A Sept. 25 press release from the Federal Trade Commission (FTC) declares a sort of victory: “FTC Secures Historic $2.5 Billion Settlement Against Amazon.” The settlement comes in the infamous “dark patterns” case, where—as the press release reminds us—the agency “alleged that Amazon used deceptive methods to sign up consumers for Prime subscriptions and made it exceedingly difficult to cancel.” (Here’s the initial June 2023 complaint, and here’s the amended complaint filed in September 2023.)
Those methods must have been sneaky indeed, given there were a reported 197 million Amazon Prime subscriptions in the United States alone by March 2025—up from a nontrivial 184 million subscribers in June 2024. That is a rather a large number, given the total U.S. population of about 340 million people.
$2.5 billion is real money, even for Amazon, and even in Washington. There are behavioral remedies too, although these don’t seem terribly burdensome, all things considered. For example, we’re told that Amazon will now have to include “a clear and conspicuous [virtual] button for customers to decline Prime. Amazon can no longer have a button that says, ‘No, I don’t want Free Shipping.’”
Of course, some former FTC officials—notably, Lina Khan and Alvaro Bedoya—have feigned shock that the case settled for such a paltry sum, letting both Amazon and individually named plaintiffs “off the hook” under the Restore Online Shoppers’ Confidence Act (ROSCA). They went so far as to suggest there may have been political pressure from the White House.
Here’s a 2019 review of the FTC’s ROSCA enforcement actions dating back to 2011. For some perspective on individual liability for corporate officers, then-Commissioner Christine Wilson’s concurring statement in FTC v. Progressive Leasing is instructive. Holding individual executives accountable for engineering and perpetrating actual frauds is one thing. The Amazon complaint describes no such thing.
The big deal is the transfer of rents. There’s to be a $1 billion civil penalty “which is the largest ever in a case involving an FTC rule violation” (the $5 billion Facebook settlement in 2019 doesn’t count, I gather, as it was supposed to settle an alleged order violation). And there’s also to be $1.5 billion in consumer redress, which seems to reflect an entirely dubious estimate of consumer harm, given what we know from the complaints.
It’s an especially big deal, as this seemed a conspicuously weak case when it was filed (although not quite as weak as a monopolization case the FTC also brought against Amazon). This was not an antitrust case, but a consumer protection case, brought under ROSCA and Sections 5 and 19 of the FTC Act. ROSCA in pertinent part stipulates:
It shall be unlawful for any person to charge or attempt to charge any consumer for any goods or services sold in a transaction effected on the Internet through a negative option feature (as defined in the Federal Trade Commission’s Telemarketing Sales Rule in part 310 of title 16, Code of Federal Regulations), unless the person—
(1) provides text that clearly and conspicuously discloses all material terms of the transaction before obtaining the consumer’s billing information;
(2) obtains a consumer’s express informed consent before charging the consumer’s credit card, debit card, bank account, or other financial account for products or services through such transaction; and
(3) provides simple mechanisms for a consumer to stop recurring charges from being placed on the consumer’s credit card, debit card, bank account, or other financial account.
How simple? ROSCA doesn’t say.
The FTC’s complaint alleged:
For years, Defendant Amazon.com, Inc. (“Amazon”) has knowingly duped millions of consumers into unknowingly enrolling in its Amazon Prime service (“Nonconsensual Enrollees” or “Nonconsensual Enrollment”). Specifically, Amazon used manipulative, coercive, or deceptive user-interface designs known as “dark patterns” to trick consumers into enrolling in automatically-renewing Prime subscriptions.
On the other hand, “dark patterns” are never subject to any clear technical definition in either computer science or the law, and there may be indefinitely many points along the spectrum running from (A) to some degree, for some consumers, and for some purposes, a suboptimal user interface to (Z) a user interface effectively designed to perpetrate a fraud on the median or typical consumer. That is, an allegation of “dark patterns” tends to raise the question of how dark are those patterns of which you speak?
The FTC’s 2020 case against MyLife.com may be a good deal closer to the fraudulent end of the spectrum. MyLife was alleged to have misled consumers into purchasing background reports (as alleged, the marketing appears to have been plainly and intentionally misleading). It was also alleged that MyLife had failed to provide any online or email option for cancellation; customers were required to telephone and, in many cases, failed to reach anyone, with long wait times often leading to their calls being dropped. Numerous customers had brought suit, alleging that it had been impossible to cancel a subscription.
