I spent twenty-two minutes canceling a subscription last winter. Signing up had taken about forty seconds.
The exit ran four screens deep. One asked whether I was sure. The next offered a discount I hadn’t requested. The third listed everything I’d be giving up. The last one gave me two choices: a large, bright button reading “Keep my benefits,” and beneath it, in small gray text, a link that said something close to “No thanks, I don’t mind paying more.”
That gray link is what I want to talk about. Somebody wrote that sentence. Somebody approved it in a review. Somebody almost certainly tested it, found it lifted retention by a point or two, and put the result in a deck. Everyone involved was doing their job well.
The Catalog
Harry Brignull, a British UX designer, gave this stuff a name in 2010. He put up a site cataloguing the tricks and listed twelve of them, then spent the next decade watching the vocabulary he’d invented get adopted by regulators. His 2023 book, Deceptive Patterns, is the closest thing the field has to a reference text.
The named ones you’ll recognize on sight:
Confirmshaming. The gray link above. You’re not blocked from leaving — you’re just made to insult yourself on the way out.
Manufactured urgency. Countdown timers that reset on refresh. Low-stock warnings on infinite digital inventory. “Fourteen people are viewing this right now.” A 2019 audit out of Princeton, led by Arunesh Mathur, crawled roughly 11,000 shopping sites and found these three — activity notifications, low-stock messages, countdown timers — were the most widely deployed patterns of all.
Roach motel. Trivial to enter, laborious to leave. My twenty-two minutes.
Pre-selected checkboxes. A signature you didn’t write.
Drip pricing. The number you saw on the product page and the number you’re charged are separated by four screens of fees.
The Princeton crawl turned up 1,818 pattern instances across 1,254 sites, sorted into fifteen types. Here’s the part that stuck with me: the researchers also found 22 third-party companies selling these techniques as ready-made services. You don’t have to invent a fake countdown timer. You can buy one. There’s a vendor, an integration, probably a customer success manager.
And that number is a floor, not a ceiling. The crawler read text on product, cart, and checkout pages. It couldn’t assess visual hierarchy, couldn’t judge which button was bigger, and never reached account cancellation flows — where the ugliest work happens.
Why Any of This Works
Because it does work, in the short run. That’s the uncomfortable premise underneath everything else.
The FTC’s case against Amazon over Prime enrollment produced a detail I think about often. According to an attorney on the trial team, whenever Amazon removed the friction from its cancellation process, Prime sign-ups fell. So the changes were rolled back. The friction wasn’t an oversight or a legacy bug. It was load-bearing.
Three things are being exploited, and none of them are stupidity.
Inertia, first. Most people don’t cancel things because canceling requires deciding, and deciding is work. A monthly charge below some personal threshold can survive for years on the strength of nobody wanting to deal with it on a Tuesday evening.
Decision fatigue, second. By the fourth screen you’re no longer evaluating the offer. You’re looking for whichever button ends the interaction. Interface designers know exactly which one that is, because they chose its color.
And shame. Legal scholars Jamie Luguri and Lior Strahilevitz ran controlled experiments showing that even mild manipulation measurably increases acceptance of dubious offers — and that the effect falls hardest on less-educated participants. Which reframes the whole thing. These techniques don’t fail on sophisticated users and succeed on careless ones. They sort your customers by how much bandwidth they had that day, and they extract the most from the people with the least.
A dark pattern is a loan taken against your customers’ goodwill. The revenue lands this quarter. The repayment schedule shows up on a dashboard nobody has built yet.
Why It’s a Losing Position Over Time
On September 25, 2025, Amazon settled with the FTC for $2.5 billion — a billion in civil penalties and $1.5 billion in refunds to roughly 35 million customers. Three days into trial. The complaint centered on Prime sign-up flows and a cancellation process Amazon had nicknamed, internally, the “Iliad Flow,” after Homer’s poem about a ten-year siege. Four pages, six clicks, fifteen options. Internal documents surfaced employees calling the practice “a bit of a shady world.”
They named it after a war of attrition. In writing. And that document became evidence.
The US regulatory picture is genuinely messy right now, so it’s worth being precise. The FTC’s click-to-cancel rule was vacated in full by the Eighth Circuit on July 8, 2025, on procedural grounds — six days before its main requirements took effect. But the Restore Online Shoppers’ Confidence Act still applies, Section 5 of the FTC Act still applies, state auto-renewal laws in California, New York, and elsewhere still apply, and the FTC filed a fresh advance notice of rulemaking on January 30, 2026. Anyone reading the vacatur as permission is reading it wrong.
Europe is moving in the same direction on a longer timeline. The Digital Services Act already prohibits specific named patterns. Beyond that, the Commission ran a public consultation on a Digital Fairness Act from July to October 2025, with a formal proposal expected in the fourth quarter of 2026 — targeting dark patterns, addictive design, and personalization that exploits vulnerability. That’s a proposal, not a law, and it’ll take years to negotiate. The direction of travel is not ambiguous.
Regulation is the slow risk. The fast one is that this work is permanent and public. A cancellation flow can be screenshotted. An internal nickname can be subpoenaed. The teams whose names circulate on shortlists of top web design agencies generally won’t ship this material, and the reason isn’t moral superiority — it’s that their portfolio is their entire pipeline. Every project they’ve ever done is browsable. One confirmshaming button in a case study costs them more than it could ever earn a client. Reputation is a strange kind of asset: it disciplines behavior more reliably than most compliance departments.
Ask what happens if your checkout flow ends up in a slide someone presents at a conference. If that thought makes you uneasy, you already know the answer.
The Honest Version Performs Better Than You’d Guess
Every business goal a dark pattern serves has a legitimate route to it.
Want fewer cancellations? Ask why people leave, at the moment they leave, in one field, and then fix what they tell you. The information is worth more than the two points of retention you’d have squeezed out with a gray link.
Want higher checkout conversion? Show the total price on the first screen. Surprise fees don’t reduce abandonment — they relocate it to the last step, where you’ve already paid for the traffic.
Want the newsletter signup? Make it a real choice. A pre-checked box gives you a list full of people who will mark you as spam, which damages deliverability to the people who actually wanted you.
Easy cancellation is the counterintuitive one. Make leaving frictionless and some people will leave who wouldn’t have bothered. You’ll also get customers who come back, because the exit didn’t feel like an ambush. Some subscription businesses have made “cancel anytime, two clicks” an actual selling point. It works because it’s a costly signal — it’s expensive to say if you’re not confident in the product.
None of this is charity. It’s just the recognition that trust compounds and tricks don’t.
The strange thing about dark patterns is how rarely anyone chooses one on purpose. Nobody sets out to design a maze. It happens one A/B test at a time, each one showing a small lift, each one approved by someone reasonable who didn’t have the full picture in front of them.
Which means the question worth asking in your next design review isn’t whether anyone here is a villain. It’s whether anyone in the room is measuring the thing you’d lose.
