Every choice architecture decision — every default, every friction asymmetry, every salience manipulation, every social norm communication — is an exercise of power over the decision-making of the person whose choice is being structured. The question of whether that power is being exercised in service of the decision-maker’s interests or against them is not merely an ethical question. It is the question that determines whether a specific behavioural design practice is legitimate, manipulative, or increasingly, legal or illegal.

The Thaler-Sunstein framework: two criteria for legitimate architecture

Thaler and Sunstein’s (2008) Nudge established “libertarian paternalism” as the ethical framework for choice architecture: libertarian because it preserves full freedom to choose any available option, and paternalistic because it designs the choice environment to nudge people toward options that are in their interest. Two criteria define the legitimate side of the line.

The first is option preservation. A legitimate nudge does not restrict any available option. The automatic enrolment pension default makes saving the path of least resistance, but the opt-out remains available and accessible. The fruit-at-eye-level supermarket design makes healthy choices more immediately visible, but no food has been removed. The nudge changes the relative ease of options without changing their availability. The dark pattern that buries the cancellation option five menus deep has not removed it, but it has made it practically inaccessible — which is functionally equivalent to restriction.

The second is transparency. A legitimate nudge is one the person could be told about without its effectiveness being materially undermined. The automatic enrolment pension design can be explained to participants — “we’ve set the default to saving 3% of your salary; you can change this at any time” — without destroying the default’s effectiveness, because the effectiveness rests on genuine preference alignment rather than on concealment. The dark pattern that exploits cognitive limitations against the user’s interest typically cannot survive the transparency test: explaining “we’ve designed the cancellation process to be deliberately confusing to reduce cancellation rates” would produce the user response that the design was trying to prevent.

The MINDSPACE welfare criterion: whose interests is the design serving?

The UK Behavioural Insights Team’s MINDSPACE framework (Dolan, Hallsworth, Halpern, King and Vlaev, 2010) added the welfare criterion that the Thaler-Sunstein framework implies but does not make explicit: a nudge is ethical if it serves the long-term wellbeing of the person being nudged, is transparent about its mechanism, and is compatible with the person’s own stated values and preferences.

This criterion is the most practically useful single question for the commercial practitioner evaluating a specific design choice: would the customer endorse this design if they understood it fully? The auto-renewing subscription is legitimate if the customer who is asked “would you like this to renew automatically unless you cancel?” would say yes — because the design is serving their revealed preference for continued subscription. The auto-renewing subscription is manipulative if the answer to that question would be no for a substantial proportion of the customer base — because the design is exploiting inertia to retain customers who would leave if the retention mechanism were not in place.

The objective design feature — opt-out renewal — is identical in both cases. The ethical status is entirely determined by whose interests the design is serving. This is why regulatory approaches to dark patterns focus not on specific design features but on whether those features are being deployed against user interests: the same feature can be legitimate or manipulative depending on the relationship between the design and the user’s genuine preferences.

The dark pattern taxonomy: what manipulation looks like in practice

Brignull’s (2010) foundational dark patterns classification identified the specific commercial practices that consistently fail the Thaler-Sunstein and MINDSPACE tests. Each exploits a specific cognitive mechanism against the user’s interests rather than in alignment with them.

The roach motel — easy to get in, hard to get out — deploys asymmetric friction to make subscription trivially easy and cancellation operationally onerous. The mechanism exploited is status quo inertia and effort-cost evaluation: the same cognitive processes that make opt-out defaults effective for beneficial purposes are used to retain customers who would leave if the effort cost were equivalent. The FTC’s 2023 complaint against Amazon’s “Iliad Flow” — a cancellation process requiring five pages and fifteen clicks — is the most legally significant commercial documentation of the roach motel pattern and its regulatory consequences.

Hidden costs deploy the commitment and sunk cost mechanisms: fees are added at the final checkout stage after psychological investment in the purchase has formed. The user who has spent twenty minutes selecting a product, configuring options, and proceeding to checkout has accumulated the sunk cost of that time investment and the psychological commitment of the near-completed purchase. Adding a significant booking fee or processing charge at this stage exploits the sunk cost’s asymmetric influence on the willingness to proceed.

Confirmshaming uses self-image threat to coerce compliance. Opt-out language like “No thanks, I don’t want to save money” exploits the self-esteem protection mechanism — the person who declines the email list is simultaneously required to affirm a self-description that conflicts with their self-concept. The friction is not practical but psychological: accepting the offer is easier for the self-concept than declining it through the shame-coded language.

Disguised advertising uses the availability and social proof mechanisms against user interests: search results designed to look like organic content, sponsored posts designed to look like peer recommendations. The exploitation is of the cognitive heuristic that makes familiar, socially validated content more credible — deployed to produce trust in content that was paid for rather than earned.

