Amazon defends its ad auction by calling its own price floors unpredictable in advance

An Amazon seller sets a maximum bid, lets a campaign run for a fortnight, and reads back an average cost-per-click. Which of the two price floors in Amazon's rebuttal produced that number is impossible to tell from the bill alone. The rebuttal names both. A hard reserve is the minimum a bid must clear to enter the auction at all, and covers costs ^[8]. A soft reserve is a real-time estimate of what the placement is actually worth ^[9]. Beat both and the advertiser pays the soft reserve; beat the hard reserve but not the soft one and the advertiser still wins the placement and pays their own bid ^[10]. The FTC alleges the second outcome happened around 80% of the time ^[4]. On the government's own figure, most advertisers paid first-price bills inside what was billed as a second-price auction. Deeper in the document, Amazon says reserves are determined in real time and are not predictable in advance by anyone, including Amazon or the advertiser ^[12]. As a legal move that aims at materiality, since an advertiser who could not have acted on the information cannot have been harmed by its absence. As a description of a working control panel, it says the clearing price is not forecastable, which leaves the maximum bid as the only number in the account with a fixed meaning. The size of the claim is easier to hold at the level of one account. More than $20bn since 2019 ^[1], spread across the 1.2 million advertisers regulators say were affected ^[2], is at least about $16,700 each ^[2]. More than 500,000 of those are small and medium businesses ^[2], so the average sits well above whatever a typical small seller would recognise in their own billing. That is an allegation, not a finding. Amazon answers with outcomes rather than rules. It says average cost-per-click on Sponsored Products search ads was flat once adjusted for inflation between 2019 and 2024 ^[5], conversion rates rose more than 24% from 2021 to 2025 ^[6], and average winning bids fell by half ^[7]. It adds that roughly 92% of selected Sponsored Products ads in 2024 were not the highest bid, with the mean winning bid ranking about 128th by amount ^[11]. TNW's reading of the two filings is the useful one: both sides can be right, since an advertiser can receive better value over time while also having been told something inaccurate about the mechanics ^[20]. The disclosure evidence gets the same treatment. Amazon says the FTC reviewed roughly 1.5 million pages and leaned on a handful of simplified communications ^[14], then counts the audience. Three training courses cited in the complaint drew 1,849 enrolments and 779 completions across their lifetimes ^[15], a completion rate of about 42% ^[1], and one cited video was watched by 928 people over two and a half years ^[16]. TNW notes that the legal test turns on whether the representations were misleading and material, not on reach ^[19]. There is a second reading for the operator: if those courses and that video were where the auction mechanics were spelled out, the mechanics reached a rounding error of the 1.2 million advertisers buying into them ^[2]. The most telling number Amazon offers is about behaviour. It says 80% of bid changes on clicked Sponsored Products ads came within a day of a previous change ^[18], offered as evidence that advertisers bid on real performance rather than on descriptions of mechanics. Buyers retuning that fast were reacting to performance numbers alone; the results carry no signal that would let them tell a relevance-weighted auction from a reserve-price one. For a team rebuilding a Sponsored Products forecast this quarter, the split worth drawing is between what the platform enforces and what the team inferred.