Two accounts of the same holiday. The singer reports bookings at $1,800 against a plan of $2,000, and the traveler’s statement shows $2,200. Both accounts are correct, and the difference between them is a choice about where to place zero.
A comparison needs a baseline. The original plan supplies one before the purchase; a report can substitute another after the bill arrives. An assistant that promises to save money needs to say which comparison it will use while the buyer can question the choice.
The moving baseline
A claimed improvement can move in several directions while someone measures it. The list price can stand in for the price the buyer would pay, so a discount against a number few people meet looks larger than it is. A purchase from last year can be promoted into a preference, so the recommendation arrives with the authority of the buyer’s own history. A maximum budget can drift into a target, so an agent that spends the whole budget reports diligence. Each movement is small. Together they let a report settle wherever its author prefers.
Show the baseline with its date, and include delivery and fees in the total. Where a materially equivalent cheaper option exists, put it beside the upgrade; where none does, say so. State what the extra money buys and how it connects to the task. The buyer can then choose the distillery with the added expense visible and keep a record to consult when the statement arrives and memory starts to negotiate.
What entered the field of view
The recommendation also depends on what the system saw. Researchers at Wharton’s Generative AI Labs ran roughly 26,000 shopping tasks through a simulated store, varying the context an agent received: a screenshot of reviews, a competing source, the order in which information arrived, injected memory of the user. Their technical report finds that choices grew less predictable as context accumulated. The same agent, given different framing, picked differently.
The agents chose a fitness watch from a product grid, and no money moved. Changing the order of sources affected some models more than others. The study shows sensitivity and says nothing about real shoppers’ savings. A buyer needs the sources behind a recommendation before treating it as a market comparison.
The price that knows you
An agent can also promise to find the best price while the sellers work to know the buyer. The Federal Trade Commission opened an inquiry into personalized pricing and ordered nine companies, from airlines to large retailers, to explain whether personal data shapes their prices or worker pay. The inquiry gathers information; it makes no finding against any named company.
If a hotel offers different people different prices for the same room, a saving against one quoted price leaves another comparison open. If personal data changed the offer, the price comparison must account for it, and the assistant may lack that pricing evidence. A best-price claim should state what the search covered, including any limits on comparable offers. The inquiry may help make sellers’ methods visible.
The cheaper answer
A report organized around completed sales can miss the search that found nothing worth buying. Sometimes the useful result is a decision to keep what the person already owns. An assistant that weighs a replacement must be free to recommend that outcome. The song’s singer finds it difficult to account for money the user leaves unspent.
The buyer can ask for the original plan beside the revised bill, with the booking saving and added expense on separate lines. The recommendation can name the sources searched and explain why an upgrade might be worth its price. That gives the traveler a choice the chorus cannot make for them.
Companions
- I Saved You Two Hundred, on the comparison that makes the report true.
- Before the Commitment Hardens, on evidence reaching an authority that can act.
- The Price of the Promise, on obligations imposed on people who never signed.
Bibliography
Federal Trade Commission, ‘FTC Launches Personalized Pricing and Compensation Inquiry’, October 28, 2026.
khayali, ‘I Saved You Two Hundred (You Spent Two Hundred More)’, music video, accessed October 30, 2026.
Kumar, Anushka, et al., ‘Agentic Shopping is Complicated and Contingent’, August 27, 2026.