Research explained · Peer-reviewed study
Why the quietest manipulation may deserve the closest review
Two large experiments by Luguri and Strahilevitz found that selected dark patterns could substantially increase sign-up for a dubious service. Mild patterns more than doubled sign-up and aggressive patterns produced an almost fourfold rate relative to the neutral condition, but aggressive designs also generated stronger backlash. The study suggests that quiet defaults, omissions and friction can deserve as much scrutiny as loud pressure. Its US experimental results do not determine section 28B or any Australian contravention.
- Original work
- Shining a Light on Dark Patterns
- Authors
- Jamie Luguri and Lior Jacob Strahilevitz
- Published
- 2021-03-23
- Venue
- Journal of Legal Analysis, volume 13, issue 1
- Method
- Two large online experiments comparing neutral, mild and aggressive dark-pattern conditions and then isolating selected techniques.
- Sample or scope
- Census-weighted samples of adult participants in the United States.
Read the evidence carefully
From research question to useful conclusion
- 1
Question
Two large experiments by Luguri and Strahilevitz found that selected dark patterns could substantially increase sign-up for a dubious service.
- 2
Method
Two large online experiments comparing neutral, mild and aggressive dark-pattern conditions and then isolating selected techniques.
- 3
Finding
The first experiment reported more than twice the neutral sign-up rate under mild patterns and almost four times the rate under aggressive patterns.
- 4
Boundary
The research uses selected hypothetical services and US samples; Australian products, audiences, prices and decision settings may produce different responses.
Evidence at a glance
Acceptance rates in the first experiment
Published Table 2 percentages among participants completing the experiment, comparing the same acceptance outcome across conditions.
Loud design creates its own warning
A page that traps the user or mocks refusal is easy to recognise as hostile. A preselected option, muted renewal term or slightly longer decline route can be much less dramatic. The experiments suggest that this difference matters: mild patterns changed behaviour while provoking less of the backlash seen in aggressive conditions.
For Australian teams, that is a reason to examine the ordinary-looking choices that pass through a design review without debate.
The experiment gives effect evidence, not a statutory formula
Participants encountered controlled interfaces for hypothetical services. The researchers could change the choice architecture while holding other features stable, then compare sign-up and participant response.
Section 28B asks a different set of questions. For conduct after commencement, teams must establish the consumer connection, analyse manipulation or unreasonable distortion and assess financial or other detriment. A US experiment does not answer those elements. It can strengthen the decision to investigate likely effect and avoid assuming that only outrageous interfaces matter.
Where quiet asymmetry hides
Subscription journeys are fertile ground. Acceptance can be one prominent button, while refusal is a text link. A renewal term can be present but visually remote from the decision. Cancellation can be available online but require several more steps than entry.
The false-hierarchy practice and hard-cancel practice show how those differences can be made visible as HTML rather than described only in prose.
A useful neutral-comparator test
Build a genuinely neutral version before debating intent. Give each material option comparable language and prominence, place the consequence beside the action and remove avoidable friction from the alternative. Then compare comprehension, selection, regret and downstream support contacts.
If the result changes sharply, the team has evidence that presentation mattered. It still needs to interpret why, for whom and under which provision.
What to record
Preserve more than the final click. Record the full journey, the variant or experiment assignment, the terms and prices displayed, the responsive state and the later account outcome. Otherwise the team may know that conversion moved without being able to reconstruct the decision environment that produced it.
Source check on 14 September 2026
The publisher’s Table 2 supports the chart’s exact acceptance rates among experiment completers. If dropouts count as rejection, the mild rate is 25.4% and the aggressive rate 37.2%, against 11.3% for the control. These remain US experimental results, so neither denominator supplies an Australian legal threshold.
What to retain
Three findings worth carrying into review
Choice architecture altered sign-up
The first experiment reported more than twice the neutral sign-up rate under mild patterns and almost four times the rate under aggressive patterns.
Subtlety reduced the warning signal
Aggressive design provoked a stronger backlash, while mild conditions could influence behaviour without producing the same resistance.
Not every tactic worked equally
The second experiment found meaningful differences among techniques, cautioning against treating a familiar label as proof of effect.
What this evidence cannot establish
- The research uses selected hypothetical services and US samples; Australian products, audiences, prices and decision settings may produce different responses.
- Measured influence is relevant evidence but does not by itself prove manipulation or unreasonable distortion, consumer connection or likely detriment under section 28B.
Questions for an Australian journey review
- Would the consumer’s decision change if material choices had comparable prominence, timing and effort?
- Are review teams looking beyond aggressive pressure to defaults, small-print consequences and one-extra-step asymmetries?
- What Australian user research and product evidence can test likely effect for the actual audience and journey?
Evidence base
Sources
- Shining a Light on Dark PatternsLuguri and Strahilevitz; Journal of Legal Analysis · Secondary · checked 2026-09-14 · DOI 10.1093/jla/laaa006