Walmart has said it will work with external AI shopping agents, and it has attached conditions to that access. The move matters because it converts a technical question into a commercial one. Until now, agentic commerce was mostly a demo: an assistant reads a product page and suggests an item. Once a retailer publishes terms, the agent becomes a distribution channel with rules, and the rules decide who gets to represent the catalog and how.
The terms Walmart published
The published conditions cover how agents may read product data, how they may present prices and availability, and what they may not do with the retailer name. The substance is less dramatic than the headline. The retailer is not handing over checkout. It is defining the boundary between an assistant that helps a shopper find an item and an assistant that speaks for the store.
That boundary is the whole game. An agent that only retrieves data is a traffic source. An agent that summarizes and recommends is a merchandiser, and merchandising decisions carry liability for accuracy, price and stock. Retailers want the first and are cautious about the second, which is why the terms read like a supplier agreement rather than an API document.
Why a retailer would open its catalog
Because discovery is moving. A growing share of shoppers now start a product search inside an assistant rather than a search box, and a catalog that cannot be read by those assistants becomes invisible at the top of the funnel. Opening a catalog to agents is a distribution decision before it is a technology decision, and the cost of staying closed is paid in lost consideration rather than in lost transactions. Retailers weighing the trade can look at report for how much margin is already moving through refunds and tariff adjustments.
There is a second reason, which is data. When agents query a catalog, the retailer learns which attributes are asked about and which are missing. That feedback loop is valuable for assortment and content teams, and it is difficult to obtain from conventional search logs because assistants ask questions in full sentences.
What the retailer keeps
The retailer keeps price authority, inventory truth and the customer relationship. Those three are non-negotiable, and every published term tends to protect them. What the retailer gives up is the right to control the narrative around a product, which is precisely what a well built assistant takes over.
The difference between reading and quoting
An agent that reads a catalog returns a link. An agent that quotes a catalog returns a sentence, and that sentence becomes the shopper impression of the product. Retailers can control the first through access rules and the second only through content quality, because a quote is assembled from whatever fields exist. Thin fields produce confident sentences that the retailer never approved.
The three clauses that matter
The first clause is attribution. If an agent shows a price, it must show the source and the timestamp, because stale prices create disputes that the retailer absorbs. The second is scope, meaning which fields an agent may read and which must stay internal, such as margin or store level stock in specific locations.
The third is conduct, meaning what an agent may say about availability and substitution. A substitution suggested by an agent can create a promise the store cannot keep, and the customer will hold the store responsible. Retailers that write this clause carefully will avoid a category of complaint that is otherwise impossible to trace.
Why access rules read like contracts
Access rules are written by legal teams because the risk is asymmetric. A single wrong price shown to a large audience can cost more than a year of incremental traffic from the same channel. That asymmetry explains why the terms are long and the technical documentation is short, and why retailers with weak catalog governance will find the rules harder to satisfy than the integration itself.
What agents cannot see
Agents cannot see what is not written down. A catalog with thin descriptions and missing attributes produces confident but wrong recommendations, and the retailer is blamed for the outcome. This is the practical reason content quality has become an operations priority rather than a marketing one.
The gap shows up fastest in categories where the decision depends on context: size, compatibility, dietary constraints, installation requirements. An assistant that lacks those fields will default to popularity, which favors large brands and disadvantages everything else in the catalog.
Opening a catalog is therefore a decision to be judged on content readiness, not on agent partnerships. Retailers with structured attributes will gain reach. Retailers without them will gain a new channel for returns. The terms Walmart published are a reasonable template for the first group and a warning for the second.
