One internal Amazon document, quoted by the Federal Trade Commission, says the company's auction pricing had "a surcharge hidden in it." The "it" was the price of an advertising auction, and the hiding place was the auction itself.

On August 31, the FTC and attorneys general from 22 states filed an antitrust and consumer-protection suit against Amazon in federal court in Washington. The complaint alleges that Amazon told advertisers it ran a generalized second-price auction, the industry-standard format for search advertising. The complaint says Amazon quietly changed that arrangement beginning in 2019. It added what the company called internally a "soft reserve price," then used a private pricing layer to collect more than the competitive auction would have produced.

The FTC alleges that Amazon's Sponsored Products advertisers paid their own winning bid close to 80 percent of the time in 2024. The alleged scheme ran for more than seven years, affected more than a million brands and sellers, and generated tens of billions of dollars in revenue. Those are allegations, not findings. Amazon has called the suit misguided and says the FTC assembled it from a handful of simplified training materials. But the case poses a question worth asking before the lawyers settle the facts: what happens to a market when the rule that sets the price changes and the sentence explaining the price does not?

To see why the distinction matters, imagine the small ceremony that an advertiser thinks it is entering. A seller wants its product to appear when someone searches Amazon for a keyword. It submits a bid, the most it is willing to pay for the placement, and other sellers submit theirs. Amazon ranks the candidates, taking account of the bid and the ad system's other inputs, and awards the placement to the highest-ranked eligible advertiser.

Under the second-price convention Amazon described, the winner does not pay the whole amount it offered. It pays the amount required to beat the next competitor, described in Amazon's public materials as one cent more than the next-highest bid. The point is not the penny. The point is that the price is supposed to be discovered in the competition among advertisers rather than set by the winner's maximum willingness to pay.

That rule changes the way people bid. In a first-price auction, the winner pays what it bid. A bidder who offers its true maximum risks paying the full value of the offer, even if nobody else came close. Repeatedly facing that risk, it may shade its bids downward, trying to learn the minimum amount that will win. In a second-price auction, the bidder has more reason to bid close to its true value. It expects the price to be determined by the next-best offer rather than by the ceiling it disclosed. The format is part of the product.

The obvious objection to the FTC's theory is not difficult to make. A digital advertising auction is not a rustic room in which independent bidders gather under a neutral auctioneer. Amazon owns the store, controls the search page, decides which ads are eligible, and operates a complicated ranking system. It may need reserves to keep low-quality ads out of valuable space, and it may price inventory differently on a quiet Tuesday and on a day when millions of people search for gifts. A reserve price, if disclosed, can be a normal part of auction design.

The question, then, cannot simply be whether Amazon used a reserve or whether it wanted to earn more. Platforms are allowed to make money. An auction can be designed around a floor. The question is what Amazon represented about the source of the price, and whether the alleged change was disclosed to the people whose bids were shaped by the old rule.

The stronger objection is the one a litigator reaches for first, and the complaint's own structure invites it. Its representation sections point at training and marketing materials, at what account managers told advertisers, and at what the head of Amazon Ads said out loud. None of those is a contract. Amazon's published advertising terms are the ordinary platform instrument, modifiable at Amazon's sole discretion, and the second-price promise does not live in them. Amazon's public response takes exactly this line, saying the FTC "cherry-picked a small number of materials, such as a few online educational videos and training content that contained older or simplified examples about how our auctions are run." If the description was never a binding term, there was no promise to break.

That objection is real and will decide part of the case, particularly the state claims asking a court to rescind or reform contracts. It is weaker against the rest. A deception claim does not need a contract. It needs a representation that was material and likely to mislead a reasonable buyer, and a marketing claim is not repaired by silence in the terms beneath it. The complaint alleges that Amazon said nothing about reserve pricing on its own website until October 30, 2025, almost a year after it learned of the FTC's investigation. And the internal language quoted in the complaint reads like a company relying on a belief rather than on a document. "Only GSP logic is known publically [sic]," one employee is quoted as writing. Another, in 2024: "We don't tell them about the surcharge, we let them assume GSP-based."

