The FTC and 22 states allege that Amazon represented Sponsored Ads as generalized second-price auctions, then used undisclosed soft reserves to charge more than the price produced by advertiser competition—sometimes the winner's full bid. Amazon confirms that soft reserves exist and that a winner can pay its bid, but disputes deception and harm, saying the reserve reflects placement value, relevance often lets lower bids win, and no advertiser pays above its maximum bid.
Those are litigants' positions in a case filed August 31, 2026, not a judicial finding. The useful map separates the mechanics both sides acknowledge from the pricing sequence the plaintiffs allege, Amazon's competing explanation and metrics, and the questions the federal court has not decided.
The question this map answers
What does the complaint say happened between an advertiser's bid and the charged cost per click, what does Amazon acknowledge or contest, and what evidence would be needed to move from a pricing difference to liability or damages?
MyMap reviewed the filed 181-page complaint, FTC case page and release, Amazon's same-day response, and Amazon's earlier DMA compliance description at 4:05 AM Pacific Time on September 1. The FTC case page labeled the matter Pending. No answer, discovery record, merits ruling, liability finding, damages award, or final injunction appeared in that controlling case timeline at observation time.
In this map, a bid is the maximum cost per click. A rank score combines bid and predicted relevance. A GSP result is roughly the amount needed to beat the next-ranked eligible ad, not necessarily the second-highest raw bid. A soft reserve is the placement-value threshold Amazon acknowledges using. Charged CPC is the click price, and first-price rate is the complaint's measure of how often it equals the winning bid.
The shared facts stop before the legal conclusion
Both sides describe a real-time process with maximum bids and relevance. Amazon's 2025 DMA annex says its rank score uses bid, relevance, and predicted click- or view-through measures. Amazon says the selected advertiser never pays more than its bid.
Amazon says a hard reserve is the minimum needed to enter, while a soft reserve estimates placement value. If a winner's bid exceeds both, it pays the soft reserve; if the bid clears only the hard reserve, it receives the placement and pays its bid.
Agreement on those components does not resolve disclosure, reliance, unfairness, or injury.
The complaint draws a represented-price path and an alleged charged-price path
The complaint quotes Amazon training and product materials that described a second-price auction in which a winner pays only what is needed to beat the next-ranked advertiser. One example used a $0.75 winning bid, a $0.60 competing bid, and a $0.61 charge. The plaintiffs say those or similar representations continued after Amazon introduced reserve pricing.
The pleaded pricing sequence is the map's upper lane. First, the system ranks eligible ads and determines a winner and second-place result. Then, the complaint alleges, Amazon applies a calculated “proxy” price or soft reserve that can replace the GSP result with a higher charge, capped by the winner's bid. The complaint characterizes that difference as an undisclosed surcharge and says internal constraints were designed to limit how visibly pricing deviated from GSP.
That is an allegation supported in the complaint by quotations and partially redacted internal materials. The public filing does not expose every model input, reserve amount, auction-level row, or underlying dataset. MyMap therefore labels the edge “plaintiffs allege”; it does not redraw an internal Amazon system as independently verified source code.
The plaintiffs' metrics describe scale, not a completed damages finding
The complaint estimates 1.2 million U.S. advertising customers, including more than 500,000 small and medium-size businesses, and alleges more than $20 billion was extracted. It says Sponsored Products advertisers paid their winning bid close to 80% of the time in 2024.
The complaint also says internal weekly reports showed the Sponsored Products first-price rate rising from 4% in a five-week 2020 period to 52–64% in a five-week 2023 period.
These are pleaded numbers. Redactions obscure values and methods, and a company total cannot be allocated to one campaign without auction-level inputs, scope, counterfactual bidding behavior, and an accepted injury model.
Amazon describes the same reserve as placement valuation
Amazon says relevance models allowed useful ads with lower bids to win, leaving premium placements undervalued. It calls soft reserves a common tool for estimating placement value and says advertisers manage bids from campaign performance.
