E095 Instacart Test Design Research Note

E095 Instacart Test Design Research Note

Investigation record

Consumer Reports published the investigation on December 9, 2025 with Groundwork Collaborative and More Perfect Union. The article was updated on December 22 after Instacart ended item-level price testing.

The investigation recruited 437 volunteers. In September 2025, the volunteers were divided into four groups for coordinated online shopping sessions. Each group used the same retailer and store location, shopped at the same time, and added an identical basket of 18 to 20 items from Safeway or Target. A fifth Safeway and Target test was conducted in person in Washington, D.C. A later November test examined additional retailers.

Participants captured screenshots but did not buy the items. Consumer Reports says roughly 200 screenshot sets without execution errors supported its central analysis. That distinction matters. The recruited sample was 437, while the main analyzed screenshot set was smaller.

Observations

Consumer Reports reported that every analyzed shopper was assigned to an item-price experiment and that about three-quarters of checked products appeared at more than one price. The reported item differences ranged from seven cents to $2.56 and reached 23 percent in one example.

In the September 4 Seattle Safeway session, 39 volunteers built the same 20-item basket. The reported basket totals ranged from $114.34 to $123.93, an $9.59 or 8.4 percent spread. Only 8 percent of that group received the lowest total.

The investigation also documented different displayed reference prices for some discounted items even when the final sale price matched. That is evidence about the presentation of the promotion. It does not establish that every promotion on the platform used the same method.

The annual estimate

Consumer Reports translated average basket variation into an estimate of roughly $1,200 per year for a household of four using an Instacart household-spending assumption. The number is a modelled illustration, not an observed annual loss for each participant.

Any public use of the figure must identify the extrapolation. It should not be described as a bill that Consumer Reports measured over a year or as a typical loss established by a representative national sample.

Instacart response

Instacart agreed that the tests temporarily showed different item prices to different shoppers. It described the mechanism as randomized, short-term A/B testing by product category and store location.

The company said personal, demographic, and user-level behavioral information did not determine item prices. It rejected the dynamic-pricing and surveillance-pricing labels, disputed the $1,200 estimate, and said retailer partners controlled base prices. Instacart separately acknowledged that it used public Target prices plus an amount for operating and technology costs and had tested different approaches on the Target storefront.

These are company statements about the internal assignment method. Consumer Reports observed the outputs but did not publish an independent inspection of the complete code, assignment logs, or data pipeline.

Later change

On December 22, 2025, Instacart announced that it was ending all item price tests on its platform. It said retailers would no longer be able to use Eversight technology for that purpose. The company said shoppers using the same store location at the same time would see the same item prices.

Instacart continued to permit store-by-store differences, retailer markups, promotions, discounts, marketing incentives, and loyalty integrations. The current record therefore supports "item-level price testing ended," not "all differentiated offers ended."

In July 2026, Instacart published a current pricing-principles page repeating that item-price tests ended in December 2025 and stating that personal information would not be used to set item prices, membership prices, or fees. That is the company's current commitment and requires attribution.

Evidence boundary

The public record supports simultaneous price variation and a company-described randomized assignment mechanism. It does not show that age, disability, income, device type, purchase history, or food-desert status caused the observed item prices.

The New York attorney general demanded information in January 2026 and warned about possible noncompliance with the state's disclosure law. That was an inquiry and warning, not a final determination that personal data caused the tested prices.