The dark pattern taxonomy
A named set of interface tricks that get users to do what they did not intend.
dark pattern: An interface built to get an outcome the user would reject if it were stated plainly. are interface choices that work against the user's own intent, by hiding, misdirecting, pressuring or obstructing. Brignull's original taxonomy and the later academic ones give the categories names, which is what makes them reviewable. The test is intent plus effect: would the user object if the mechanism were explained to them.
How it shows up in software
The patterns cluster in the money and consent surfaces: checkout, upgrade, consent banners, cancellation, permission prompts. They usually arrive one at a time, each defended as a conversion win, and the taxonomy is what lets a reviewer see the accumulation as a class of harm rather than a series of small decisions.
Using it well
- Run a pattern audit against the taxonomy. Walk signup, upgrade, consent, permissions, refund and cancellation, and name every instance you find by its category.
- Check what your third-party widgets do. Consent management platforms, urgency plugins and review widgets ship patterns you did not write and are still responsible for.
- Add a removal line to the roadmap with an owner and a date, and accept the conversion loss as a correction rather than a regression.
- Put the taxonomy into design review as a checklist, so new instances are caught before launch rather than found in an audit.
Where it turns manipulative
- These patterns are dishonest by construction. They produce outcomes the user would reject if asked, which is the definition of manipulation rather than persuasion.
- They are increasingly illegal. The EU Digital Services Act Article 25 prohibits them on platforms, the GDPR invalidates consent obtained through them, California's CPRA states that consent obtained by dark patterns is not consent, and the FTC has brought cases under Section 5 and ROSCA.
- The internal defence that the numbers improved is not a defence. Revenue from a user who did not intend to buy is a liability recorded as income, and it returns as refunds, chargebacks, complaints and enforcement.
Where you have seen it
Consent banner obstruction
Accept all is one prominent button; rejecting requires opening a preferences panel and toggling individual vendors. Widely documented across the CMP market by Nouwens et al. (2020) and the Norwegian Consumer Council.
Checkout sneaking
An optional insurance, donation or warranty is added to the basket by default and only visible in the line-item total.
Disguised advertisement
A download button styled to match the page's real control, so the click goes somewhere the user did not choose.
What the research says
- Brignull, 2010 onward (deceptive.design)Well evidenced
Coined the term and built the working taxonomy still in use: sneaking, urgency, misdirection, social proof: People look at what others did to decide what is correct here., scarcity, obstruction, forced action.
Practitioner-built rather than experimental. Its value is a shared vocabulary that survived into regulation.
- Mathur et al., 2019 (CSCW)Well evidenced
An automated crawl of roughly 11,000 shopping sites found over 1,800 instances of dark patterns across more than a thousand sites, and identified the third-party vendors selling several of them as a service.
The vendor finding matters most: many of these patterns are bought as a widget, not designed in house, which changes where to intervene.
- Luguri and Strahilevitz, 2021 (Journal of Legal Analysis)Well evidenced
Two large experiments found mild dark patterns more than doubled acceptance of an unwanted subscription, and aggressive ones raised it further but produced measurable user backlash.
Randomised and preregistered. It also found less educated participants were disproportionately affected by the milder patterns.
Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.