Same groceries, five prices, one checkout experiment

Farming is older than the wheel by several thousand years, and for most of that run, price was its information technology. The futures page tells the grower what to plant; the shelf tag tells the shopper what to buy; the whole system talks to itself in numbers. So when a grocery app starts showing different shoppers different numbers for the same eggs in the same minute, I pay attention. Last September, 437 volunteers ran five live tests on Instacart, filling identical baskets at the same stores, simultaneously. Here's what they found, ranked by how far upstream the trouble travels.

The checkout became a lab you never signed up for

Instacart's pricing engine arrived with a 2022 acquisition: Eversight, an AI-pricing firm whose marketing stated plainly that "End shoppers are not aware that they're in an experiment." That was the design brief, out loud. Across the tests, about 74% of items rang up at more than one price. Running an experiment is fine — publishing the protocol is what makes it science.

One cart, two totals, same minute

In Seattle, identical Safeway baskets totaled $114.34 for some shoppers and $123.93 for others — about a 7% spread, which compounds to roughly $1,200 a year for a family of four by Instacart's own spending estimate. Anyone who keeps a compost pile knows small daily deviations are where systems leak. This one leaks toward somebody's margin, and it isn't the shopper's.

Five prices for one carton of eggs

At a Washington, DC Safeway, Lucerne eggs appeared at $3.99, $4.28, $4.59, $4.69 and $4.79 — one store, one moment. In grain markets there's a word for local price divergence: basis, and it's usually explainable by freight or storage. Here the maximum spread hit 23% on peanut butter, crackers and deli turkey. Nobody's freight costs vary five ways at once.

The discount was a variable too

Shoppers saw the same $4.99 sale price sitting atop different strike-through "original" prices — Heinz ketchup from $5.29 to $5.59 — so the identical deal looked deeper depending on who was looking. A receipt is a user interface, and this interface had been quietly edited. Design decisions are pricing decisions.

The documented harm is opacity, not targeting — say it precisely

The study found no meaningful price differences by age, race, gender or income, and Instacart says its tests used no personal data. The proven problem is that shoppers couldn't see the experiment. To be careful: this is a transparency failure with receipts, not proven profiling, and the authors themselves caution their volunteers weren't a representative sample. Both things can be true.

Regulators moved at record speed, for regulators

Senators including Amy Klobuchar pressed the FTC to investigate; on December 18 the agency announced a $60 million settlement over separate alleged deceptive practices — masked service fees, a "guarantee" that paid credits instead of refunds, auto-enrolled trials. Days later, Instacart ended its item price tests, though partners can still test promotions, and New York's attorney general is now probing whether its disclosures satisfy a new algorithmic-pricing notice law (NY AG).

What it means upstream, where food actually comes from

Price is the instruction set agriculture reads. Farmers already sell into thin margins on global exchanges; if the last inch of the chain — the checkout — gets randomized per shopper, the demand signal gets noisy for everyone. My opinion, plainly labeled: when a middleman finds margin by testing shoppers rather than improving the food itself, the incentive to grow anything better quietly erodes.

We composters have a rule: if the pile surprises you, you weren't watching it. Food systems are honest exactly as far as they're traceable. A checkout running hidden experiments on the person holding the carrots fails that test. The fix is old technology — one tag, one price, printed where everyone can read it.