Competition Law and Recommendation Engines and Competition Concerns

1. Introduction

Recommendation engines are algorithmic systems that determine or influence what users see, purchase, listen to, watch, read, or otherwise engage with. They are used by e-commerce marketplaces, search engines, social-media platforms, streaming services, app stores, travel platforms, food-delivery platforms and digital advertising intermediaries.

A recommendation engine can be pro-competitive because it reduces search costs, improves matching between consumers and products, enables personalization and helps new products reach consumers. However, where the operator of the recommendation system also competes with the businesses whose products are being ranked, the system may create significant competition-law concerns.

The central competition-law question is therefore not whether an algorithm recommends products, but whether the design, operation or manipulation of the recommendation system is capable of excluding competitors, exploiting users, facilitating coordination, or entrenching market power.

2. Meaning of a Recommendation Engine

A recommendation engine is an automated system that ranks, recommends, filters or personalizes products, services or content.

Typical inputs include:

• previous purchases;
• browsing history;
• searches;
• clicks;
• ratings and reviews;
• location;
• price;
• availability;
• conversion rates;
• seller performance;
• advertising payments;
• engagement levels;
• customer characteristics; and
• platform-specific commercial objectives.

The output may determine:

• "recommended for you";
• "best match";
• "featured";
• "popular";
• "top seller";
• "sponsored";
• "similar products";
• "customers also bought";
• "people you may know";
• "next video";
• "recommended app"; or
• the order in which search results appear.

Consequently, ranking and recommendation can become an economically significant gateway to consumers.

3. Relevant Competition-Law Framework

Recommendation engines can potentially engage several branches of competition law.

A. Abuse of dominance

A dominant platform may potentially abuse its position where its recommendation system systematically disadvantages competing businesses.

Relevant theories include:

• self-preferencing;
• discriminatory ranking;
• exclusionary ranking;
• demotion of rivals;
• preferential treatment of affiliated products;
• manipulation of search results;
• tying;
• refusal or degradation of access;
• leveraging dominance from one market into another.

B. Restrictive agreements

Recommendation systems can also facilitate:

• resale-price maintenance;
• parity clauses;
• territorial restrictions;
• exclusionary agreements;
• discriminatory platform contracts;
• coordinated restrictions between sellers.

C. Algorithmic collusion

Algorithms can potentially make coordination easier by:

• rapidly observing competitors' prices;
• automatically responding to price changes;
• implementing common pricing rules;
• reducing the need for direct human communication; and
• making deviations from coordinated behaviour easier to detect.

However, mere parallel algorithmic pricing does not automatically establish an unlawful cartel. Competition authorities generally need to establish the relevant legal elements of an agreement, concerted practice, or other prohibited conduct.

D. Merger control

Recommendation engines may become relevant in digital mergers where a transaction combines:

• consumer data;
• user attention;
• search/ranking infrastructure;
• marketplace access;
• advertising data; and
• complementary recommendation technologies.

A merger may therefore increase the ability of the merged entity to control the information through which consumers discover competing products.

4. Self-Preferencing Through Recommendation Engines

One of the most important concerns is self-preferencing.

Suppose a marketplace operates:

1. an online marketplace;
2. a recommendation engine; and
3. its own private-label products.

The platform could theoretically manipulate the recommendation algorithm so that its own products appear:

• higher in search results;
• more frequently in recommendations;
• as "best match";
• in default selections;
• in personalised suggestions; or
• in "featured" positions.

The competition issue becomes particularly significant where rivals cannot obtain equivalent visibility despite having comparable or better commercial characteristics.

The relevant assessment may consider:

• market power;
• objective ranking criteria;
• transparency;
• actual ranking effects;
• foreclosure of competitors;
• consumer harm;
• efficiency justifications; and
• whether less restrictive alternatives exist.

5. Discriminatory Ranking

A recommendation engine may discriminate between businesses using apparently neutral criteria.

For example:

Seller A and Seller B have similar prices, quality, delivery times and consumer ratings, but Seller A is repeatedly promoted because it is affiliated with the platform.

Such conduct can raise concerns where the discrimination:

• materially affects access to consumers;
• disadvantages competing suppliers;
• protects an affiliated business;
• increases barriers to entry; or
• exploits dependence on the platform.

The competition authority may therefore examine the actual operation of the algorithm rather than merely its stated terms and conditions.

6. Opaque Recommendation Algorithms

Opacity itself is not necessarily an antitrust violation.

Nevertheless, opacity can make it difficult for competitors and regulators to determine whether discriminatory conduct exists.

Competition concerns may arise where the platform:

• changes ranking criteria without notice;
• secretly penalizes competitors;
• gives preferential treatment to affiliates;
• uses advertising payments to influence allegedly organic rankings;
• combines commercial and non-commercial ranking criteria; or
• prevents independent auditing.

The legal challenge is to distinguish legitimate algorithmic optimization from exclusionary manipulation.