The complaint's distinction is between a stated competitive mechanism and a concealed price-setting one. A disclosed reserve gives advertisers a new fact to put into their bids. A hidden one leaves them bidding against a description of the market that no longer matches the machinery beneath it.

The FTC says Amazon's internal language makes that machinery unusually plain. The executive in charge of Amazon Ads is quoted as saying that in Amazon's auctions "the second price isn't set by an actual bidder, but rather by" Amazon itself, in the form of a "proxy 2nd price that we calculate." Another internal document, according to the complaint, referred to an "invented auction participant." An employee is quoted saying that the surcharges produced prices "beyond what [can] be achieved through advertiser competition."

That is the auction after the auction. The visible contest still happens. Advertisers submit bids, Amazon ranks them, and a seller wins a placement. But after the apparent contest produces a price, the platform allegedly runs a second calculation against a participant it created for the purpose, and the number that reaches the invoice comes from there. The vocabulary matters because of what it shows the company understood. Amazon's own engineers could not describe their system without saying, to each other, that the price had stopped coming from a bidder.

The economics deserve more care than the phrase invites. A reserve price is the seller's own valuation entering the auction, and auction theory does not treat that as a distortion; it puts it at the center. The standard way to implement a revenue-maximizing reserve is to model the seller as a phantom bidder offering it. Every ad exchange has one. Amazon now says something close to this in public, describing the soft reserve as "a real-time minimum value that seeks to better reflect what each placement is actually worth."

So the invented participant is not the offense. It is the evidence. A phantom bidder is unremarkable in a market that knows it is there, and becomes something else in a market that has been told the next advertiser sets the price. The internal documents are valuable to the FTC because they show Amazon describing to itself, in precise language, a pricing mechanism it was not describing to its customers.

The complaint says that misunderstanding was worth money. Amazon's own auction team is quoted observing that advertisers "are operating under the assumption that Amazon uses a GSP auction." A disclosed change collects more on each transaction once. A concealed change collects more and preserves the bidding behavior that made it collectible.

The figures in the complaint describe that behavior shifting over time. For Sponsored Products, the FTC says, the share of clicks priced at the advertiser's own winning bid rose to between 30 and 40 percent in 2021, then 70 percent in 2022, and approximately 80 percent in 2024. The complaint attributes the increase to the surcharges.

A trend like that does not prove its own cause, and there are better alternatives than advertisers turning generous. Over the same years the market moved onto automated and target-return bidding, where a submitted "bid" is a control setting that software adjusts rather than a ceiling a human disclosed, so "paid their own bid" may mean something different in 2024 than it meant in 2021. A shift in the mix toward thin keyword auctions, where no second bidder exists and the winner pays the floor by default, would move the same line. What makes those explanations less likely is that Amazon allegedly ran the experiment itself. The complaint says the company repeatedly tested surcharge increases on some advertisers and not others to see whether anyone noticed, including a large-scale test in 2023 later applied across its United States customer base. Those advertisers faced the same bidding tools and the same auctions as everyone else, so the tests measured the instrument's own effect with the drift held constant on both sides.

The alleged timing matters too. The FTC says Amazon began the change in 2019, when it was dissatisfied with advertising revenue, raised the surcharges on high-volume shopping days such as Prime Day and Black Friday, and ramped them in ways intended to disguise the increase. A platform that openly changes its rate card has made a commercial decision. A platform accused of doing it hardest when demand is high, and then hiding it, is accused of managing the customer's understanding as well as the auction.

The complaint quotes a 2024 discussion among senior executives, including the head of Amazon Ads and Amazon's chief digital economist, describing a "clever non-transparent way to charge first price" as an "incredibly effective way to drive revenue." The court has not decided what the words prove. But "first price" describes a different bargain from second price, and "non-transparent" is not an incidental adjective if the business depends on bidders behaving as though the system were transparent.

