Two news stories that broke just a few days apart—one legal, the other product-related—actually convey the same message. As of August 2, the European AI Act requires all conversational agents to identify themselves as such. In late July, Kroger rolled out a shopping assistant across all its stores that doesn’t search for items in a catalog but starts with the meal. This is no coincidence in timing. It confirms that the AI shopping assistant is becoming a key piece of infrastructure—with its own obligations and design standards—and that the best way to address these is to keep it in-house.
Transparency is becoming a product requirement, not just a way to build trust
Article 50 of the EU AI Regulation took effect on August 2, 2026. Specifically, any provider of a chatbot, conversational agent, or virtual avatar must now clearly inform the user that they are interacting with an AI, unless this is already obvious. This deadline was confirmed by the “Digital Omnibus on AI,” published in the Official Journal of the EU in late July. A grace period has been granted until December 2, 2026, solely for the technical labeling requirement for content already on the market; however, the obligation to inform users applies immediately. In France, the DGCCRF, CNIL, and Arcom can impose penalties for noncompliance, with fines of up to 15 million euros or 3% of global revenue (Blog du Modérateur).
For a retailer considering offering a shopping assistant to its customers, this change in status matters. Until now, it was just one trust-building factor among many: saying that the AI is indeed AI is reassuring, but there was no real obligation to do so. It is now a compliance requirement—documented, enforceable, and subject to penalties. And compliance is structurally easier to maintain with an agent designed and operated for the retailer—one that’s integrated into its customer experience—than with a third-party, general-purpose agent over which the retailer has no control regarding behavior or update schedules.
Kroger proves that a responsible agent starts with the meal, not the list
That same month, Kroger rolled out an AI shopping assistant across all of its brands’ websites and apps—a feature that stands in contrast to traditional product search. The tool helps users plan a week’s worth of meals, browse recipes, and put together a shopping cart that fits a budget or dietary restrictions. It can also take a photo of a handwritten list, a recipe card, or a link and turn it into a shopping cart in just a few seconds (Progressive Grocer, The Shelby Report).
This is a strong signal—coming from a player of this size—about how a grocery shopping assistant should be designed. A large portion of the shopping cart is filled out of habit, with favorite products or items that have already been ordered, and that’s perfectly fine. But that’s precisely the area where current search engines already excel: finding a known product. The real added value of an agent lies elsewhere—in the portion of the shopping cart driven by a craving or a recipe, which simple product search currently serves poorly. An agent that can only find products in a catalog leaves this potential untapped. Kroger is building its assistant on Google Cloud’s AI infrastructure, but presents and operates it as a Kroger tool, integrated into its customer journey, its brand, and its product lineup. The technology may come from outside; the experience, customer data, and relationship remain with the retailer.
The common thread: the winning agent is a full-time employee
These two developments point to the same conclusion for mass-market retailers: A relevant shopping assistant must both meet a legal requirement for transparency and be driven by the customer’s desires rather than a shopping list. Both of these goals are much easier to achieve with a white-label-specialized agent—designed for the retailer and integrated into its customer experience—than with a third-party generalist agent that intervenes in the shopping journey without the retailer having control over either the messaging or the data. Compliance is managed from within; culinary relevance is built using the retailer’s own product and recipe data.
The challenge for the coming months, therefore, is no longer whether an AI shopping assistant has a place in the customer experience, but rather in what form. An assistant that clearly identifies itself as AI, that starts with the meal rather than the aisle, and that remains within the retailer’s ecosystem: this three-pronged approach is beginning to emerge as the standard.
