Dark pattern
Definition
A dark pattern — increasingly called deceptive design — is a deliberate design choice that leads users to do something that serves the service provider's interest rather than their own. Intent is central to the definition: this is not clumsy ergonomics, but manipulation designed as such.
The goals vary: making people spend more, keeping them on a service longer, or obtaining acceptance of terms or tracking they would have declined had the choice been presented plainly.
The term was popularised in 2010 by British user experience specialist Harry Brignull, who produced the first classification. Regulators and recent literature increasingly prefer "deceptive design" as a clearer, less loaded phrase.
Typology
Classifications vary, but a stable core of families recurs.
Concealment
- Hidden costs: service, handling or delivery fees revealed only at the final checkout step.
- Drip pricing: the announced price becomes complete only after several screens.
- Sneak into basket: an option, insurance or subscription pre-added that the user must remember to remove.
Manufactured urgency and scarcity
- Countdown timers that reset, "only 2 left" or "14 people are viewing this page" claims that cannot be verified.
- Exploitation of scarcity bias and fear of missing out (FOMO).
Misdirection and shaming
- Misdirection: a visual hierarchy that promotes the option favouring the service and erases the other — a coloured button against a tiny grey link.
- Confirmshaming: phrasing the decline option so as to induce guilt ("No thanks, I'd rather pay full price").
- Trick questions: double negatives, checkboxes whose meaning flips from one line to the next.
Obstruction
- Roach motel: subscribe in two clicks, cancel in eight steps, a phone call and a letter. This is the pattern most explicitly targeted by regulators.
- Nagging: repeated prompts — notifications, modal windows — until consent is given.
Coercion and privacy
- Forced action: mandatory account creation for a trivial task, permissions requested with no bearing on the feature.
- Privacy Zuckering: design that leads users to share more personal data than intended, with protective settings buried and off by default.
- Preselection: pre-ticked boxes standing in for consent.
- Disguised ads: commercial content presented as editorial material or as an interface element.
Mechanisms exploited
Dark patterns create no new behaviour; they exploit well documented cognitive biases.
- Scarcity bias and loss aversion: the fear of losing an opportunity weighs more than the prospect of an equivalent gain.
- Anchoring: a high price shown first makes the next one look reasonable.
- Cost of effort: beyond a few steps users give up — precisely the calculation behind obstruction.
- Default effect: the pre-set option is kept in the vast majority of cases, whatever it contains.
- Attentional load: a dense, pressing interface degrades decision quality, which is the domain of the attention economy.
Consent banners
Cookie banners are the most common observation ground: a prominent "Accept all" button against a refusal buried in a submenu, loaded wording, re-display on every visit. Since its 2022 decisions, the French data protection authority (CNIL) has applied a simple requirement: refusing must be as easy as accepting, with a visible refusal button at the first level.
Regulatory framework
- GDPR: consent must be freely given, specific, informed and unambiguous. A pre-ticked box, a costly refusal path or consent obtained through fatigue do not meet that bar.
- Digital Services Act (DSA), Article 25: interfaces that deceive, manipulate or otherwise impair users' ability to make free and informed decisions are prohibited. In December 2025 the European Commission issued its first DSA fine — €120 million against the platform X, in part for the deceptive design of its verification badge.
- Digital Markets Act (DMA): governs gatekeepers' interfaces, notably choice screens and cross-service consent.
- Consumer law: depending on the case, dark patterns fall under misleading or aggressive commercial practices.
Why it matters for sustainable IT
The topic is usually filed under ethics alone; it also bears directly on sobriety.
- Resource consumption. Artificially lengthened journeys, repeated prompts and engagement features multiply requests, heavy screens and time spent — hence the energy drawn by devices, networks and servers.
- Imposed usage time. Engagement mechanics extend usage without creating user value: the opposite of the useful / usable / used principle that underpins ecodesign of digital services.
- Convergence with accessibility. A deceptive interface is first of all an unreadable one. People with disabilities, people uneasy with digital tools or with the language bear the consequences first, which ties the subject to digital accessibility and inclusion.
- Reference frameworks. ISO/IEC TS 20125 and its French predecessor AFNOR SPEC 2201 explicitly call for discarding attention-capture features and design biases that mislead users or artificially extend their usage time. The GR491 covers the subject in its strategy and UX/UI families.
Designing without dark patterns
- Symmetry of choice. Accepting and refusing should take the same number of actions and receive the same visual treatment.
- Protective defaults. The pre-set option is the one that protects the user: no pre-ticked boxes, no data sharing by default.
- Desirable friction. Not all friction is bad: deliberately slowing an irreversible or costly action protects the user. Manipulative friction obstructs what the user wants; desirable friction helps them decide.
- Plain language. No double negatives, no loaded wording, full price shown as early as possible.
- Entry and exit symmetry. Cancelling, deleting an account or exporting data must be as accessible as signing up.
- Ethical review and user testing. Have journeys reviewed by people outside the project, with one question: would an informed user have made the same choice?
See also
- Ecodesign · Digital sobriety · Digital accessibility · GR491 · RGESN · ISO/IEC TS 20125 · Sustainable IT · Algorithmic bias
