Japan Built a Fake Streaming Site to Measure What Dark Patterns Actually Do
Most dark pattern research asks people what they think. Japan's Consumer Affairs Agency built a fake video streaming service instead, put 1,600 people through signing up and cancelling, and recorded what they actually clicked.
The results were published in June 2026. They are more modest than the headlines around them, and more useful.
What they built
There were two studies. The first was a recognition survey: 1,200 people from their teens to their seventies, shown mocked up screens for 28 different dark patterns, asked how often they see something like this. It ran over the new year holiday, from 26 December 2025 to 4 January 2026.
Two numbers came out of it. 76.2% had seen at least one of the 28. Asked whether they had personally run into one while shopping online in the last twelve months, 37.5% said yes.
The second study is the one I care about. They built a working fake site for a subscription streaming service and gave 1,600 people a task. Imagine you want this for about three months. Choose between a premium plan, a standard plan, an ad supported standard plan, or don't sign up. Then, later, imagine the free trial is ending and you want out. Cancel, or stay.
Different groups got different versions of the site. The control group saw everything disclosed the whole way through, including the cancellation fee. The other groups got specific manipulations, one at a time, with the behaviour logged rather than self reported.
The signup result, with the caveat first
Against the clean control, one manipulation moved the plan choice. Hiding the subscription terms, presenting the free period prominently and keeping the recurring charge and cancellation fee out of sight, raised the share choosing the most expensive premium plan from 1.0% to 6.5%.
Each of those groups had 200 people. So we are looking at roughly two people versus roughly thirteen. I am not going to build a theory on eleven people, and the agency does not either. It reports a tendency and says the combination effects need further work.
The second manipulation, hiding the account management fee until the final confirmation screen, barely moved plan choice at all. It moved something else. When people were asked afterwards what the monthly account management fee on their chosen plan was, the correct answer rate was lower in both dark pattern groups than in the control.
That is the finding I would put on a wall. The pattern did not necessarily change what people bought. It changed whether they knew what they had bought. Those are different harms and only one of them shows up in your conversion numbers.
The cancellation result is the clean one
The cancellation phase had four conditions, 400 people each, with obstruction escalating in defined steps.
Control was a normal cancellation flow. Weak obstruction made the cancellation page hard to find and added a confirmation popup. Medium added a false hierarchy on the final screen, visually emphasising the keep my subscription button. Strong added repeated popups that kept interrupting.
The stronger the obstruction, the lower the share of people who managed to cancel. Monotonically, across the whole ladder.
Nobody who has built a cancellation flow will be surprised. What is new is that it is now measured, in Japanese, by the Japanese regulator, with behavioural data rather than opinion. The agency notes that verifying subscription cancellation dark patterns this way is a new undertaking and that few comparable studies exist here.
If you run a subscription business in Japan, that sentence is the one to read twice. Somebody is now building the instrument that measures your cancellation flow.
Recognition is not protection
Put the two studies next to each other and you get the thing I find genuinely uncomfortable.
Three quarters of people recognise these patterns. They know the countdown timer is fake. They have seen the only 2 left in stock badge enough times to be bored by it. And then, given a task and a deadline and a screen that buries the cancellation link, the obstruction still works.
Knowing about a pattern does not give you the patience to fight it. Design does not persuade you. It taxes you, and most people pay the tax rather than spend twenty minutes proving a point to a website.
Which is why I have stopped finding the ethics argument interesting. Nobody who ships a three step cancellation gauntlet believes it is honest. They believe it is legal and it hits the retention target.
The money
Separately, in August 2026, the Dark Pattern Countermeasures Association published its own survey. Among general consumers aged 21 to 64, 29.0% reported a financial loss in the past year from an unintended purchase, charge or contract, essentially flat against 30.2% in 2024.
Their estimate of total annual harm in Japan came out between roughly 2.4 and 3.2 trillion yen, about double their 2024 estimate of 1.06 to 1.68 trillion. The rate did not move. The scale of what people lose did.
Among those who reported harm, about 32.3% lost 50,000 yen or more and 6.8% lost more than 500,000 yen. That is an estimate built on a survey of 500 people, so treat the trillions as an order of magnitude rather than an accounting figure. Even discounted heavily, it is a large number for a problem that lives entirely in interface decisions.
What I would change on Monday
Four things, and none of them require a redesign.
Find your cancellation path and count the taps from the account screen. Not from the help centre. From where a person who wants out would actually start. If it is more than three, someone chose that.
Put every recurring charge on the same screen as the button that commits to it. Not one screen earlier and not in a link. The account management fee finding says people will confidently misremember a price they were shown late.
Look at your final confirmation screen and ask which button a stranger would press by accident. If the visual weight is on the option that makes you money, that is the false hierarchy condition from the experiment, and it is now a documented manipulation rather than a style choice.
Then run the ugliest test there is. Take five people who have never seen your product, tell them to subscribe and then cancel, and watch without helping. You will learn more in an hour than from any amount of survey data, including this one.
Why I think this matters more in Japan
Japanese consumers complain less. Not because they mind less, but because the social cost of complaining is higher and the expectation of a company behaving properly is higher too. So the feedback loop that would tell a Western product team something is wrong is quieter here, and the damage accumulates without ever reaching a dashboard.
A regulator running controlled experiments is what fills that gap. This study is small, honest about its limits, and clearly the first of several. I would rather be ahead of it than explaining myself to it.
Sources
- Consumer Affairs Agency, Strategic Headquarters for Frontier Research, Webページ等におけるいわゆるダークパターンに対する消費者意識調査について (概要版), June 2026, based on the report by NTT Data Institute of Management Consulting. Survey 1: 1,200 respondents, 28 dark pattern types, fielded 26 December 2025 to 4 January 2026, 76.2% recognition and 37.5% experience in the previous twelve months. Survey 2: 1,600 respondents, fielded 20 to 25 February 2026, contract phase groups of 200 and cancellation phase groups of 400.
- Dark Pattern Countermeasures Association, 2026年ダークパターン被害調査, 18 August 2026. General consumers aged 21 to 64, 500 respondents, fielded 26 May to 8 June 2026. 29.0% reporting loss against 30.2% in 2024, estimated annual harm of roughly 2.4 to 3.2 trillion yen against roughly 1.06 to 1.68 trillion in 2024.
- OECD's definition of dark patterns, cited in the Consumer Affairs Agency document as the working definition for the study.