The FTC says Amazon charged advertisers their own bid 79 percent of the time. The lesson is about auctions, not lawsuits

The US Federal Trade Commission and 22 state attorneys general filed a 181-page complaint against Amazon on 31 August 2026, alleging that the company ran a generalised second price auction for its ads and then applied an undisclosed soft reserve on top of the result. The case, United States v. Amazon.com, Inc., number 2:26-cv-03097 in the Western District of Washington, is a deception case under Section 5(a) of the FTC Act and the deceptive trade practices statutes of 22 states, not an antitrust case.

Everything in that complaint is an allegation. Nothing in it has been tested in court, no liability has been found, and no remedy has been ordered. Amazon rejects the case. The reason it is worth reading anyway has little to do with the litigation and a lot to do with auction mechanics, because the filing describes in unusual detail what an extra pricing layer does to cost per click over seven years, and why the advertisers paying it could not see it in their own reporting.

An allegation, not a finding

This point is easy to lose once the numbers start moving, so it is worth fixing in place first. A complaint is one side's account of events, written to survive a motion to dismiss. The internal quotations in it were selected by the agency bringing the case. Amazon has not had its version tested either. Every figure and every quotation below is what the FTC says, as reported by PPC Land, and should be read that way.

PPC Land reports that Amazon published a rebuttal on the same day, arguing that average cost per click stayed flat in inflation-adjusted terms and that advertisers saved more than 8 billion dollars because of the same pricing system. Amazon's position, per that reporting, is that soft reserves represent its estimate of the true market value of an ad placement, that reserves of this kind are common across the industry, and that no advertiser ever pays more than its own bid.

How a second price auction is supposed to work

In a sealed-bid second price auction, the highest bidder wins and pays what the runner-up bid, plus a token increment. The point of that rule is not fairness in the abstract. It is that the rule makes honest bidding rational: because the price is set by someone else's bid, there is no gain from bidding below what the click is worth to you, and shading your bid only risks losing an auction you would have wanted to win.

A generalised second price auction extends that idea to multiple slots and usually ranks bidders by bid and a relevance score rather than by bid alone. The important property survives: the price paid is meant to be determined by another bidder, not by the seller.

The complaint's central allegation is that the price stopped being determined that way. According to the filing as reported by PPC Land, Amazon calculated the auction price, then applied a soft reserve on top of it, converting the auction result into the amount actually charged.

What a soft reserve is, and why the timing matters

A conventional reserve is a floor published before an auction runs. A bid that fails to clear it does not compete at all. Every bidder knows the floor exists, and the floor cannot take money from a bid that already won.

The mechanism described in the complaint operates after the winner and runner-up have already been determined, and its only ceiling is the winning bid itself. That ordering is the whole story. Applied before the auction, a reserve is a term of participation. Applied after the auction, with the winner's own bid as the cap, it changes what is being charged rather than who competes.

Amazon's own staff appear to have named it plainly. A senior scientist described it as an invented auction participant representing how much Amazon thinks a particular ad slot is worth, according to the complaint. A senior vice president in the advertising organisation wrote that the second price is not set by an actual bidder but is a proxy second price that Amazon calculates.

The number that travels: the first price rate

The first price rate is the share of clicks charged at the advertiser's own bid rather than at anything resembling a runner-up bid. It is the cleanest single measure of how far a second price auction has drifted from its own description, and the complaint tracks it across four years.

For Sponsored Brands in 2024, winners paid their own bid roughly half the time, according to the same filing. Digiday's explainer of the case, published by Seb Joseph on 1 September 2026, describes the same trajectory as a surcharge occurring around 80 percent of the time by 2024, up from 30 to 40 percent in 2021.

Read the table as a mechanism rather than a scandal. When four percent of clicks are charged at the bid, the auction still mostly behaves as described. When 79 percent are, the advertiser is functionally in a first price auction while still being told the rules of a second price one. The bidding strategy appropriate to each is different, and that difference is the harm the complaint describes.

Where it was applied, and when

The mechanism was not switched on everywhere at once. According to the complaint, soft reserve pricing began in Sponsored Brands during the 2018 Christmas period, extended to Sponsored Products second price auctions in mid-2019, and reached Display Ads in 2023. Three formats, five years, one direction of travel.

The staged rollout matters for anyone reconstructing their own historical performance. A cost per click series that spans 2018 to 2023 crosses several changes in how the price was formed, and none of those changes were announced in a way an advertiser could have annotated a chart with.

Why seasonality made it hard to detect

This is the part with the most transferable lesson, and it does not require believing anything about Amazon's intent.

A cost per click number moves for many reasons. Competitors raise bids. Demand spikes at Christmas. Query mix shifts toward more expensive terms. Relevance scores change. Against that noise, a pricing layer that adds a few percent is not visible in a chart, because there is always a plausible seasonal explanation available for a rise.

The complaint quotes internal remarks acknowledging exactly this dynamic. A Sponsored Brands employee wrote in 2018 that price increases were masked by holiday increases in advertiser demand. The head of Sponsored Products auctions noted that surcharge constraints could be relaxed when there were enough confounders, seasonality among them, to make bid shading difficult to detect.

