Walmart will allow third party AI shopping agents to interact with its catalog under new terms, a decision that puts the retailer on the opposite side of a debate most platforms have tried to avoid. Some marketplaces block outside agents to keep traffic and data inside their own interfaces, while others open their catalogs and accept that a growing share of purchase decisions will start somewhere else. The terms Walmart published are the interesting part, because they define what an agent may do on its behalf.
Two routes to AI mediated shopping
The closed route treats the marketplace as the interface. The platform builds its own assistant, controls the recommendation logic, and captures the data generated by every conversation. The open route treats the catalog as an input that any agent can query, accepting disintermediation in exchange for staying in the consideration set when a shopper delegates the search. Walmart will let third party AI shopping agents interact with its catalog under new terms, and the choice signals which risk the company considers larger.
The trade off is straightforward. A closed assistant protects margin per session and customer data, but risks being skipped entirely if shoppers prefer a general purpose agent that compares across retailers. An open catalog gives up some control of the conversation, but keeps the product data and the checkout relationship, which are the parts of the transaction that actually generate revenue.
What the new terms allow
The published terms focus on behavior rather than technology. Agents must identify themselves, respect rate limits, use accurate product data and avoid misrepresenting prices or availability. Those conditions matter because an agent that quotes a stale price creates a customer expectation the retailer cannot honor at checkout, which converts a discovery win into a service failure. Guardrails are cheaper than disputes after the fact.
Guardrails written into the terms
The most consequential clauses are usually the least discussed: how often an agent may query, whether it can cache product data, and what happens when its information conflicts with the live catalog. Caching rules determine whether the retailer can change a price without an agent quoting the old one for days. Query limits determine whether a popular agent can degrade site performance during peak hours.
Marketplace sellers feel it first
Third party sellers on the platform will experience the change before anyone else. If an agent surfaces products based on structured attributes, sellers with incomplete listings become invisible regardless of price. That creates an incentive to invest in product data quality, which is exactly what brands spreading inventory across more marketplaces ahead of the peak season will need to manage, since each marketplace applies different attribute standards.
The data trade behind the door
Opening a catalog to agents changes the direction of data flow. In a closed interface, the retailer observes the full journey from query to checkout. In an open one, the query happens elsewhere and the retailer may only see the resulting product view. That loss of context is the real cost of the open route, and it is why some platforms demand that agents pass through a sanctioned interface rather than scraping pages.
What stays defensible
For retailers, the compensating move is to make the first party relationship stronger than the agent relationship. Loyalty pricing, saved preferences and reliable delivery windows are things an external agent cannot replicate, and they keep the customer attached to the retailer even when discovery happens elsewhere. The strategic question is not whether agents will mediate shopping, but which assets remain defensible once they do.
How brands should respond
The first response is data hygiene. Agents select on structured attributes, so titles, dimensions, materials and compatibility fields must be complete and consistent across every marketplace. The second is price integrity, since an agent that finds a mismatch between a cached price and the live one will route the shopper to a competitor. The third is monitoring, because agent traffic looks different from human traffic and can distort conversion metrics if it is not separated.
The fourth response is content strategy. If a general purpose agent summarizes reviews and specifications, brands need their product information to be quotable and internally consistent across pages. Walmart is extending its marketplace and advertising flywheel into more markets, which means the same catalog will be queried by more agents in more languages, and inconsistencies that were harmless in a single market become conversion losses at scale.
What to watch next quarter
Three signals will show whether the open route works. The first is whether agent referred traffic converts at a similar rate to organic traffic, which indicates the catalog data is accurate enough to survive delegation. The second is whether advertising revenue holds up when the conversation moves off platform, since sponsored placement inside a chat interface is a different product. The third is whether other large retailers follow or explicitly refuse.
The longer term question is who owns the customer relationship when the interface is a conversation. Retailers that treat agents as another traffic source, with the same discipline applied to feeds and search, will adapt fastest. Those that treat agents as a threat to be blocked may protect short term metrics while losing the discovery layer where an increasing share of purchase decisions now begins.