7. Data Advantage and Recommendation Engines

Recommendation systems become more powerful as they obtain additional data.

A large platform may possess:

• transaction data;
• consumer preference data;
• seller performance data;
• clickstream information;
• search data;
• product-level conversion data;
• price information; and
• behavioural information.

A dominant platform can potentially use this information to improve its own competing products while depriving rivals of comparable access to data.

This creates a possible data foreclosure theory.

Competition authorities may therefore ask:

Does control over commercially valuable data create or reinforce market power, and is the platform using that data to disadvantage competitors?

8. Recommendation Engines and Consumer Lock-In

Recommendation systems may increase switching costs.

For example, a consumer who has spent years generating:

• playlists;
• purchase histories;
• ratings;
• preferences;
• saved products;
• viewing histories; and
• personalised recommendations

may become less willing to move to another platform.

This can contribute to:

• network effects;
• ecosystem effects;
• consumer lock-in;
• reduced multi-homing; and
• increased barriers to entry.

Competition analysis should therefore consider not only price, but also data portability, interoperability, switching costs and access to users.

9. Recommendation Engines and Advertising

A platform may combine organic recommendations with sponsored placement.

Potential concerns arise where consumers cannot distinguish between:

• genuinely personalised recommendations;
• paid placement;
• platform-owned products; and
• algorithmically selected results.

This may be particularly important where advertising becomes effectively necessary for suppliers to obtain visibility.

The competition-law analysis can involve:

• self-preferencing;
• tying;
• discriminatory access;
• leveraging;
• exclusionary conduct; and
• exploitation of business-user dependence.

10. Recommendation Engines and Algorithmic Collusion

Recommendation technology can also affect the horizontal competition between suppliers.

Consider several competing sellers whose pricing algorithms continuously observe the market.

An algorithm could:

1. monitor competitors;
2. identify price changes;
3. react immediately;
4. maintain a predetermined pricing rule; and
5. punish deviations through automated responses.

This can potentially make coordinated outcomes more stable.

However, competition law distinguishes between:

Independent algorithmic conduct

and

algorithmic implementation of an unlawful agreement or concerted practice.

The existence of sophisticated software alone is insufficient to establish a cartel.

11. Case Laws and Major Decisions

Case 1: Google Shopping – European Commission

Google Search (Shopping), Commission Decision, 2017

This is one of the most important authorities for analysing algorithmic ranking and self-preferencing.

The European Commission found that Google had systematically given prominent placement to its comparison-shopping service while applying generic search algorithms to competing comparison-shopping services.

The case concerned the interaction between:

• search algorithms;
• ranking;
• visibility;
• traffic;
• dominance; and
• preferential treatment.

The General Court substantially upheld the Commission's decision in 2021, while the litigation subsequently continued through the EU judicial system.

Competition significance

The case demonstrates that an algorithmically determined ranking system can become relevant to abuse-of-dominance analysis where a dominant platform allegedly treats its own service more favourably than competing services.

Principle

Control over an important ranking gateway can give a dominant undertaking the ability to influence competitive visibility.

Case 2: Google Android – European Commission

Google Android, European Commission Decision, 2018

The Android case primarily concerned tying and exclusionary practices rather than recommendation engines themselves. However, it is highly relevant to recommendation ecosystems.

The Commission examined Google's position in:

• general search;
• mobile operating systems;
• app stores; and
• mobile ecosystems.

The case illustrates how control over an ecosystem can allow a dominant undertaking to influence the distribution and visibility of complementary services.

Competition significance

Recommendation engines should therefore be analysed together with:

• operating systems;
• default settings;
• app stores;
• search functions; and
• ecosystem access.

Principle

Algorithmic recommendations may have stronger exclusionary effects when combined with control over a broader digital ecosystem.

Case 3: Google Search – European Commission

The European Commission's investigations into Google's general-search practices provide important background for assessing search ranking and visibility.

Search engines determine which websites consumers encounter and in what order.

Where a dominant search provider allegedly uses its position to advantage affiliated services or disadvantage competing services, ranking becomes an important competitive parameter.

Competition significance

The case illustrates the economic importance of:

• ranking;
• traffic allocation;
• visibility;
• default positioning; and
• access to consumer demand.

Principle

Control over information-ranking infrastructure can confer substantial competitive significance even where the underlying service is provided without a monetary price.

Case 4: Amazon Marketplace – German Competition Authority

Bundeskartellamt proceedings concerning Amazon's terms and marketplace practices

The German competition authority investigated Amazon's conduct as the operator of a major marketplace and subsequently obtained significant changes to Amazon's marketplace terms.

The proceedings examined Amazon's relationship with third-party sellers and the company's role as both:

• marketplace operator; and
• retailer.

This dual role is particularly important for recommendation systems.

