Retailers spending time and money to get their products recommended by ChatGPT may be solving the wrong problem first. Showing up in an AI-generated answer doesn't mean a customer can actually buy anything — that requires plugging into one of several competing, still-immature checkout systems built for AI agents.
What happened
Over the past year, a handful of major players have each rolled out their own version of an "agentic commerce" protocol — the plumbing that lets an AI chatbot or shopping assistant complete a purchase on a customer's behalf instead of just describing a product. OpenAI has been building checkout capability directly into ChatGPT, working with a payments partner to let users buy items without leaving the chat window. Google has its own agent payments framework tied to its search and Gemini products. Visa and Mastercard have each introduced protocols meant to verify that an AI agent making a purchase is authorized to do so, and PayPal and Shopify have layered their own agentic tools on top of existing merchant infrastructure.
The problem for retailers is that these systems don't talk to each other. A product feed formatted for one platform's shopping assistant may not meet the requirements of another. Tax calculation, shipping rules, inventory syncing, and — critically — fraud liability all work differently depending on which protocol handles the transaction. A sale that goes through smoothly on one platform can stall or get flagged on another because the underlying checkout logic wasn't built to match.
This matters because the industry is still deciding who eats the cost when something goes wrong. If an AI agent completes a purchase using stored payment credentials and the transaction turns out to be fraudulent, current card network rules and each platform's own terms determine who's liable — the retailer, the payment processor, or the platform running the AI agent. Those rules are not yet standardized across the major protocols, which means a retailer connected to multiple systems is effectively agreeing to multiple different sets of liability terms at once.
Why it matters
This follows a familiar pattern in tech infrastructure: several large companies race to define a new category before a dominant standard emerges, and merchants are asked to integrate with all of them just to hedge their bets. It happened with mobile wallets, with early web payment gateways, and with messaging standards. History suggests consolidation eventually happens — but usually after a period where smaller businesses absorb the cost and complexity of supporting multiple incompatible systems.
What this means for small businesses
For a small retailer, the temptation is to treat every new AI shopping feature as another marketing channel to plug into quickly. The checkout layer is where that logic breaks down. Connecting product feeds to ChatGPT, Google, or any other AI shopping surface without first confirming how a completed sale will actually be processed, taxed, and protected against fraud can create reconciliation headaches that outweigh any traffic gained.
Before connecting to a new agentic commerce protocol, it's worth running three checks: whether the protocol's fraud liability terms are clearly spelled out and acceptable, whether your existing point-of-sale or e-commerce platform can sync inventory and pricing accurately with that specific system, and whether a completed AI-driven sale actually shows up in your normal order management and accounting workflow without manual cleanup. Skipping these checks in favor of speed is how retailers end up with orphaned orders, tax mismatches, or disputed charges months later.
What to watch
Watch for whether card networks or major platforms move toward a shared verification standard for AI-initiated purchases, and whether any single protocol starts pulling ahead in retailer adoption numbers. Also worth tracking: early reports of chargeback or fraud disputes tied specifically to agent-completed purchases, which will signal how the liability question is actually being resolved in practice rather than in policy documents.
The bottom line
Being visible in an AI shopping assistant's answer is only half the transaction — small businesses should confirm how checkout, tax, and fraud liability actually work on a given protocol before connecting their product catalog to it.