Win the click, lose the trust.
Interfaces crafted to trick people into buying, subscribing, or sharing more than they meant to. They lift a metric for a quarter, then cost the trust of the people who were tricked, and increasingly they draw fines.
Harry Brignull, a UK user experience designer, coined the term on 28 July 2010 when he registered darkpatterns.org, a pattern library meant to name and shame deceptive interfaces. The site has since moved to deceptive.design, and in 2023 he published the book Deceptive Patterns.
Research and regulators followed. A 2019 study by Arunesh Mathur and colleagues at Princeton crawled about 11,000 shopping sites and found 1,818 instances of dark patterns. In 2022 France’s CNIL fined Google and Facebook over cookie banners, and in 2025 Amazon settled the FTC’s case over Prime sign-up and cancellation for $2.5 billion, without admitting wrongdoing.
A dark pattern is an interface crafted to trick people into doing something they didn’t mean to do, such as buying, subscribing, or sharing data. It works by exploiting habits and biases, and it trades short-term conversion for long-term trust and, more and more often, legal risk.
Easy to get in, hard to get out. The FTC said Amazon’s Prime cancellation flow, known internally as Iliad, routed people through repeated pages of offers. Under EU pressure in 2022, Amazon made cancellation two clicks in Europe.
A low headline price, with fees and charges added step by step until checkout. By the time the real total appears, people have invested enough effort that many pay it anyway. Show the full price from the first screen.
A decline link worded to shame: No thanks, I don’t like saving money. It pressures people with guilt instead of persuading them with value. Word the decline option as plainly as the accept option.
A bright button for what the business wants and a faint link for what the user wants, or a consent box with a double negative. Software installers made it famous by bundling unrelated programs behind a prominent Accept button.
Named after Mark Zuckerberg: designs that get people to share more than they intended. In 2022 France’s CNIL fined Google €150 million and Facebook €60 million because accepting cookies took one click and refusing took several.
Countdown timers that reset, and Running out! labels on stock that isn’t running out. In 2024 the UK’s Competition and Markets Authority secured undertakings from Wowcher over exactly these claims. Show only real deadlines and real stock.
Advertise something free or cheap that most people can’t actually get, then steer them to a paid option. In 2022 Intuit agreed to a $141 million settlement over TurboTax ads for free filing that most filers didn’t qualify for.
The test for any flow: would people make the same choice if they fully understood it? Give yes and no equal weight, show the real total, and make leaving as easy as joining. Several jurisdictions now require exactly that.
AI can tailor copy, timing, and offers to each person. That makes persuasion more effective and manipulation cheaper and harder to see.
A model can generate thousands of variants of a message and learn which ones work on whom. Hold generated copy to the same rules as handwritten copy, and never optimize for conversion alone.
In 2024 Meta planned to train AI on Facebook and Instagram posts, with an opt-out reached through misleading emails, login redirects, and a form that asked for a reason. Refusing consent for AI training should be as easy as giving it.