Artificial intelligence is playing an increasingly important role in how consumers shop for food. Today, it is mainly used to search for information, compare products, check prices and plan grocery purchases. As AI agents evolve, however, consumers could gradually begin delegating some purchasing decisions directly to these systems.
For brands and retailers, this raises an important question: how might loyalty change when some choices are delegated to an AI agent? According to FMI - The Food Industry Association, one concept to watch is "protected preferences", meaning choices that consumers decide to keep fixed rather than leave to the algorithm.
AI is already part of the grocery shopping journey
Consumers have already started using artificial intelligence when shopping for food. According to McKinsey's State of Grocery Retail Europe 2026, around 12% of consumers surveyed have used AI tools as part of their grocery shopping journey.
Among these early adopters, 40% use AI to create meal plans, 37% to research specific products and 36% to find the best prices or recipes that match their dietary preferences.
For now, AI is primarily being used for research, comparison and purchase planning. McKinsey notes, however, that AI agents could take on a broader role in the future, from preparing complete shopping carts to placing orders directly based on preferences set by the consumer.
From purchase frequency to "protected preferences"
If consumers delegate some purchasing decisions to an AI agent, the way loyalty is understood may also change. FMI - The Food Industry Association introduces the concept of "protected preferences" in this context. Consumers could retain control over selected parts of the shopping cart by setting permanent or particularly firm preferences.
They might, for example, instruct the agent to always buy a specific coffee brand, never substitute a particular yogurt, choose only certified organic products or always begin the search with a preferred retailer.
Repurchase frequency may therefore no longer be the only measure of loyalty. It could become equally important to understand which products, attributes or stores consumers choose to protect from alternatives identified by the algorithm.
When one product can shape the entire shopping cart
A protected preference can influence more than the sale of a single item. FMI notes that certain products can become a key factor in shaping the overall shopping cart, potentially influencing which retailer receives the rest of the order.
The implications differ depending on whether the preference concerns a manufacturer brand or a private label. If consumers protect a manufacturer brand, the agent may still choose among several retailers that carry it. If they protect a private label product, however, choosing that product also determines the retailer that sells it.
The same principle can apply to organic products, specialty foods, fresh items, exclusive preparations or other products that are difficult to substitute. Even an item representing only a small share of total spending could influence the rest of the basket and, in some cases, the choice of store.
Repurchase and loyalty are not always the same
One of the most important distinctions is between genuine preference and habitual behavior. Retailers and manufacturers often measure loyalty through purchase frequency and repeat purchases, but some of this behavior may simply reflect convenience, availability or habit.
An AI agent capable of quickly comparing alternatives, prices and stores could reduce the practical barriers to switching. Some habits may remain, while others could disappear once choosing an alternative no longer requires additional time or research from the consumer.
Frequent purchases alone may therefore become less informative. A more meaningful question could be whether consumers see a product as replaceable or want to keep choosing it even when the algorithm identifies other options.
Brands and retailers may have different objectives
Protected preferences also highlight a potential difference between the interests of manufacturers and retailers. For a manufacturer, the ideal outcome may be for consumers to instruct the agent to buy a specific product regardless of where it is sold. For a retailer, the priority may instead be to become the consumer's preferred shopping destination before individual products are selected.
The strongest combination occurs when both preferences align: consumers want a specific product from their preferred retailer. If neither choice is protected, the agent has greater freedom to optimize the purchase based on price, nutritional attributes, convenience or other priorities set by the user.
Promotions do not always indicate lasting preference
The same logic could change how promotions are evaluated. A consumer might, for example, allow the agent to buy a particular brand only when its price falls below a specified threshold.
In that case, a promotion that performs well in the short term does not necessarily mean the product has earned lasting preference. FMI therefore suggests looking beyond immediate campaign results and assessing whether customers continue buying the product without a promotion, whether the item becomes an explicit choice and whether the relationship with the brand or retailer strengthens over time.
Product visibility is extending to AI agents
This shift is also beginning to affect measurement tools. NIQ and Similarweb recently announced the development of new metrics for agentic commerce, designed to connect AI-generated recommendations with traffic, conversions and actual purchases.
One of the planned indicators is Agentic Shelf Visibility, which measures which products appear when an AI assistant presents shopping options and how their visibility compares with competing products. Another area concerns the quality and structure of product data, which AI systems need in order to correctly identify, understand and recommend an item.
For brands, retailers and distributors, this introduces another dimension of commercial visibility. Alongside the physical shelf, e-commerce search results and the digital shelf, companies may increasingly need to understand whether a product is included among the options considered by an AI agent and under what conditions.
A new perspective for brands and retailers
AI-agent-driven shopping is still at an early stage, and it is difficult to predict how quickly consumers will delegate more complex purchasing decisions. McKinsey's State of Grocery Retail Europe 2026 also points to several characteristics that make grocery shopping particularly complex, including high purchase frequency, highly varied baskets, large numbers of low-priced items and the challenges involved in delivering orders to the end customer.
Even at this early stage, however, the shift offers a new way to think about loyalty. For manufacturers and retailers, attention could gradually move from "how many times has the customer repurchased this product?" to "which choices are important enough that the customer does not want to delegate them to the algorithm?"