Two episodes in the filing show the limit of that cover. On 10 December 2021, the auction team removed a click-level surcharge constraint and prices climbed over five days. By 13 December more than 20 agencies and advertisers had escalated. On 17 December the vice president of Sponsored Products convened a war room, and on 20 December the team drafted a response attributing the movement to Black Friday and Cyber Monday patterns. An internal summary revised on 9 February 2022 had removed the discussion of generalised second price mechanics and reserve pricing entirely; a product management colleague had objected on 1 February that the revision made it sound as though Amazon Ads had no role in the fluctuations.

Prime Day 2023 runs the same sequence in the other direction. On 11 July the auction team reported cost per click below expectations. On 12 July it obtained approval to raise the surcharge constraint, on an order form stating the purpose as Prime Day ROAS maintenance. An account manager relayed a grocery client's complaint that CPCs were up more than 90 percent and killing return on ad spend.

What an advertiser can actually do about this

None of what follows is a claim about Amazon or any other platform. It is what the allegations imply about measurement practice generally, and it holds whether or not the case succeeds.

Watch cost per click against your own bid, not just against last month. The signal in the complaint is not that CPC rose. It is the ratio between what advertisers bid and what they were charged. Most reporting workflows track CPC as a trend and never compute it as a fraction of the maximum bid, which is the one view in which a first price pattern would show up.

Treat a platform's reported prices as unaudited. They are the seller's account of a transaction the seller also ran. That is not an accusation, it is a description of the structure, and it applies to any closed auction where the same company sells the inventory, operates the auction and reports the result.

Be suspicious of seasonal explanations for cost changes you did not cause. A rise that arrives with a ready explanation is the hardest kind to investigate. Keeping a record of your own bid changes, so you can separate movements you caused from movements you did not, costs nothing and is the only baseline you fully control.

Automated bidding weakens all of this, because a bid you did not set is harder to compare a charge against. That is a reason to keep some manually bid inventory as a control, not a reason to abandon automation.

What this means for Thai marketers

PPC Land reported no figure for Thailand or Southeast Asia, and the complaint covers United States advertisers. The relevance here is structural rather than jurisdictional. Retail media and marketplace ad platforms are the fastest growing part of the regional mix, and most of them share the property at the centre of this case: the seller of the inventory, the operator of the auction and the reporter of the results are the same company.

For teams running ecommerce marketing across marketplaces, the practical response is a reporting habit rather than a platform decision. Track charged price against submitted bid wherever the platform exposes both. Keep a channel where the auction rules are published, so you have something to compare against. Advertisers who also run Google Ads already have a second dataset with a different auction and different disclosure practices, and the comparison between them is more informative than either alone. Brand owners whose organic marketplace and Shopify SEO performance is measured separately from paid should be wary of any single platform's numbers standing as the only evidence of what happened.

What the source did not say

PPC Land did not report any finding of liability, because there is none. It reported no remedy, no settlement and no ruling. It reported no figure for Thailand or Southeast Asia. It reported nothing establishing that other advertising platforms operate the same mechanism, and no such claim should be read into any of this. Several dollar estimates, average surcharge rates and the size of the reserve guardrails are redacted from the public version of the complaint.

FAQ: the FTC complaint about Amazon ad pricing

What exactly did the FTC accuse Amazon of doing?

The complaint alleges Amazon ran a generalised second price auction and then applied an undisclosed soft reserve on top of the result, converting the auction price into the price actually charged. It was filed on 31 August 2026 by the FTC and 22 state attorneys general as case 2:26-cv-03097 in the Western District of Washington.

Has Amazon been found guilty of anything?

No. The filing is a complaint, which is an allegation and not a finding, and Amazon rejects the case. PPC Land reports that Amazon argues average cost per click stayed flat in inflation-adjusted terms and that advertisers saved more than 8 billion dollars under the same pricing system.

What is the difference between a soft reserve and a normal reserve price?

A conventional reserve is published before the auction and a bid that fails to clear it does not compete, while the mechanism described in the complaint operates after the winner and runner-up have been determined and is capped only by the winning bid. That ordering is what the FTC's case turns on.

Why could advertisers not detect this from their own data?

Cost per click moves for many reasons at once, so a pricing layer is hard to separate from seasonal demand. The complaint quotes a 2018 internal remark that price increases were masked by holiday increases in advertiser demand, and a note from the head of Sponsored Products auctions that surcharge constraints could be relaxed when confounders such as seasonality made bid shading hard to detect.

Is this an antitrust case?

No. It is a deception case brought under Section 5(a) of the FTC Act and the deceptive trade practices statutes of 22 states. The subject is what advertisers were told about how prices were set, not market power.

If you buy on marketplace or retail media auctions and your only record of what a click cost is the platform's own dashboard, this is a reasonable week to add charged price against submitted bid to your reporting. Relevant Audience helps ecommerce teams in Thailand build measurement that does not depend on a single platform's word for it.