Research explained · Peer-reviewed study
Why automated dark-pattern detection still needs a journey review
The 2019 “Dark Patterns at Scale” study analysed roughly 53,000 product pages across about 11,000 shopping websites and identified 1,818 pattern instances spanning 15 types. It proved that visible, recurring interface signals could be collected systematically and reviewed at market scale. For Australian implementation, the equally important lesson is what a page crawl misses: authenticated states, price progression, subscription lifecycle, support channels and the evidence needed for a statutory assessment.
- Original work
- Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites
- Authors
- Arunesh Mathur, Gunes Acar, Michael J. Friedman, Elena Lucherini, Jonathan Mayer, Marshini Chetty and Arvind Narayanan
- Published
- 2019
- Venue
- Proceedings of the ACM on Human-Computer Interaction, CSCW
- Method
- A semi-automated crawl and expert review of shopping-site product pages, followed by taxonomy and deceptive-practice analysis.
- Sample or scope
- Approximately 53,000 product pages on about 11,000 shopping websites.
Read the evidence carefully
From research question to useful conclusion
- 1
Question
The 2019 “Dark Patterns at Scale” study analysed roughly 53,000 product pages across about 11,000 shopping websites and identified 1,818 pattern instances spanning 15 types.
- 2
Method
A semi-automated crawl and expert review of shopping-site product pages, followed by taxonomy and deceptive-practice analysis.
- 3
Finding
The research combined automated discovery with expert review to identify 1,818 instances across 15 pattern types and seven categories.
- 4
Boundary
The crawl focused on shopping product pages in 2019 and did not reconstruct every account, mobile, support, recurrence or cancellation state relevant to Australian implementation.
Evidence at a glance
From market crawl to reviewed candidates
The published counts describe different parts of the study rather than a single conversion funnel or legal rate.
The breakthrough was a repeatable collection method
The paper changed the field because it moved dark-pattern research beyond memorable examples. The team crawled thousands of shopping sites, identified candidate signals and reviewed recurring mechanisms. That made prevalence and distribution questions researchable at a new scale.
The workflow is still a useful model: broad collection first, structured classification second, expert interpretation last. Problems begin when those stages are collapsed into one automated verdict.
What is easy to collect
A product page exposes text, prices, timers, stock claims and activity messages. Those features can be captured repeatedly and compared. A system can flag changed wording or a timer whose advertised expiry does not match later observations.
Australian pricing review can benefit from that coverage. Yet section 48A work is not limited to one page. Teams need to follow the base price, extras, payment method, transaction charge, final total and confirmation. The pricing journey guide makes those states explicit.
What a page crawl leaves behind
Subscription scope can depend on the contract structure. Cancellation can require login, account state and later confirmation. A support obstruction may move from web to phone. Detriment may appear only after renewal or an unsuccessful exit.
These are journey questions. Evidence needs the sequence, timestamp, configuration, responsive state and result, not merely the page on which a candidate was first noticed.
The right role for automation
Use automation to widen observation and make repeated checks possible. Give each candidate a reason code and preserve the underlying state. Then ask a person to confirm the mechanism and its boundary. Finally, allow the responsible legal team to determine which provision and facts matter.
The practice examples illustrate that boundary by pairing a problematic scenario with a neutral design. They are learning aids, not machine-labelled business accusations.
Why this matters before commencement
Teams have time to build reliable evidence operations before 1 July 2027. A well-designed system will be more valuable than a late rush for a compliance score. It can show what changed, which journey was tested, which candidate appeared and how a human resolved it. That record supports review without claiming that detection alone proves the law.
Source check on 14 September 2026
The researchers’ revision log records that they narrowed their trick-question classification in July 2019. An opt-out checkbox was not enough on its own: misleading wording mattered. Australian review should preserve that context and the full choice presented. The historical crawl does not measure Australian services in September 2026.
What to retain
Three findings worth carrying into review
Coverage can scale
The research combined automated discovery with expert review to identify 1,818 instances across 15 pattern types and seven categories.
Visible signals are uneven
Scarcity, social-proof and textual claims lend themselves to collection more readily than a context-heavy subscription or cancellation experience.
Patterns were productised
The study identified third-party providers offering pattern-enabling services, showing that some mechanisms could propagate through reusable commercial tooling.
What this evidence cannot establish
- The crawl focused on shopping product pages in 2019 and did not reconstruct every account, mobile, support, recurrence or cancellation state relevant to Australian implementation.
- A research candidate is not an Australian legal finding and does not establish the complete section 28B test, transaction-charge treatment or subscription scope.
Questions for an Australian journey review
- Which visible claims and defaults can be checked automatically across releases, devices and locations with a low false-positive burden?
- Which legal or harm questions require an authenticated journey, configuration record, backend state or human interview beyond the page?
- Does the review workflow preserve candidate, evidence, human classification and legal assessment as separate stages?
Evidence base
Sources
- Dark Patterns at Scale: Findings from a Crawl of 11K Shopping WebsitesMathur et al.; arXiv · Secondary · checked 2026-09-14 · arXiv:1907.07032