Dark patterns are interface designs that trick, pressure or trap people into choices they would not otherwise make, such as fake countdown timers, pre-ticked add-ons, hidden fees and cancellation mazes.
They can lift a short-term metric, but our database shows the same tactics often lose once you measure revenue per visitor, refunds or repeat purchases.
Regulators in the US and EU also treat many of them as illegal.
The term comes from UX designer Harry Brignull, who registered darkpatterns.org in 2010 to name and shame deceptive interfaces.
The site now lives at deceptive.design, and Brignull's newer term for them is "deceptive patterns."
This page covers common examples, what the research and the law say, and what real A/B tests show when stores and apps swap a manipulative pattern for an honest one.
It sits alongside our guide to conversion psychology.
Dark patterns examples
The largest crawl of shopping sites, "Dark Patterns at Scale" by Mathur et al. (2019), checked about 53,000 product pages on about 11,000 shopping sites.
It found 1,818 dark pattern instances of 15 types in 7 categories, and 183 sites using outright deceptive ones.
It also found 22 third-party vendors selling these patterns as ready-made plugins.
The patterns ecommerce and subscription teams run into most often:
| Pattern | What it looks like | Honest version |
|---|---|---|
| Fake urgency | A countdown timer that resets on reload, or a "sale ends tonight" that never ends | A real deadline, shown the same way to everyone |
| Fake scarcity | "Only 3 left" when stock is plentiful | Real inventory counts, or nothing |
| Sneak into basket | Insurance, protection or a donation pre-ticked in the cart | The add-on offered clearly and left unticked |
| Hidden costs / drip pricing | Fees that only appear at the last checkout step | The full price shown up front |
| Forced continuity | A free trial that turns into a paid plan with no reminder | A clear billing date and a reminder before the charge |
| Roach motel / obstruction | Signing up takes one click; canceling takes a phone call | Canceling is as easy as signing up |
| Confirmshaming | "No thanks, I don't like saving money" | A neutral "No thanks" |
| Nagging / no exit | Popups or paywalls with no close button | A visible, working close option |
Several of these overlap with tactics our database files under normal names, such as countdown timers, low-stock messages, upsells and popups.
None of those is a dark pattern in itself.
A timer with a real deadline is information. A timer that resets is a lie.
Deceptive design patterns: what research and the law say
Do they work? The lab evidence
The cleanest experiment on dark patterns is Luguri and Strahilevitz, "Shining a Light on Dark Patterns" (Journal of Legal Analysis, 2021).
A representative sample of US adults was told they had been enrolled in an identity-theft protection plan, and the design of the opt-out screens was varied at random.
In the working-paper version, the results were:
| Condition | Accepted the plan | Respondents |
|---|---|---|
| Control (simple accept/decline) | 11.3% | 73 of 644 |
| Mild dark patterns | 25.8% | 155 of 600 |
| Aggressive dark patterns | 41.9% | 217 of 518 |
So yes, in a lab, they work. The authors also found that aggressive patterns caused a backlash among participants while mild ones did not, and that less-educated participants were more susceptible to the mild ones.
That is the uncomfortable part: the subtle versions are the ones people don't notice.
Keep the limits in mind. This was one survey experiment about one unwanted product, with no real money at stake.
It measured whether people stayed enrolled, not whether they later asked for refunds, complained or stopped buying.
What regulators say
The US Federal Trade Commission's 2022 staff report "Bringing Dark Patterns to Light" defines them as design practices that trick or manipulate users into choices they would not otherwise make and that may cause harm.
The report groups them into four types: inducing false beliefs, hiding or delaying information, obstructing user action, and subverting privacy choices.
Enforcement has followed. In September 2025 the FTC announced a $2.5 billion settlement with Amazon: a $1 billion civil penalty plus $1.5 billion in refunds.
The FTC had alleged that Amazon used confusing interfaces to enroll people in Prime without clear consent and made canceling hard on purpose.
The FTC's broader "click-to-cancel" rule was vacated by the Eighth Circuit in July 2025 on procedural grounds.
In March 2026 the agency opened a new rulemaking on negative-option plans.
Existing law, including the Restore Online Shoppers' Confidence Act used in the Amazon case, still applies.
In the EU, Article 25 of the Digital Services Act bars online platforms from designing interfaces "in a way that deceives or manipulates" users or "materially distorts or impairs" their ability to make free and informed decisions.
This is not legal advice. If your store sells into the EU or runs subscriptions, have a lawyer review your checkout and cancellation flows.
What A/B tests show when you remove the trick
Lab studies measure one choice. A/B tests on live stores measure money, and that's where many manipulative tactics fall apart.
Here is what our database shows when teams test the honest version against the pushy one.
Pre-ticked add-ons and defaults
Oliver Kenyon's team unticked a pre-ticked accidental damage cover in the cart of a store selling cameras and sold it explicitly instead.
Fewer people bought the cover, but more people bought cameras, and the test was called a win.