The serious allegation is that the interface taught one strategy while a hidden mechanism rewarded another, so that the advertiser's mistaken belief became an input to the pricing system. The arrangement is hard to see from outside, because the transaction gives the customer so little evidence. An advertiser sees a bid, a placement, and a charge. It does not see the bid that would have won under the old rule, the reserve that entered after the ranking, or the counterfactual price produced by genuine competition. It may see that its ad won and that it paid less than its maximum bid, which can look like confirmation that the second-price rule worked.

Consider the simple case Amazon's public explanation invites an advertiser to imagine. The seller bids $4. The next seller bids $2.70. Under the described rule, the price is $2.71. Now put a hidden reserve of $3.50 into the same auction. The seller still wins, still pays less than its $4 ceiling, and still sees no rival who appears to have beaten it. But the bill is now $3.51. The two systems can produce the same placement and the same winner while charging different prices. The customer cannot recover that difference from the receipt. It needs the records of what happened between ranking and billing, and those records belong to the platform.

This is why the alleged conduct cannot be reduced to a complaint about an unexpectedly large invoice. The central fact, if the FTC proves it, is that the advertiser's own bid helped conceal the change. Paying below the ceiling looks like evidence that the second-price bargain survived. It may instead be evidence that the platform found a new way to approach the ceiling without crossing it.

Amazon's answer to all of this is that nobody was stranded. Advertisers watch what they pay and what it returns and re-optimize continuously, so a change in the price rule is absorbed within weeks. Amazon points to its own numbers, saying that between 2019 and 2024 the average cost-per-click, adjusted for inflation, stayed flat, and that in no scenario does an advertiser pay more than its bid.

The platform's informational advantage is what that answer leaves out. Amazon keeps the auction's records and can watch how advertisers respond when it changes the mechanism. According to the complaint, when advertisers escalated complaints about unexplained cost-per-click spikes, in December 2021 and again after Prime Day 2023, Amazon denied having played any role in raising the prices. The complaint also alleges that advertisers exposed to higher surcharges in the 2023 test did not reduce their bids or budgets. A market corrects when it can see what it is correcting. And a flat average cost-per-click does not close the question, because the complaint alleges the surcharges were calibrated against advertisers' return on ad spend, not the headline price.

The scale of the customer base gives the allegation its less cinematic force. The FTC says Amazon's roughly 1.2 million American advertising customers include more than 500,000 small and medium-sized businesses. For a national company, a hidden difference between $2.71 and $3.51 may disappear into a marketing report. For a small seller buying a placement every day, the difference is a decision about whether the product can be advertised at all. The auction's rules travel directly into the price of staying visible.

A hard price floor is visible as a rule. A soft reserve sounds adjustable, almost atmospheric. It can move with demand, category, time of day, or a revenue target. That flexibility may be useful for a platform. It also makes the price harder for a customer to challenge. The reserve does not announce itself as a separate charge; it appears as the result of the auction, wearing the authority of a rival who was never there.

The FTC will have to connect the hidden pricing rule it alleges to the promises made to advertisers and the prices they paid. Amazon will argue that its mechanics, reserves, and commercial terms were permissible, and that the complaint mistakes a complex market for a simple promise.

That is the right argument to test. Complexity is not suspicious merely because it is difficult to explain. But it cannot license a simple public explanation after the private system has changed. If an advertiser needs to understand that Amazon can insert a reserve, adjust it for demand, and calculate a proxy price no actual competitor offered, then those facts are terms of the auction and belong where it is described.

The ordinary seller does not need to know every line of the ranking code. It does need to know whether the price came from a competitor, from a disclosed floor, or from the platform's own estimate of what the slot was worth. Amazon was always free to set that floor. What the complaint says it did not do was say so. Those are different markets, even when they produce the same page and the same winning ad.