The FTC’s 2025 case against the education-technology firm Chegg also seems a sound consumer protection case. For one thing—at least as alleged—the considerably more complicated (and promised and required) online-cancellation process that Chegg provided did not actually work for consumers using a mobile web browser; consumers inquiring about cancellation on mobile could find their way to a message suggesting they try via a desktop or laptop machine instead. For another, the complaint alleged that some customers who successfully navigated the online self-cancellation system continued to be charged after cancellation. The settlement in Chegg, also this month, stipulates that the defendant pay $7.5 million, in addition to providing a simple cancellation mechanism and compliance reporting.
Of course, the details matter. An occasional glitch in billing should be rectified, but it’s not obviously a species of fraud or a Section 5 violation. But the allegations in the complaints against MyLife and Chegg at least describe more serious consumer difficulties.
The complaint against Amazon was fundamentally different—or, at least, very different in degree. If readers will forgive some self-plagiarism, I provided a tolerably quick overview back in June 2023:
Prior to recent changes, the complaint alleges that “the primary purpose of the Prime cancellation process was not to enable subscribers to cancel, but rather to thwart them. Fittingly, Amazon named that process ‘Iliad,’ which refers to Homer’s epic about the long, arduous Trojan War.”
Now we get it. Canceling one’s Prime membership is akin to reading a 700-page account of a 10-year war, if not precisely like fighting in a 10-year war. That seems bad, not that I want to dissuade anyone from reading one of the classics of Western lit. But tastes vary, and it is indeed a long book (or 24 “books,” as the poem is organized).
I’ve not tried to cancel Prime myself (because I don’t want to cancel Prime) but others have told me that it’s easily accomplished in a minute or less. And having taken a fresh look at the website, I find that entirely plausible. The commission’s account of the horrors of ancient war—sorry, Prime cancellation—begins on page 43 of the complaint, where we learn that, prior to recent streamlining, cancellation involved navigating the Scylla and Charybdis (sorry, that’s the Odyssey) of “the Iliad Flow,” which “required consumers intending to cancel to navigate a four-page, six-click, fifteen-option cancellation process. In contrast, customers could enroll in Prime with one or two clicks.”
That is, one or two clicks, once they’ve registered with Amazon and taken the time to input the required information. If you express (by clicking) an interest in cancellation, you are asked whether you mean it, presented with a pitch to change your mind, an offer of a discount, and, eventually, an effective click to cancel. To be sure, six clicks is more than two, but I wouldn’t expect it to take 10 years or the death of Achilles.
Does it take 10 minutes? Does it take a single minute or less, as some have said? Does it take the time it takes to read the Iliad (translated into a language I can actually read)? Are we talking about the time required by the median webpage designer, the median consumer, or some substantial class of disadvantaged consumers? Some simple consumer testing might tell us how long it takes the median or modal consumer (or how long it takes on average). It might also tell us what percentage of users attempting to cancel are thwarted—or, at least, diverted—by the “dark patterns.” But the complaint doesn’t say anything about such testing or, if it was done, its results.
While we’re at it, how many millions of consumers were “duped” into signing up for Prime in the first place? “Millions” seems rather vague. Amazon says that it has about 200 million Prime members. One wonders (a) how many (or what percentage) were enrolled (and charged) without their knowledge and (b) how many (or what percentage) of them were unable to withdraw from Prime, having been so duped. Roughly?
The complaint doesn’t say. It does provide examples of the alleged “dark patterns,” but these seem to include both the price of Prime membership and the terms. I don’t recall feeling duped myself, but maybe that goes hand-in-hand with being duped.
There was, to be clear, some sort of staff analysis. The FTC has also posted an expert report prepared for the commission by Neale Mahoney, an economist at Stanford and at Bates White. That report, with appendices and figures, runs 118 pages, but it’s of little help, because nearly everything technical of interest has been redacted. There are no numbers, but for the page numbers. Tables and figures have been blacked out, and whatever analysis may have been described in the report is also redacted.
Some of this was no doubt necessary—the redaction of nonpublic business information is not in itself unusual. It’s just that we have a 118-page report from a PhD economist, but no way to evaluate the parts of it that are . . . economics. We are told, for example, that Mahoney used a “‘differences-in-differences’ regression approach, common in empirical economics, to quantify economic harm from unsuccessful cancellations where subscribers believed they had cancelled.” And that’s fine. Diff-in-diff methods are common enough in peer-reviewed empirical articles. But we don’t see it. No data, no analysis, and no results. It’s a step above a computer scientist telling us that she used “an algorithm,” and a small step at that.
