The Roach Motel: Easy In, Hard Out - and What the Measurements Show

The roach motel is the interface pattern where getting in takes one click and getting out takes eleven, a phone call, and a form asking why you are leaving.

It is one of the oldest named patterns in the field - the term comes from Harry Brignull's taxonomy, which he began in July 2010 under the label "dark patterns," now more commonly called deceptive or manipulative patterns. And it is one of the few where researchers have gone out and actually measured the thing rather than cataloguing it.

What they found is real, and smaller and more specific than the folklore. It is also, in one important respect, much less well established than it is usually presented.

Somebody actually counted

Sheil and colleagues (CHI 2024) did the obvious experiment that nobody had done: they signed up, then tried to leave.

Working across 34 news websites in four countries - the United States, United Kingdom, Germany and the Netherlands - and accessing them from five locations, they took out 67 subscriptions and cancelled 63 of them online, recording roughly a hundred screen captures of the process.

The core measurement, for their American personas on American sites:

Action	Average clicks
Subscribing	3.5
Cancelling	5.8

For European media accessed from Europe, the same figures were 3.5 to subscribe and 4.9 to cancel. Comparing cancellation directly: 4.9 clicks in Europe against 6.2 in the US.

The friction was not only clicks.

• Exit surveys. In the US, almost every site displayed one, and 82.7% made answering mandatory. In Europe, 67.1% asked for reasons, with about 72.5% of those making it compulsory.
• Retention offers. 77.8% of US sites and 51.9% of European sites offered a discount to make you stay.
• Phone only. Four of the 67 subscriptions could not be cancelled online at all - you had to call a human.

The German cases are the sharpest: German law requires a two-click termination option, and only three of eight German sites met it.

What causes what

Counting clicks establishes that the asymmetry exists. It does not establish that the asymmetry keeps people subscribed. For that you need an experiment, and there is essentially one.

Luguri and Strahilevitz (2021), in the Journal of Legal Analysis, ran two large randomised studies on US census-weighted samples. In the first (n = 1,963), they offered an unwanted data-protection subscription through interfaces of escalating manipulativeness:

Interface	Accepted
Control	11.3%
Mild manipulation	25.8%
Aggressive manipulation	41.9%

The second study (n = 3,777) separated the individual techniques. Obstruction - making the decline path require additional work - produced a significant increase in acceptance, χ²(1, N = 2,065) = 14.84, p < .001. Making acceptance the default was weaker but still significant, χ²(1, N = 2,145) = 4.02, p = .045.

For scale, the strongest patterns in the same study were hidden information (30.1% vs 14.8%) and a trick question (33.4% vs 19.2%). Obstruction sits in the upper group.

The most recent systematic review (Schaffner, Heysen and Chetty, CHI 2026) reports that obstruction and "adding steps" have been tested 45 times across the literature and reached significance in 7 of 10 of the head-to-head experiments it categorised. So the effect replicates reasonably well.

The money

There is one large-scale estimate of what this friction is worth, and it comes with a caveat.

Einav, Klopack and Mahoney analysed payment-card network data covering 635,021 account-service pairs across ten subscription services between 2017 and 2021. Their design is clever: when a card is replaced - expiry, loss, fraud - the subscriber is forced to actively re-authorise, which functions as a quasi-random shock to attention.

Cancellation rates were four times higher in card-replacement months: 0.08 against 0.02 monthly. Their modelling attributes an average revenue increase of 87% to inertia, ranging from 14% to 230% across services, and estimates that requiring active renewal every six months would cut that inertia revenue by 45-48%.

The caveat: this is a working paper, and we could not confirm it has completed peer review. The figures are reported here as its authors state them, with that status attached.

What the pattern is called, and why the name matters

"Roach motel" is Brignull's original label, borrowed from an American cockroach trap advertised on the promise that the insects check in and do not check out. In the current research vocabulary it usually sits under the more neutral heading of obstruction, sometimes "adding steps," within the family of manipulative or deceptive patterns.

The renaming is not pedantry. "Roach motel" describes an intent - someone built a trap. "Obstruction" describes a measurable property - the decline path requires more actions than the accept path. Only the second can be counted, and counting is what turned this from a complaint into evidence.

It also makes the audit findings harder to dismiss. Sheil and colleagues did not have to establish that anyone intended to trap subscribers. They only had to count clicks, and the counts came out asymmetric in every country they looked at.

What has not been shown

This is the part that gets skipped, and it matters.

No experiment has randomised cancellation difficulty and measured whether people give up. Luguri and Strahilevitz manipulated obstruction on the way in - declining an offer. Sheil and colleagues measured how hard it is to get out, but ran an audit, not an experiment. The bridge between them is an inference, and a reasonable one, but it is an inference.

Mathur and colleagues (2019) is not evidence about cancellation, despite being cited that way constantly. That crawl of about 11,000 shopping sites found 1,818 instances of 15 dark-pattern types on 183 sites, supplied in part by 22 third-party vendors - a genuinely important measurement. But its stated methodology covers product pages, cart pages and checkout pages only, and the paper lists this as an explicit limitation. It never followed a cancellation flow. We are not going to borrow it.

Gray and colleagues' 2018 taxonomy is a taxonomy. It gave "obstruction" its name and its place in the literature. It did not measure an effect.

Nothing here is legal advice. Rules on cancellation differ by country and change; the German two-click finding above is a measurement of compliance in 2024, not a statement of your rights today.

Where the boundary sits

There is a related pattern that is often mixed in with this one and should not be: confirmshaming, the decline button worded to make you feel bad about clicking it. They are different mechanisms with different evidence, and Luguri and Strahilevitz tested them separately in the same study.

Obstruction works by raising the cost of leaving. Confirmshaming works by raising the discomfort of saying no. Obstruction was the stronger effect. That distinction is covered separately in the article on confirmshaming, which publishes after this one; the link will be added once that page is live.

What this actually tells you

Three things worth carrying.

The asymmetry is measured, not folklore. Roughly 1.7 extra clicks in Europe and 2.7 in the US, plus a mandatory exit survey most of the time and a retention discount more often than not. That is not enormous per cancellation. Multiplied across a subscriber base and a year, it is the entire business model of several companies.

The strongest evidence points at attention, not resistance. The card-replacement finding is the most striking result in this area precisely because it is not about difficulty at all. Cancellations quadrupled when people were merely forced to notice they were subscribed. That suggests most of what keeps subscriptions alive is not that leaving is hard - it is that leaving never comes to mind.

Which makes the practical move a calendar entry, not a fight. If the binding constraint is attention rather than friction, then the intervention that matches the evidence is the one that recreates the card-replacement shock deliberately: a recurring reminder to look at what you are paying for. The five extra clicks are annoying. They are not what is costing you money.

The roach motel is one of a family of interface patterns that shape decisions by structure rather than argument - see the persuasive design guide for the wider set, should you accept cookies for the same architecture applied to consent, and your phone isn't the problem, its defaults are for what a pre-supplied answer does.