Algorithmic Grocery Pricing: What Shoppers Can See

Algorithmic Grocery Pricing: What Shoppers Can and Cannot See

Algorithmic grocery pricing is not one practice. A price test assigns controlled variants to learn how a group responds. Dynamic pricing changes with conditions such as time, supply, or demand. Personalized pricing changes an offer for an individual or segment. Surveillance pricing is the FTC's term for using personal data to target individualized prices or offers.

Seeing two shoppers receive different prices proves variation. It does not automatically reveal the rule, data, or intent behind that variation. To evaluate a system, a shopper or regulator needs the product, time, store, channel, comparison group, assignment method, data inputs, duration, and resulting difference.

Four terms describe different mechanisms

Loose language makes a pricing dispute harder to resolve. The important distinction is not whether software was involved. Most modern retail systems use software. The distinction is what changed, for whom, why, and with which data.

These categories can overlap. A retailer could test personalized discounts, or a dynamic system could use location as a market condition without using a named person's history. The label should follow the demonstrated mechanism, not the emotional force of the claim.

flowchart TD
    A["Different prices observed"] --> B{"Was assignment randomized?"}
    B -->|Yes| C["Price test"]
    B -->|Unknown or no| D{"Did price follow time, supply, or demand?"}
    D -->|Yes| E["Dynamic pricing"]
    D -->|Unknown or no| F{"Did individual or segment data affect the offer?"}
    F -->|Yes| G["Personalized pricing"]
    G --> H{"Did personal data drive the individualized result?"}
    H -->|Yes| I["Surveillance pricing in the FTC sense"]
    H -->|Unproven| J["Personalization mechanism remains uncertain"]

The diagram is a research path, not a legal test. Applicable statutes can define terms differently.

The 2025 investigation established simultaneous variation

Consumer Reports, working with Groundwork Collaborative and More Perfect Union, organized a grocery-shopping investigation in 2025. Its published account says 437 volunteers were divided across four city groups and asked to add the same 18 to 20 items from Safeway or Target to Instacart carts at coordinated times.

Participants submitted screenshots. Consumer Reports says roughly 200 error-free screenshot sets were used for the central analysis, with additional retailers examined later. Around three-quarters of checked items appeared at multiple prices. Reported item-level differences ranged from seven cents to $2.56 and reached 23 percent in the examples. One analyzed basket ranged from $114.34 to $123.93.

That design gives the report a strong answer to one question: shoppers could see different prices for the same item from the same store during coordinated sessions. It does not directly observe the internal assignment logic.

The widely repeated $1,200 annual figure came from extending an observed basket difference across assumptions about household grocery spending and repeated exposure. It was an estimate, not an annual bill collected from a representative family. It is reasonable to report the calculation as an illustration of possible cumulative impact. It is not reasonable to state that every family paid $1,200 more.

Instacart described the mechanism as randomized testing

Instacart's December 2025 response agrees that different users were temporarily shown different item prices. It describes the practice as short-term randomized A/B testing conducted by retail partners to understand price sensitivity at a macro level.

The company says participants were randomly assigned by product category and store location. It says personal, demographic, and user-level behavioral information was not used to set item prices. It also says retail partners control base prices and that some retailers apply online markups.

Those statements are the counterparty's account, not an independent audit of the code or data. They matter because they propose a specific mechanism that can be tested. If assignment was random within category and store location, the practice was a price test. If individual income, shopping history, or predicted willingness to pay affected assignment, the practice would move into personalized or surveillance pricing.

The investigation observed the output. Instacart described the input and rule. Public evidence did not include a complete independent inspection of the assignment system. The correct conclusion preserves both facts.

Surveillance pricing requires evidence about personal data

The FTC's January 2025 surveillance-pricing study found that pricing intermediaries can use precise location, browsing behavior, demographics, shopping history, and other consumer information to target different prices or offers. The agency reviewed documents from eight intermediary companies and described hypothetical examples based on aggregated findings.

That work establishes that the capability exists and is offered in the market. It does not establish that every algorithmic price difference uses personal data. It also does not prove that the Instacart experiment used the inputs described in the broader FTC study.

This distinction matters for consumer protection. If any unexplained variation is called surveillance pricing, a company can refute the label by showing a randomized test while leaving the fairness of that test unanswered. Precise terms allow stronger questions. Was the control price visible? How long did the test last? Could a shopper opt out? Were essential goods included? Did higher and lower variants balance? Were users told they were in an experiment? Which party controlled the rule?

The legal record was still developing

New York's Algorithmic Pricing Disclosure Act took effect on November 10, 2025. The New York attorney general's January 2026 Instacart announcement says the law requires a prominent disclosure when personal data is used to set individualized prices.

The attorney general demanded information about Instacart's price experiments and warned that its disclosures might not comply. A demand for information is not a final finding of violation. It establishes a live regulatory question and the evidence the office wanted to inspect.

The FTC announced a separate Instacart settlement in December 2025. The agency alleged deceptive practices involving “free delivery” advertising, satisfaction guarantees, refunds, and subscription enrollment. The proposed order and refunds addressed those allegations. They did not adjudicate the Consumer Reports price-testing methodology or prove that personal data set item prices.

This separation should remain visible in any article, search snippet, or AI-generated summary. Enforcement against the same company is not automatically enforcement for the same conduct.

Instacart ended item price tests, but the broader question remains

Consumer Reports later reported that Instacart stopped item price testing after the investigation. The update says the platform could still permit partners to test promotions and discounts.

Ending a practice changes the current consumer risk but does not erase the governance question. Retail experiments can affect people who never consented to become research participants. A lower variant can benefit one shopper while a higher variant makes another person fund the experiment. Essential goods sharpen the concern because delaying or abandoning the purchase may not be realistic.

Retailers and platforms can make these systems more inspectable by publishing the purpose, responsible party, eligible products, assignment method, data exclusions, control price, variant range, duration, consumer notice, opt-out path, monitoring, and post-test decision. A regulator or independent auditor needs access to the underlying logs. A shopper needs a shorter record close to the price.

What a shopper can check today

A shopper cannot reverse engineer a pricing model from one cart. A more useful check compares the same exact item, store location, account state, fulfillment channel, time window, membership, and coupon conditions. Screenshots should include the product size, seller, price, fees, and timestamp.

Even a careful comparison proves only what it records. Different results may come from a test, promotion, inventory update, store markup, account eligibility, location, or an error. The next step is to ask the platform which rule explains the difference and retain the response.

[[How to Compare Grocery Prices Across Stores]] provides a method for comparing the household decision rather than investigating the platform. [[E095 Content Plan|Episode 95]] preserves the price-testing investigation as its own evidence package, while [[E096 Content Plan|Episode 96]] focuses on the broader use of consumer data and willingness-to-pay models.

The practical standard is simple: name the mechanism only as precisely as the evidence permits. “Different prices were observed” is a meaningful finding. “Personal data caused the difference” is a separate claim that requires separate proof.

Sources and editorial notes

This explainer uses the Consumer Reports investigation and follow-up, Instacart's published response, the FTC surveillance-pricing study, the New York attorney general's January 2026 inquiry, and the FTC's separate Instacart settlement announcement.

It provides consumer education, not legal advice. The legal status, investigation record, company policies, and terminology require refresh before publication. AI assisted with research organization and drafting under editorial review. Publication remains unauthorized.