Nobody can predict Amazon’s ad reserve prices, including Amazon. That is the defence.

The Federal Trade Commission and 22 states sued Amazon in Seattle on 31 August, alleging it inflated auction prices for Sponsored Products, Sponsored Brands and Sponsored Display, and generated more than $20bn in additional advertising costs since 2019. As CNBC reported, regulators say 1.2 million advertisers were affected, including more than 500,000 small and medium businesses.

Amazon published a detailed rebuttal the same day, calling the lawsuit misguided and saying it will make its case in court. The document is worth reading closely, because the two sides are not arguing about the same thing.

What each side is actually claiming

The FTC’s case is about disclosure. It alleges Amazon described one set of auction rules to advertisers while operating another, and that winners were charged their full bid rather than the second-price amount around 80% of the time.

Amazon’s answer is about outcomes. It says average cost-per-click for Sponsored Products search ads was flat adjusted for inflation from 2019 to 2024, conversion rates rose more than 24% from 2021 to 2025, and average winning bids fell 50%.

Both positions can be true simultaneously. Advertisers can have received better value over time and also have been told something inaccurate about how their money was being spent.

The mechanism, in Amazon’s own words

The company explains two floors. A “hard reserve” is the minimum a bid must clear to enter the auction and covers costs; a “soft reserve” is a real-time estimate of what a placement is actually worth.

When a winning bid exceeds both, the advertiser pays the soft reserve. When it clears the hard reserve but not the soft one, the advertiser wins anyway and pays their own bid, which is how a second-price auction can produce first-price outcomes without anyone paying above their maximum.

Amazon says roughly 92% of selected Sponsored Products ads in 2024 were not the highest bid, and that the mean winning bid ranked about 128th by amount. That is a genuine argument that relevance, not price, drives placement.

The line worth pausing on

Deep in the rebuttal is a sentence that does more work than Amazon may intend. Reserves, it says, are determined in real time and are not predictable in advance by anyone, including Amazon or the advertiser.

That is offered as a reason the disclosure could not have mattered, since an advertiser could not have acted on it. Read another way, it is a statement that the price floor in a $68.6bn advertising business is not knowable from outside or, apparently, from inside.

Amazon’s strongest defence is therefore that nobody could have gamed the auction because nobody can see it. That may well succeed in court, and it is not reassuring as a description of a market.

The disclosure evidence, and Amazon’s scale argument

Amazon says the FTC reviewed roughly 1.5 million pages and leaned on a handful of simplified communications. It puts hard numbers on the reach: three training courses cited in the complaint drew 1,849 enrolments and 779 completions over their lifetimes, and one video cited was seen by 928 viewers over two and a half years.

Those figures are checkable and, if accurate, are a real problem for a claim of a company-wide effort to deceive. Amazon also says the campaign builder has stated since 2018 that a bid is the maximum an advertiser could be charged.

The counter is that reach is not the test for a deception claim. What matters legally is whether the representations were misleading and material, not how many people completed the training course.

Why advertisers may not have noticed

Amazon argues advertisers bid on real-world performance rather than descriptions of mechanics, using automated tools that adjust to actual outcomes. It says 80% of bid changes on clicked Sponsored Products ads happened within a day of a previous change.

This is probably accurate and cuts in an uncomfortable direction. If sophisticated programmatic buyers cannot distinguish a relevance-weighted auction from a reserve-price one by watching results, then results are not an adequate substitute for disclosure.

Amazon also notes that advertiser contacts about auction mechanics stayed very low even after media coverage, and that it saw no change in spending. Low complaint volume in an opaque market is weak evidence of satisfaction.

The consumer question

On consumer harm, Amazon has the better of the argument on the current record. It points out that the FTC’s own complaint mentions consumers only a handful of times across more than 150 pages, that its damages model assumes no pass-through, and that redress is directed at advertisers.

That is a fair reading of what regulators filed. This is a case about who captured value in an intermediary market, not about retail prices.

It matters for how the case gets covered. Amazon’s treatment of its own products in search results is a separate long-running complaint and should not be blended into this one.

The wider pattern

Advertising is where the antitrust action now sits. A US judge declined to force Google to divest its ad exchange this week, leaving Brussels alone in arguing that only a breakup fixes the conflict of interest.

The pattern across both cases is that establishing conduct has proved easier than landing a remedy. Antitrust action against the digital advertising stack keeps arriving at the same wall.

Amazon has exposure beyond this. Penalties in the auction case could run to billions, and a separate antitrust trial over pricing pressure on brands is listed for early 2027.

What would settle it

The evidentiary question is narrow and answerable. Courts will decide whether Amazon’s published descriptions of its auction were misleading in a way that mattered, and Amazon says it properly described its pricing throughout.

The structural question will survive the verdict either way. An auction whose clearing price nobody can predict in advance is one that advertisers must simply trust, which is a considerable amount of trust to ask for $68.6bn a year.