Another point of interest is Mahoney’s use of survey data from Amazon’s own “cancellation survey.” Survey data are not ideal, but well-designed surveys can provide useful information where direct measures are unavailable. Still, surveys have their limits, and Amazon’s “ordinary course” surveys were not really designed to get at the question of whether consumers—in any number—were deliberately or negligently misled or deceived. And while Mahoney’s defense of the survey evidence seems defensible at some level, it also seems very plainly to oversell the extent to which the reported magnitudes (and Mahoney’s sampling and analysis of the survey data) may be accurate or even “conservative.”
The report’s vague conclusion that “a substantial number of subscribers exited the cancellation process with the mistaken belief that they successfully cancelled their Prime subscriptions” may be true, but now we really want some numbers. How many—or what tranche of—consumers left with such mistaken beliefs? Under what conditions? How long did those mistaken beliefs persist? Independent of the question of fault, what harm was done on average (or what median harm was done)?
Feel free to disagree—and to provide some evidence to support your disagreement—but to my mind, the standard eight-sided stop sign is a very clear sign. It’s highly legible and it means stop. Nonetheless, people run stop signs every day. If 200 million drivers are presented with a stop sign, some “substantial number” of them will fail to “see” it (or fail to register that they’ve seen it), misread it, or otherwise act as if they have not seen or understood it. Many might subsequently report—intending no falsehood—that they did not see the bright red sign.
Suppose that 2% of Prime subscribers attempt to cancel their membership every year, representing roughly 4 million consumers. That’s a lot of people. If 95% of those 4 million leave the cancellation process correctly understanding that they have, indeed, cancelled their Prime membership, whereas 5% leave with the erroneous belief that they’ve cancelled, then 200,000 will have exited the process with a false belief. That is, by any stretch of the imagination, a “substantial number” of people.
But, of course, I’ve pulled these numbers out of thin air. Is 200,000 in the ballpark? Is it within an order of magnitude? Is it within two orders of magnitude? Is the incidence of confusion 5%, 50%, or 0.5%? The fact is, I don’t know the real numbers, and the redacted version of the expert report does not provide them.
I’m also not arguing that the cancellation process was as clear as a stop sign, that it was ideal, or that it was objectively easy. It seems pretty easy to me. The web-page cancellation slides provided by the FTC and in its expert report do not seem confusing. They do not make an instant cancellation mechanism as clear and conspicuous as possible, but they are legible. There’s nothing obviously false in the text and it’s not as if the options are buried in multiple pages of dense legal text, as one might find with consumer software licensing agreements, privacy policy notices, or various consumer warranties.
Moreover, the six clicks described in the FTC complaint (the FTC complaint acknowledges that Amazon had revamped the process, making it easier “shortly before” the FTC filed its complaint) were not six (of the twelve) labors of Hercules. They were six clicks of a mouse or six taps on a pad or screen.
It doesn’t seem that hard to me. By way of comparison, try clarifying insurance coverage or a health-care provider bill with customer service, whether online or by phone. Try calling an airline’s customer-service number. My own perceptions, intuitions, and personal experience are not the issue. I am not the typical, median, or “reasonable” consumer made flesh. That’s why I’d like to know the numbers, to start. And I’d like to know where the numbers came from. That’s why I’d like to see some actual consumer testing in a case of this importance.
I also don’t know what the legal standard is supposed to be. “Deceptive” practices under Section 5 may include materially misleading marketing claims as well as false ones, but that prompts not only the question of how to determine materiality, but the question “misleading to whom?” It’s not an “all or nearly all” standard and it’s not necessarily a median consumer standard either. If it’s to be a sort of risk-based approach, under which a lower incidence of confusion might justify finding a violation where the confusion causes a great deal of harm, we might still want some clarity. What’s the model or the legal standard? And in this case, what percentage of consumers were alleged to be misled, and what was the average or median harm suffered by those consumers?
None of that’s at all clear in the FTC’s amended complaint or in the redacted, publicly available version of the FTC’s expert witness report. What’s not there seems directly relevant to the question of whether the FTC’s consumer protection case was sound, anti-tech nonsense, or something in between. More than that, what’s missing robs the FTC’s multi-year allocation of staff resources, its legal complaint, and the settlement itself of guidance value.
Without a full trial on the merits, we’re left without a precedent. But a settled case can still provide significant insight into an agency’s perspective on something as potentially important, but fundamentally vague, as “dark patterns” and the standards and methods the agency employed in determining what sort of online-marketing practices violate Section 5 or ROSCA.
To be clear, the FTC’s stipulated order does include some concrete requirements, such as these:
If the Negative Option Feature has an auto-renew feature, indicate that feature by using the word “renews” (or a similar word) on all sign-up pages; and E. Always disclose the price and autorenewal feature on the sign up page for the Negative Option Feature.
