In September 2024, a manager on Amazon's ad pricing team had a research paper pulled from the company's public science page and a major conference listing. His message to colleagues: "I should have been on top of this. Crap." What the paper described sits at the center of a federal case that names the most common piece of advice in e-commerce advertising as structurally false.

The advice is one sentence long. Set your max bid, let the platform run the auction, and you will pay just one cent more than the next highest bidder. Every PPC guide on the internet repeats some version of this. It sounds right. It is not.

The advice rests on real economics. In 2007, economists Benjamin Edelman, Michael Ostrovsky, and Michael Schwarz published a paper in the American Economic Review on how search ad auctions work. Their finding: in a generalized second-price auction, the rational move is not always to bid your true maximum. You will almost never pay it. The gap between your bid and the next bid down protects you.

That paper became the base for how platforms sold their ad systems to small businesses. Name your real number. The math keeps you safe. Amazon told more than 500,000 small and mid-size businesses they were in that kind of auction.

The research was real. The math was sound. The assumption underneath both turned out to be wrong.

In late August 2026, the FTC and 22 state attorneys general filed suit against Amazon, Case 2:26-cv-03097, Western District of Washington. The charge is not antitrust. It is consumer protection under Section 5(a) of the FTC Act.

The complaint alleges that in 2019, Amazon added a hidden surcharge to its ad auctions without telling anyone. Internal documents called it a "soft reserve price." One document used a more precise term: "invented auction participant." A phantom bidder. Not a real person. A number Amazon's system placed into the auction after the close.

The numbers trace the ramp. In late 2020, about 4 percent of winning bids resulted in the advertiser paying close to their full bid. By 2024, that share was 79.1 percent. The alleged total overcharge: more than $20 billion.

A senior Amazon executive, quoted in the complaint from a 2024 internal talk, called the system a "clever non-transparent way to charge first price." Internal documents show staff discussed raising prices while "hoping that advertisers don't notice and decrease bids."

The research is clear on this. A second-price auction only works when the second price comes from a real bidder. When the platform inserts its own number after the close, the auction is not second-price. It is first-price with a label that says the opposite.

The problem is not greed. The problem is that the surcharge was set after the auction closed, using the winner's own bid as the ceiling, and the bidder saw only their own result. The score and the scorekeeper were the same entity. No operator could spot the shift from the data Amazon showed them. That is not a pricing problem. It is a measurement problem.

Stop tracking cost per click as a trend line. Start tracking it as a fraction of your submitted bid. That ratio is the one view in which a first-price pattern shows up in your own data. Once that is clear, three moves follow from it.

Pull your campaign data for the past 90 days. For each ad group, divide your cost per click by your max bid. If the ratio sits above 0.90 on more than half your clicks, you are in a first-price auction no matter what the platform calls it. That single number asks more of you than any trend chart, because it forces you to look at the gap between what you offered and what you lost.

Most bid tools adjust based on CPC trends. That method assumes the auction is honest. Instead, set your max bid so that even at a 0.95 ratio, the cost per sale still clears your margin floor. Work from the worst case, not the average. If the worst case kills the margin, the campaign was never profitable. It just looked that way on a trend line.

Track the ratio, not the raw CPC. If the ratio stays flat after the cut, the platform is pricing to your bid, not to real competition. If it drops, a real bidder was setting the price. Both answers are useful. Neither is available from a trend line alone.

Running this for 90 days does something the standard advice never did. You see which campaigns are priced by real competition and which are priced by the platform itself. You see where your margin floor holds and where it was being cut without a trace. You see the gap between what you bid and what you paid, stated as a number you can act on. And you stop blaming your keyword list for a cost structure that had nothing to do with keywords.

At the end of 90 days, ask three things.

→ What moved the ratio?

→ What looked like a cost drop but left the ratio flat?

→ What friction showed up more than once across campaigns?

That is the difference between advice that sounds right and a system that proves itself.

The pulled paper is gone from Amazon's science page. The conference listing is gone. But the ratio between what you bid and what you paid is still in your campaign data. The mechanism that was hidden at the platform level is visible at the spreadsheet level. The data are in your account. The question is whether you have been reading them.