13 of 25 Vendor Claims About AI Commerce Did Not Survive Fact-Checking
I adversarially verified the highest-stakes claims shaping how merchants think about AI-agent commerce. Just over half held up.
I ran 25 of the highest-stakes claims vendors are making about AI commerce through adversarial fact-checking. Thirteen did not survive. Every week another press release lands claiming a platform now has agentic checkout, an AI-readiness score, or a live integration with a major shopping agent. Merchants read these and make roadmap decisions on the strength of a paragraph they cannot verify. I wanted to know how much of it was real.
highest stakes for merchants
majority-refutation
survived adversarial review
What I did
I ran a multi-agent research pass across 23 sources, 7 of them primary (vendor documentation, official filings, first-party announcements, not secondary writeups). That pass surfaced the claims currently shaping how merchants think about AI-agent commerce: what is live, what is standard, what is coming.
From that set I picked the 25 claims with the highest stakes for a merchant's actual decisions, the kind that would change what you build or who you sign with. Then I tried to kill each one. Several independent skeptic passes per claim, each looking for the primary source, the fine print, the gap between the press release and the shipped product. A claim survived only if it held up against a majority of those checks. One convincing rebuttal was enough to flag it; a majority of rebuttals killed it outright.
The rounding-up pattern
The pattern in the killed claims is consistent: a real vendor initiative gets described, in the secondary coverage or in the AI-generated summary a merchant reads, as more finished and more universal than the primary source supports.
Six representative failures
- PIM vendors and MCP. Salsify or Akeneo do not ship MCP servers for third-party agents as claimed.
- The free audit that isn't. Shopify does not offer a free agent-readiness audit tool as described.
- "All stores" claims. Shopify Catalog does not auto-provide standardized schema and cross-AI syndication for every store on the platform.
- Google Merchant Center's phantom features. "AI performance insights" and conversational attributes are not live in the form described.
- Microsoft and UCP. Microsoft has not adopted Google's Universal Commerce Protocol as its transaction layer.
- The universal cart. Google's universal cart, live with named retailers, is not confirmed as described.
What's actually happening
- A real initiative exists behind nearly every killed claim.
- Each gets rounded up a notch per retelling: journalist, then newsletter, then AI summary.
- "Piloting X with one partner" becomes "X is live everywhere" by the third or fourth hop.
- None of these are vendor fabrications, in most cases.
Confirmed against the same scrutiny
The confirmed claims are worth taking seriously precisely because they survived the same scrutiny that killed the other thirteen.
Feedonomics launched Agentic Catalog Exports for enterprise customers. Productsup ships an AI Enrich feature. SalsifyIQ shipped an AEO module, enterprise-priced from roughly 40,000 to 200,000 USD a year, with long implementation timelines. Shopify co-developed UCP with Google, and its Agentic Plan lets brands on any platform list in Shopify Catalog. Microsoft Copilot Checkout is a live US pilot with Shopify auto-enrolled. And there is one direct SMB competitor in this exact space, with roughly 13 reviews in ten months.
That last point matters as much as the enterprise moves. The category is real, the incumbents are moving, and there is startup activity at the SMB end too, just not much of it yet. This is an early market, not a crowded one, and not a fictional one either.
Distrust the summary, ask for the primary source
If you are a merchant trying to decide whether to invest time in agent-readiness now or wait, the honest read of this research is: the underlying shift is real, but a lot of what you will read about it is not what it seems.
The practical move is to distrust the summary and ask for the primary source. When a vendor says "AI agents can now buy from your store," ask which agent, in what pilot, with how many retailers, since when. When a platform claims universal support, ask what "universal" actually covers. Half the confirmed claims above still come with real limits: enterprise-only pricing, single-market pilots, one platform's ecosystem. Those limits are the useful part of the claim, and they are exactly what gets dropped in the retelling.
I publish this kind of check because I build in this category myself, and I would rather my own roadmap be wrong for a week than wrong for a quarter because I trusted a headline. Pollen exists on the same premise this research does: agents can only act on what they can actually verify, and neither agents nor merchants are well served by claims that don't survive a second look.
Twelve claims survived. Thirteen did not. Ask for the primary source before you build your roadmap on either count.