And perhaps, in the agency’s view, there’s some guidance implied there about minimum standards, if not all that much. It also leaves the residual question of why these technical changes—or something like them—could not have been negotiated much earlier in the process.
I can see the impact of the settlement on Amazon’s bottom line ($2.5 billion plus other compliance costs, with some of that perhaps passed on to Amazon customers down the line). I can see the impact on the government’s coffers, with only $1.5 billion going into the consumer fund and the remaining $1 billion going to the government. I can even see the impact on those consumers who receive compensation from the fund, although the initial round of payouts will be capped at $51 per consumer. That is something, to be sure, but it underscores the question what sort of median harm was done to those consumers alleged to have been misled by Amazon. And it leaves aside the question of how many consumers may be “misled” by the “consumer fund distribution process” that’s to be overseen by a court-appointed supervisor. But I cannot see all that much guidance as to what counts as a “dark pattern” or an acceptable user interface, or as to an economically sound foundation for such guidance. Perhaps it’s forthcoming?
There’s one more nagging issue about the FTC’s expert economics report. In a section on “the economic literature on consumer choice and choice architecture,” Mahoney summarizes the behavioral-economics research as follows:
Substantial economic research supports the proposition that consumers may take actions that do not reflect their underlying preferences due to a combination of their cognitive constraints and how decisions are framed for them, with the latter often being referred to as the “choice architecture” they face. One important implication of this research for the current matter is that even seemingly small changes in how Amazon presents its Prime enrollment and cancellation options to consumers can potentially have a substantial impact on the decisions those consumers make, and can induce them to make decisions that are against their own preferences. (internal citations omitted)
Perhaps that’s right, but that gloss seems to beg some hard questions about how it is we determine consumers’ “underlying preferences.” Consider, for example, two stages in the “Iliad flow” highlighted in the FTC’s complaint. Consumers looking into cancellation were first provided with a screen listing certain benefits of Prime membership that they might want to consider (as well as a button to continue the cancellation process). That’s marketing, to be sure, but the information might have been useful to some of those consumers, and there was no allegation that the features themselves were misrepresented.
Next, consumers continuing in their “quest” might encounter the offer of a discount (as well as a button to continue the cancellation process). This, too, was marketing, but marketing that might have been appealing to a rational, well-informed, and self-interested consumer. It’s really a filter to identify price-sensitive consumers and, thereby, to enable a sort of price differentiation that’s common across many goods and services: offer a lower price to price-sensitive consumers.
Does a choice architecture that includes such choices steer (or nudge, or coerce, or deceive) consumers away from their “underlying” or true preferences? That’s hardly obvious. Certainly, we can imagine a consumer with a demand curve such that she enters into the cancellation process preferring to cancel her Prime membership at the standard price but not at the discounted price. And the whole idea in the gloss on the behavioral-economics literature is that consumers’ “cognitive limitations” matter; decisions may be the result of things (say, features of Prime) that consumers did not know, or did not recall, when beginning to cancel their subscription. Providing such information does not obviously steer consumers away from their true preferences.
Of course, we can ask about the details. As I said above, we can imagine cancellation processes so onerous or confusing as to be practically impossible; and we can imagine processes that are so onerous or confusing for all consumers, most consumers, the median consumer, or a significant and identifiable tranche of consumers. I don’t have a countervailing economic analysis on offer, or a good strong argument that Amazon should have done precisely what they did, and not something else, at any time certain. But these really do seem like some very basic questions about the assumptions and assessments made by the agency that has the burden of making a case that actual conduct violates the law.
And we can ask about the efficacy and efficiency of the remedy. Independent surveys do, after all, indicate extremely high degrees of consumer satisfaction with Amazon. That does not give Amazon or anyone else a free pass on the FTC Act, ROSCA, or any other law, but it does call to mind that remedies may impinge on procompetitive offerings, and that efficient remedies ought to be tethered to demonstrable consumer harm, lest they serve more as punishment, or a tax, than as remediation.
FTC Chairman Andrew Ferguson has blasted EU digital regulations as “taxes on American firms,” and not without reason. EU digital regulations are not just that, but it’s hard to explain the contours of “gatekeeper” status under the Digital Markets Act (DMA)—or the European Commission’s designations of gatekeepers—without supposing that the process was reverse engineered to target large U.S.-based tech firms especially. And, as the old saw goes, what’s good for the goose is good for U.S. antitrust-enforcement agencies.
