AI Glossary / Governance, ethics and safety / Beginner
Dark pattern
Definition
A dark pattern is a design practice that misleads, pressures or obstructs people in ways that undermine informed choice. It can appear in wording, layout, defaults or interaction flows, including the interfaces and conversations surrounding an AI service.
What it means
A user interface can influence decisions without making a false statement in isolation. Important conditions may be hidden, one option may be much easier to select than another, or an apparently helpful conversation may repeatedly divert a person from their expressed goal.
Not every persuasive design or inconvenient interface is automatically a dark pattern. The concern is how the overall interaction affects a person's ability to understand and act on their choice. Evidence about the design and its effects matters more than casually assigning motives to an individual designer.
AI can participate in these interactions through personalized messages or conversational flows. It is not required: ordinary websites have long used such patterns. Legal treatment depends on the practice and jurisdiction; the concept alone does not determine a legal conclusion.
How it works
A design shapes what users notice, how they interpret it and which actions are practical. A prominent button may imply that optional sharing is necessary, while the alternative is hidden behind confusing navigation. A cancellation flow may introduce repeated diversions after the user has clearly decided to leave.
In an AI interface, a conversational assistant might present one commercial choice as neutral advice without making the relevant relationship clear. Personalization can also change which prompts different users encounter, making a single screenshot an incomplete view of the experience.
Review the entire journey, including entry, explanation, confirmation, reversal and exit. Test whether people understand the consequences and can carry out their stated intentions. Conversion rates alone cannot distinguish informed preference from confusion or obstruction.
Three practical examples
- An AI service makes an optional training-data choice appear necessary to access an unrelated basic function.
- A subscription assistant repeatedly responds to cancellation requests with promotional questions while hiding the direct cancellation action.
- A recommendation interface presents paid placement as if it were an independent assessment of the best option.
Think of it this way
Think of a building whose exit signs point toward a sales desk while the real exit is concealed. People may reach the desk, but their arrival is poor evidence that they wanted a sales conversation.
The analogy highlights interference with an expressed intention. Digital interactions require contextual assessment; a longer route can sometimes have a legitimate purpose, such as confirming an irreversible action, if it is clear and proportionate.
Worked example
A click rate does not establish informed choice
In an invented usability test, 100 participants are asked whether to share their conversations for an optional purpose. Seventy click the most prominent continue button. A follow-up comprehension check finds that only 30 of those 70 understood that the button also enabled sharing.
The product team cannot describe all 70 clicks as evidence of an informed preference. Forty participants who clicked did not understand the relevant consequence in this test.
The team separates the choices, explains the optional use clearly and tests again. The example does not establish legal validity from a comprehension percentage; it shows why a behavioral metric needs interpretation. A design can increase recorded acceptance while reducing the quality of the decision behind it.
Illustrative case study
Fictional company Birch Notes offers an AI meeting assistant. Its setup screen labels the button enabling optional reuse of transcripts as improve my meetings, while the alternative is difficult to find.
A usability review shows that participants confuse this optional reuse with the core transcription function. The company rewrites the explanation, presents the choices clearly and ensures the basic service behaves as described when reuse is declined. It also checks that later settings reflect the original choice.
The review focuses on the actual understanding and control offered to users. A change in button color alone would not address the ambiguous purpose or the possibility that the backend ignored the selected preference.
Common misconceptions and mistakes
- A high acceptance rate proves users want the option. Confusion and friction can influence the rate.
- A disclosure somewhere on the site solves every issue. Timing, clarity and the complete interaction matter.
- Every reminder is manipulative. A clear, relevant reminder differs from pressure that obstructs a decision.
- Only visual design can create dark patterns. Conversational wording and repeated diversions can also undermine choice.
Business use
Businesses should review AI onboarding, permissions, purchases and cancellation paths for user understanding and practical control. This is especially important when automated experimentation optimizes only conversion or engagement.
Include measures of comprehension, unintended choices and successful completion of users' stated goals. Review personalized variants and the behavior after a choice, not only the initial interface. Clear design supports more credible evidence of what customers actually wanted and reduces the risk of building business decisions on misleading interaction data.
People also ask
Does a dark pattern require generative AI?
No. AI may deliver or personalize the interaction, but the underlying issue is design that undermines choice.
Is a default always a dark pattern?
No. Its context, explanation, consequences and ease of changing it matter.
Can a chatbot create one?
Yes. A conversation can conceal options, misrepresent consequences or obstruct an expressed request.
Check your understanding
A cancellation chatbot records long conversations as customer engagement, even when users repeatedly ask to cancel. What is wrong with that interpretation?
Answer: The duration may reflect obstruction rather than interest. Evaluate whether users can complete their stated goal, not just how long they remain in the interaction.
Related terms
- Consent
- Transparency
- Informed choice
- user interface
- persuasive design
- human agency
Learn next
Read consent and transparency. Walk through one optional-data or cancellation journey and record what a person must understand at each step. Test whether they can make and carry out either choice without misleading pressure.
More in Governance, ethics and safety
- Criterion validity - Criterion validity concerns how well a measurement relates to an appropriate external reference or outcome.
- Crowdsource - To crowdsource is to invite contributions from a distributed group of people to complete a task, supply information or solve a problem.
- Cybersecurity - Cybersecurity is the management of risks to digital systems, services and information from unauthorized access, disruption, misuse or other harmful activity.
- Data control - Data control is the practical ability to set and enforce conditions on how data is collected, accessed, changed, used, shared and retained.
- Data dredging - Data dredging is searching through many analyses or patterns and presenting selected favorable findings without adequately accounting for the search.
- Data fusion - Data fusion combines information from multiple sources to form a joint estimate, representation, or assessment.
Sources
U.S. Federal Trade Commission, Bringing Dark Patterns to Light: https://www.ftc.gov/reports/bringing-dark-patterns-light. Official report overview: https://www.ftc.gov/news-events/news/press-releases/2022/09/ftc-report-shows-rise-sophisticated-dark-patterns-designed-trick-trap-consumers. This entry explains design concerns using original scenarios; it does not give a jurisdiction-independent legal test.
Previous: Dark data Next: Data access