The pre-ticked box was quietly costing main-product sales.
Convertibles tested a shipping protection toggle set to off by default on an electronics accessories store.
Conversion went from 1.53% to 1.90% (+23.7%) and revenue per visitor rose 16.4%, at 97% probability to beat control.
Average order value fell 6%, partly from fewer protection fees, and among cart openers the on-by-default version edged ahead.
That caveat matters: the honest version won overall, not on every slice.
Defaults are not automatically dark. Blend Commerce preselected the subscription option on Jackson's product page and saw a +50% subscription rate.
Dave Diederen's team pre-checked a newly introduced subscription option and saw subscription share of orders rise 1,295%, while revenue per visitor fell 9.17% upfront.
A visible, clearly labeled default is a default effect.
It becomes a dark pattern when the shopper can't tell they're signing up for recurring charges.
Clear disclosure of recurring billing
Vasta tested making a supplement's auto-refill terms obvious: a bold "every 30-day auto-refill, cancel anytime" bullet, icons by the price and billing micro-copy under the button.
Conversion rose 4.35% and revenue per visitor rose 6.53% at 95% confidence over two weeks.
Telling people they'd be billed monthly didn't scare them off.
In apps, Marcus Burke's team at Blinkist tested an "ethical" paywall showing a trial timeline and the date of the first charge.
Trial opt-ins rose significantly and trial cancellations fell 4%.
Paywalls and popups you can't close
BoldVoice ran two tests that look like wins on the dashboard.
A hard paywall with the close button removed and an unclosable discount offer after dismiss both increased trial starts.
In both, refunds rose by more than the gain.
BoldVoice now judges tests on net revenue per exposed user, measured on matured cohorts.
Pressure at the wrong moment
Kasper Dolk's team ran a post-purchase survey where Dutch shoppers said the checkout timer felt like too much pressure.
Hiding the checkout countdown timer lifted conversion 25.61% and revenue per visitor 26.18%.
Dave Diederen's team saw a similar pattern: adding a countdown timer for a non-US market cut revenue per visitor 35.61% and conversion 17.35%.
The team reports the same pattern across more than 20 brands in markets outside the US.
Ruben's team tested timers across several brands: about +10% conversion for a low-priced brand that had really sold out before, no clear lift for a higher-priced brand after four weeks, and -9% for another. Urgency worked where it was true.
Real scarcity versus scary scarcity
Clean Commit added a "stock is running low" announcement bar to product pages.
Conversion fell 17.5% and revenue per visitor fell 11.3%.
It was a short test (5,161 visitors, 6 days), so treat it as a warning, not proof.
Dave Diederen's team found that "In Stock" beat "Almost Out of Stock" with +16.78% revenue per visitor.
Meanwhile Oliver Kenyon's team used authentic scarcity on product pages for a brand built on limited runs, with copy that matched real supply, and forecast +$46,900 per month.
@oliverkenyon
Vasta
Convertibles
@kasper_dolk
Clean CommitThe pattern across these tests: honest urgency, clear defaults and plain billing terms can lift sales.
Fabricated pressure and traps tend to lose once you look past the first click.
These are individual tests on specific stores, though, not laws of nature.
For more on why tactics flip between sites, see our guide to scarcity and urgency.
A/B testing ethics: how to test without crossing the line
Experimentation itself is not unethical. Showing two honest versions of a page and keeping the better one is just product work.
The ethical problems start when the variant deceives people, or when the metric you're optimizing hides the harm.
Rules of thumb before you launch a variant
- Would it still work if the shopper understood it? If a tactic only works because people miss it, such as a pre-ticked fee or a hidden renewal, it's a trick.
- Is every claim true? Stock counts, deadlines, "people viewing now" and review numbers must reflect reality, for every visitor.
- Is leaving as easy as joining? Close buttons, opt-outs and cancellation should take no more effort than signing up.
- Would you show it to a regulator? If the FTC report's four categories describe your variant, don't ship it.
Measure the harm, not just the click
Most dark patterns "win" because the test stops at the wrong metric.
Trial starts go up while refunds go up more.
Add-on attach rate goes up while main-product sales fall.
Fix that with guardrail metrics: refunds, chargebacks, cancellation rate, support tickets, unsubscribes and repeat purchase rate.
For subscriptions and apps, judge tests on matured cohorts, as BoldVoice now does, and consider holdout groups to catch long-term effects.
Pick a primary metric close to profit, like revenue per visitor or net revenue per user, rather than a single click.
When a manipulative variant "wins"
Sometimes a pushy variant wins on every metric you measured.
Don't ship it if it deceives people. The short window of an A/B test rarely captures the trust cost, the review-site complaints or the legal risk, and as the Amazon case shows, that risk can be large.
Instead, look for the honest version of the same idea: a real deadline instead of a fake one, a clearly offered add-on instead of a pre-ticked one, a reminder before billing instead of a surprise.
Our tests suggest the honest version often wins anyway.
Browse more results, including tests that lost, in the full test database.