KFTC Announces Results of a Randomized Controlled Trial (RCT) Involving 3,072 Consumers
June 28, 2026
Economic Analysis Division
The Korea Fair Trade Commission has published a report titled "Consumer Behavior Experimental Study on Platform Algorithm-Based Self-Preferencing," which experimentally investigates the effects of digital platforms' algorithm-based self-preferencing practices on consumer choices.
This study was conducted to experimentally verify how consumer choices shift when digital platforms preferentially expose their own products through search, recommendation, and ranking algorithms. The KFTC built a virtual online shopping mall that accurately replicated the interface of an actual e-commerce platform and conducted a Randomized Controlled Trial (RCT) on 3,072 consumers.
Note: An RCT is a research methodology that randomly assigns participants to a treatment group and a control group to identify the specific causal effect of a treatment by comparing the outcomes between the two groups.
In particular, the study analyzed not only the causal effect of self-preferencing on the distortion of consumer choices but also whether information-providing corrective measures (such as labeling and ranking criteria disclosures) effectively mitigate this distortion. The experimental design and empirical analysis were jointly conducted by internal KFTC researchers and an external expert in experimental and behavioral economics (Professor Shin Eun-cheol, KAIST College of Business).
Summary of Key Findings
The results revealed that the vast majority of consumers complete their product search and purchase within the top-ranked items of the platform's default sorting order. This demonstrates that platforms can drastically increase the purchase rate of their own products through artificial algorithmic manipulation alone, regardless of intrinsic competitive factors like price and quality.
Furthermore, the study confirmed the structural limitations of information-based remedies. Attaching a "Label" indicating a platform's own product paradoxically increased the purchase rate of that product. Additionally, a "Disclosure" detailing the sorting criteria failed to attract consumers' attention, proving ineffective at systematically correcting choice distortion.
Experimental Design
To test the effects of algorithm-based self-preferencing as realistically as possible, the KFTC created a virtual online shopping mall. Participants performed two separate shopping tasks.
Round 1: All groups shopped in the same environment (no self-preferencing manipulation).
Round 2: Participants were randomly assigned to one of four groups based on the presence of labels and ranking disclosures (self-preferencing manipulation applied).
Three product categories with different purchasing characteristics were used: Bluetooth speakers, Vitamin C, and toilet paper. Each participant was randomly assigned two of these three categories. This randomized assignment of product categories and treatment groups was designed to causally identify the impact of the experimental treatment on consumer behavior, isolating it from individual preferences or shopping habits.
The study heavily focused on verifying whether consumers perceive a platform's search ranking as an objective signal of quality and relevance, and whether this perception leads to actual choice distortion. In Round 2, the platform artificially placed a "cloned" product—identical but 10% more expensive—at the top of the search results. The KFTC then recorded the entire consumer journey (clicks, scrolling, page movements, sorting changes, filter usage) via behavioral logs.
Experimental Results
1. Strong Ranking Dependency
Consumers rely heavily on the rankings suggested by algorithms.
51.7% of all purchases were concentrated on the top 5 products.
94.6% of consumers completed their purchases within the first page of search results.
Only 25.2% of consumers changed the default sorting order.
83.8% of consumers did not use the filter function (a tool to search for products meeting specific criteria like features or price ranges) at all.
This indicates a clear tendency for consumers to accept the default sorting order and rankings provided by the platform without active adjustment.
2. Choice Distortion Due to Algorithm Manipulation
When the platform placed a 10% more expensive self-preferenced product (a clone that ranked in the middle/lower tiers in Round 1) at the top of the search results, the purchase rate for that product surged by 34 percentage points (from 1% before self-preferencing to 35% after).
Conversely, competing products that were originally top-ranked were pushed down the list, and their purchase rate plummeted by 32 percentage points (from 52% to 20%).
This shows that self-preferencing not only increases the sales of specific products but also significantly diminishes the selection opportunities for competing products. Consumers often misinterpret platform rankings as objective quality signals reflecting product superiority or suitability. Consequently, simple rank manipulation can severely distort final purchase choices in favor of the platform's intentions.
3. Limitations of Information-Providing Corrective Measures
The study analyzed the effectiveness of information-providing measures (labels and disclosures) designed to mitigate choice distortion and found their impact to be highly limited.
Labels (e.g., "SCpay" tags): Adding a label to identify a self-preferenced product actually decreased active consumer search behavior and further increased the purchase rate of the self-preferenced product by an additional 4.5 percentage points.
Disclosures: A banner explaining that business interests may be reflected in the sorting criteria was noticed by only 10.7% of consumers, meaning it failed to reach the vast majority. However, among the small subset of consumers who actually read the disclosure, the purchase rate of the self-preferenced product dropped by approximately 18.4 percentage points.
Interestingly, consumers who purchased the more expensive self-preferenced product did not perceive it as a loss of welfare. On the contrary, their purchase satisfaction and trust in the ranking system actually increased. This implies that choice distortion driven by algorithms possesses structural characteristics that make it exceptionally difficult for consumers to recognize on their own.
Expected Effects and Future Plans
This report is highly significant as the KFTC's first experimental study identifying the causal effects of digital platforms' algorithm-based self-preferencing on consumer choice.
Given the secrecy and opacity of algorithms in platform markets, it is often difficult to prove a causal relationship between a platform's conduct and market outcomes. Methodologies like Randomized Controlled Trials (RCTs) are expected to become highly valuable analytical tools, supplementing future competition policy research and law enforcement.
Moving forward, the KFTC plans to actively utilize diverse economic analysis methodologies—including experimental research, econometric analysis, and behavioral economics approaches—to proactively address emerging competition issues in digital markets.