A December 2025 investigation by Groundwork Collaborative, Consumer Reports and More Perfect Union found that identical items from the same stores could appear at different prices on Instacart at the same time. In the study’s usable sample, 74% of tested items appeared at multiple prices, with some differences reaching 23%; identical baskets varied by about 7% on average.
The researchers analyzed 193 complete and valid data sets submitted by 437 participants. They compared displayed offers rather than completing purchases, so the findings establish observed price differences but do not by themselves identify why each difference occurred.
Instacart said on December 22, 2025 that it was ending price tests on its platform. The company said those tests did not use personal data, demographics or an individual’s shopping behavior. That leaves an important distinction: measured price variation and acknowledged testing do not prove that a shopper was charged more because an algorithm inferred how much they wanted an item.
Price changes are not all personalized
Online prices have long shifted with stock, demand, competition and timing. AI systems can add far more signals to that process, but a market-driven price change, a randomized experiment and a price tailored to a shopper’s profile are separate practices.
| Mechanism | What changes | Role of the shopper profile |
|---|---|---|
| Dynamic pricing | The price responds to demand, inventory or competition | May not matter at all |
| Personalized pricing | The offer differs based on customer characteristics or behavior | Influences the offer |
| Price testing | Groups of users receive different prices | Assignment may be random |
| Personalized discount | One customer receives a coupon while another does not | May determine who gets the benefit |
| Personalized recommendation | Products and their display order change | Influences what the shopper gets to compare |
The Bank of England has drawn a similar line between adjustments based on market conditions and offers based on a customer profile. Its April 2026 analysis said these practices had not pushed inflation systematically in one direction: personalization can lower prices for some customers while producing more expensive offers for others, depending in part on competition and the information available to buyers.
The US Federal Trade Commission’s surveillance-pricing study describes tools that can potentially use signals including location, browsing and purchase histories, abandoned cart items and mouse movements. The public research materials are anonymized and include hypothetical examples, documenting capabilities rather than proving that every scenario is in use at a particular retailer.
A personalized offer need not appear as a visibly different product-page price. The FTC report also describes a system that could withhold promotions from customers predicted to buy anyway while extending discounts to shoppers judged likely to abandon a purchase. Two shoppers could therefore see the same listed price while paying different totals after one receives a targeted coupon.
AI shopping assistants add new data signals
AI shopping agents could help consumers by comparing stores, delivery charges and available discounts before placing an order. But a retailer-controlled assistant may also receive unusually direct information: a shopper’s budget, preferred model and urgency. Those details can make recommendations more relevant, while also creating a potential pricing signal if reused for commercial targeting. There is no evidence in the material reviewed here that every retail chatbot uses such information to set an individual price.
Some companies have publicly set boundaries. In a letter published September 25, 2026, Walmart said income, purchase history, urgency and estimated willingness to pay would not influence price, including when using its Sparky assistant. The commitment is a company statement rather than an independent audit.
McDonald’s has also said its pricing tool makes recommendations at the restaurant level and that franchisees set prices. The company denies using an individual customer’s willingness to pay when determining them.
The key question for AI-assisted commerce is not simply whether a retailer uses data. It is whether that data changes the price, access to a discount or the set of offers presented to a particular buyer. Establishing that requires controlled comparisons of the same item, seller, delivery terms and benefits, while separately tracking product prices, coupons and the final checkout total.