Amazon says it updated campaign-building content in 2018 to explain the bid cap. It calls the cited second-price language older or simplified training content with low reach and denies a companywide effort to deceive.
Amazon says 92% of selected Sponsored Products ads in 2024 were not the highest raw bid; average winning bids fell 50% from 2019 to 2025; inflation-adjusted CPC was flat from 2019 through 2024; conversion rose more than 24% from 2021 through 2025; and its relevance approach saved more than $8 billion during 2021–2025.
Those are company estimates, not court findings or an independent audit. Lower winning bids or improved conversion can coexist with a reserve component on a click. Net effects require the underlying data and an agreed counterfactual.
A bid cap is not the disputed baseline
One common misreading is to compare charged CPC only with the maximum bid. Both sides agree the charge does not exceed that cap. The complaint's theory instead compares the charge with the price the represented GSP mechanics would have produced. Those are two different differences:
bid − charged CPCasks how much of the advertiser's maximum was unused;charged CPC − reproducible GSP resultasks whether another pricing layer raised the price above the claimed auction baseline.
Neither calculation by itself proves deception, materiality, reliance, causation, consumer pass-through, or legally recoverable damages. The map keeps mechanics, communication, behavior, outcome, and legal remedy as separate nodes so one difference does not silently answer all five questions.
The pending case asks for relief; it has not granted it
The complaint was filed in the U.S. District Court for the Western District of Washington. It asks for a permanent injunction, monetary relief, state penalties, restitution, disgorgement or forfeitures where authorized, fees, costs, and other relief.
A complaint establishes what the plaintiffs allege and request. It does not establish that Amazon violated a law or that the requested remedy will be entered. Amazon's public response establishes the company's position; it does not defeat the pleading by publication. The map ends at a pending adjudication state rather than drawing an arrow from allegation to liability.
Confirmed, derived, disputed, and still unknown
The sources jointly confirm bids plus relevance, a maximum bid, soft reserves, the possibility of paying one's bid, and a pending case. The complaint confirms its allegations and requested relief; Amazon's response confirms its reserve description, denials, and estimates.
MyMap derives the four-layer evidence model: shared mechanics, plaintiffs' alleged sequence, Amazon's response, and unresolved adjudication. It also derives the two price-difference tests and the audit record described below. These are analytical joins, not findings about a particular auction.
The parties dispute whether Amazon's communications were materially misleading, whether the soft-reserve path was adequately disclosed, how advertisers formed and adjusted bids, whether the challenged mechanism injured advertisers or consumers, and which counterfactual properly measures value or loss.
The sources leave unknown Amazon's answer and defenses in court; complete unredacted model logic and auction data; the GSP baseline and reserve for a named click; exposure to each representation; advertiser reliance; bid-shading behavior absent the challenged mechanism; campaign-level injury; consumer price pass-through; admissible damages; liability; and final relief.
Reproducible mapping method and next step
For each sampled click, preserve five dated records: the exact advertiser-facing representation and channel then in force; submitted and automatically adjusted bid; eligible ads plus rank inputs; the reproducible GSP calculation; and every hard-reserve, soft-reserve, constraint, and charged-CPC value. Join them with one auction or click identifier. Keep placement, device, keyword, ad product, campaign, and peak-day status separate because the complaint says pricing and visibility can vary across those dimensions.
Then calculate both price differences without labeling either “damages.” Attach evidence of which communication the advertiser encountered, how bids changed over time, conversion and sales outcomes, and any reporting limits. Aggregate only after defining a stable cohort and counterfactual. Where auction-level fields are unavailable, mark the record unavailable instead of reconstructing them from averages.
The practical next step for an advertiser is to preserve campaign-console language, bid histories, placement/device reports, click charges, and performance exports for the relevant period, then wait for the litigation record to test the missing mechanics and legal links. This explainer maps the dispute; it is not a damages estimate or legal conclusion.