand side, Shopify reports that orders from AI-powered searches grew 15x year-over-year through 2025. And the volume behind it is not speculative: by 2030, nearly 50% of online shoppers are expected to use AI agents, accounting for roughly 25% of their spending, adding $115B to the US ecommerce sector.
Now apply the Cost of Doing Nothing. The CODN here is not a slide. It is the discovery you forfeit every quarter your catalog stays human-only while a competitor with worse products but cleaner data gets the recommendation. If your store isn’t set up for these agents to find and understand your products, they’ll recommend your competitors instead. Not because your products are worse, but because the agent couldn’t read your data. That is margin leaking to a rival’s data hygiene, and it compounds.
The market structure is still unsettled, which is exactly why this is a build window and not a wait window. OpenAI deprecated its Instant Checkout program in March 2026 after fewer than 30 Shopify merchants had gone live, and pivoted to discovery plus a redirect into the merchant’s own checkout. Other protocols are still live and being tested, so the market is unsettled rather than fixed. For now, agents mostly drive discovery while merchants keep the transaction.
Good. That means you keep the customer relationship and the loyalty data. What you do not keep, if your data is unreadable, is the shortlist.
What survives the business case
Fund this: a protocol-compliant feed measured on agent citation rate, complete canonical attributes on your top revenue categories first, real-time price and inventory served through APIs rather than rendered page text, and a single owner with a P&L stake in agent legibility. Stop funding: another round of hero-image A/B tests for pages an agent will never look at.
The buyer at your point of sale is now a machine reading structured inputs. You can keep writing beautiful copy for a shopper who has already left the room, or you can write for the reader who actually decides. The retailers who treat product data as revenue infrastructure will be on the shortlist. Everyone else will be a rounding error in someone else’s recommendation.
The shopper changed. Change the catalog before the quarter does it for you.