The price only you can see

$5.69. $6.89. Two company-run McDonald's in Fresno, two miles apart, selling the same Big Mac. Reuters found the gap on the McDonald's app in a story about the machine-learning engine that now recommends "the optimal price" for every item at nearly 14,000 US restaurants. Among its inputs is an estimate of what the franchisee dashboard calls "customer willingness to pay in your area." Reuters couldn't confirm that the engine set that particular spread.

The same day, Lindsay Owens's book on surveillance pricing got a long review in the Prospect and a Life Kit segment on NPR. The EFF also wrote up DraftKings training a model on its customers' betting records to find the ones most likely to lose and lure them back with targeted promotions.

Most of the argument about AI pricing is being had over discrimination: may a seller charge different people different prices? That's the wrong line. Sellers have always done it, and a lot of it is fine. The line that matters is whether the price is public, meaning whether anyone other than the buyer can see it.

A price anyone can look up is information. A price only you can see is a negotiation you don't know you're in.

Hayek's old point was that the price system works like a telecommunications network. One number goes out to everyone and carries knowledge no single buyer or seller has. The Fresno Big Mac is still a price in that sense. It's on the menu, Reuters compared two of them without a subpoena, and anyone with a car can beat the higher one. What Owens documents, in the Prospect's summary, is a different animal:

That discount exists for one person. Nobody else sees it, so nobody else can compare it, complain about it or undercut it. Per the same review, Starbucks trimmed a Washington Post reporter's rewards the more he spent. A study by Owens's Groundwork Collaborative, Consumer Reports and More Perfect Union found that about 75% of the items in identical Instacart baskets, bought at the same time, were priced differently from one shopper to the next. DraftKings is the limit case: an offer computed for, and visible only to, the customer least able to refuse it.

Read NPR's advice for fighting back and you can see what's been lost. Check the price on several devices, logged in and logged out, with and without incognito mode, with and without a VPN, and ask a friend to look it up on their phone. That's shoppers rebuilding the posted price by hand, a distributed audit run by the side without the model. One side analyzes millions of transactions a day. The other has a private browsing window.

The economists' defense deserves a real hearing. Price discrimination can expand output: the student fare, the senior coffee and the matinee let in price-sensitive buyers who would otherwise be priced out. Per Reuters, McDonald's engine has lately pushed some prices down, because headquarters takes a cut of revenue and wants volume. Grant all of it. Then notice that every example in that defense is a posted price for a posted group. The senior discount is on the menu board. The student fare is on the website with its rules attached. You can see what the other group pays and check whether you qualify. The case for price discrimination was always built on prices that stayed public.

That points to a narrower rule than the ones on the table. Maryland's law, taking effect Thursday, bans personalized and dynamic pricing at grocery stores and delivery apps. Owens wants the price tag back. Both reach for prohibition. The better test is publicity: charge whatever you like, to whichever group you like, as long as anyone can see what anyone pays. The Fresno store passes that test. The payday discount doesn't.

A price you can drive two miles to beat is still a price. One only you can see is just the seller's estimate of you.