The regulatory direction of travel

The ICO’s cookie consent guidance, the FTC’s Amazon complaint, and the EU’s Digital Services Act all reflect a regulatory convergence around the welfare criterion: choice architecture that exploits cognitive biases against user interests is increasingly illegal, not merely unethical. The regulatory direction of travel matters commercially because it changes the risk calculation for dark pattern adoption. Practices that were previously a commercial optimisation decision — adding friction to cancellation to reduce churn — are becoming legal liability decisions.

The commercial case for ethical choice architecture is therefore not only about brand reputation or customer trust, though both are genuine. It is increasingly about legal compliance in the regulatory environment that behavioural economics’ own success has produced: by demonstrating that cognitive biases are powerful and predictable, the research has simultaneously provided commercial practitioners with design tools and regulators with the conceptual framework to identify when those tools are being deployed against the people they affect.

The practical test: four questions for evaluating a specific design

The research and regulatory frameworks together support four practical questions for evaluating whether a specific choice architecture intervention is on the legitimate or manipulative side of the line.

First: does the design preserve genuine access to all available options, or does it make one option practically inaccessible through asymmetric friction or obscured navigation? Legitimacy requires genuine option preservation, not merely theoretical availability.

Second: could the design be explained to the user without undermining its effectiveness or provoking their objection? If transparency would destroy the design’s effectiveness, the design is depending on concealment — which is the transparency test’s failure mode.

Third: would the customer endorse the design if they understood it fully? If a substantial proportion of the users affected by the design would object to it once aware, the welfare criterion is not met.

Fourth: is the design exploiting a cognitive limitation to produce an outcome the user would not choose under ideal decision-making conditions — or is it aligning the choice architecture with the user’s genuine preferences and making the preferred choice easier to act on? The mechanism is identical in both cases; the ethical status depends entirely on the relationship between the design’s effect and the user’s genuine interest.

The Patagonia anti-nudge as the commercial case for ethical architecture

Patagonia’s “Don’t Buy This Jacket” campaign and Worn Wear programme deliberately add friction to its own purchasing process — communicating the environmental cost of purchase, encouraging repair over replacement, and directing customers to second-hand alternatives. The commercial logic is counterintuitive: Patagonia is designing against its own short-term revenue by making purchase more effortful and alternatives more visible. The strategic logic is that this choice architecture, designed explicitly in the customer’s long-term interest rather than the company’s short-term commercial interest, is the foundation of the brand loyalty and customer trust that is Patagonia’s primary long-term commercial asset. The alignment between customer interest and commercial design is the source of the brand’s durability, not a sacrifice of it.

Books worth reading on this

The Ethical Algorithm by Michael Kearns and Aaron Roth is the most rigorous available account of how computational and algorithmic design choices — including the choice architecture decisions that are increasingly made by machine learning systems rather than human designers — can systematically exploit the same cognitive biases that dark patterns exploit, at scale and often without the designer’s explicit intent. For the entrepreneur whose products involve algorithmic recommendation, personalisation, or optimisation systems, Kearns and Roth’s account of how to build fairness, privacy, and welfare constraints into algorithmic systems is the most directly applicable available complement to the choice architecture ethics framework.

If the dynamics described here are significantly affecting your wellbeing, speaking with a psychologist is the right next step. UK: Samaritans (116 123, free, 24/7). Mind (0300 123 3393). BACP: bacp.co.uk/search/Therapists. Crisis Text Line — text HOME to 741741 (US, UK, Canada, Ireland). International: internationaltherapistdirectory.com.

This article is for educational and informational purposes only. Sources: Thaler, R.H. & Sunstein, C.R. (2008), Nudge: Improving Decisions About Health, Wealth, and Happiness, Yale University Press. Dolan, P., Hallsworth, M., Halpern, D., King, D. & Vlaev, I. (2010), MINDSPACE: Influencing Behaviour through Public Policy, Institute for Government. Brignull, H. (2010), Dark Patterns: User Interfaces Designed to Trick People, darkpatterns.org. Luguri, J. & Strahilevitz, L.J. (2021), Shining a Light on Dark Patterns, Journal of Legal Analysis, 13(1), 43–109. Hansen, P.G. & Jespersen, A.M. (2013), Nudge and the Manipulation of Choice, European Journal of Risk Regulation, 4(1), 3–28. Sunstein, C.R. (2015), The Ethics of Nudging, Yale Journal on Regulation, 32(2), 413–450. Mertens, S., Herberz, M., Hahnel, U.J.J. & Brosch, T. (2022), The Effectiveness of Nudging, PNAS, 119(1), e2107346118. Kahneman, D. (2011), Thinking, Fast and Slow, Farrar, Straus and Giroux. Kearns, M. & Roth, A. (2019), The Ethical Algorithm, Oxford University Press. O’Neil, C. (2016), Weapons of Math Destruction, Crown.