Competition significance

A platform that controls seller access and simultaneously competes with those sellers may face conflicts between:

• neutral ranking;
• platform-owned products;
• seller data;
• marketplace rules; and
• recommendation mechanisms.

Principle

A platform's dual role as intermediary and competitor can make discriminatory marketplace practices particularly significant under competition law.

Case 5: Amazon Marketplace – European Commission

European Commission Amazon Marketplace investigation

The Commission investigated Amazon's use of non-public seller data and its role as both marketplace operator and retailer.

The investigation focused on whether Amazon could use information generated by independent sellers to compete with those sellers.

Although the principal issue was data use rather than recommendation ranking itself, the decision is highly relevant because seller data can influence:

• product selection;
• ranking;
• recommendation;
• product development;
• inventory decisions; and
• competitive positioning.

Principle

Access to commercially sensitive marketplace data can affect the neutrality and competitive significance of platform recommendation systems.

Case 6: Naver – Korea Fair Trade Commission

Naver search-result manipulation case

The Korean competition authority examined Naver's manipulation of search results in connection with its own affiliated services.

The proceedings are particularly relevant to recommendation and ranking systems because Naver operated an important search and platform ecosystem while also having affiliated businesses.

The authority examined whether ranking arrangements favoured Naver's own services over competing services.

Competition significance

The case demonstrates that concerns about self-preferencing and manipulation of rankings are not confined to European competition law.

Principle

A dominant search or platform operator may face competition scrutiny when ranking mechanisms systematically favour affiliated services.

Case 7: Google Shopping – General Court of the European Union

Google and Alphabet v European Commission, Case T-612/17

The General Court's judgment is especially important because it considered whether Google's conduct constituted an abuse of dominance.

The Court upheld the essential finding that Google had treated its comparison-shopping service more favourably than competing comparison-shopping services.

Competition significance

The judgment reinforces the importance of examining:

• traffic;
• visibility;
• ranking;
• algorithmic placement;
• competitive foreclosure; and
• the economic characteristics of digital markets.

Principle

Preferential treatment within a dominant digital platform can have competition significance even when the mechanism producing the preference is algorithmic.

12. Additional Relevant Authorities

Other important authorities for developing the legal framework include:

1. Microsoft – Commission

Microsoft's tying and interoperability litigation illustrates how control over one digital layer can be leveraged into adjacent markets.

2. Intel – European Commission

Intel demonstrates the importance of analysing exclusionary strategies through their effects on competitors and customers rather than looking only at formal contractual language.

3. Qualcomm – European Commission

Qualcomm's rebate-related proceedings demonstrate the importance of analysing conduct by dominant firms through economic effects and foreclosure theories.

4. Apple App Store / Epic Games litigation

The disputes surrounding Apple's App Store illustrate competition issues arising where a platform controls:

• discovery;
• distribution;
• payment;
• ranking;
• access conditions; and
• competing applications.

These cases are particularly relevant when recommendation systems are integrated with app-store governance.

13. Algorithmic Self-Preferencing: Competition Test

A useful analytical framework is:

Step 1 – Define the market

Identify:

• product/service market;
• geographic market;
• platform side;
• consumer side; and
• business-user side.

Step 2 – Establish market power

Examine:

• market share;
• network effects;
• switching costs;
• data advantages;
• entry barriers;
• multi-homing;
• interoperability; and
• ecosystem effects.

Step 3 – Identify the recommendation mechanism

Determine:

• what the algorithm ranks;
• how ranking occurs;
• what data it uses;
• whether commercial payments influence ranking;
• whether affiliated products receive preferential treatment.

Step 4 – Compare treatment

Compare:

Platform's own products
versus
independent competitors.

Step 5 – Examine foreclosure

Determine whether rivals suffer:

• reduced visibility;
• reduced traffic;
• lower conversion;
• loss of customers;
• increased acquisition costs; or
• exclusion from the market.

Step 6 – Examine consumer effects

Consider:

• reduced choice;
• higher prices;
• lower quality;
• reduced innovation;
• reduced privacy;
• reduced variety; or
• reduced service quality.

Step 7 – Consider objective justification

The platform may argue that its ranking reflects:

• relevance;
• quality;
• delivery performance;
• consumer preferences;
• security;
• fraud prevention;
• innovation; or
• technical efficiency.

These justifications must be assessed against the actual design and effects of the recommendation system.

14. Recommendation Engines and Essential-Facility Concepts

In some circumstances, a recommendation or discovery interface may become an indispensable route to customers.

This does not automatically make the recommendation system an essential facility.

Traditional essential-facility principles generally require careful examination of factors such as:

• indispensability;
• lack of realistic alternatives;
• feasibility of access;
• elimination or substantial restriction of competition; and
• objective justification.

Digital markets complicate the analysis because consumers can sometimes use several platforms simultaneously.

Therefore, multi-homing and alternative discovery channels become particularly important.

15. Recommendation Engines and New Entrants

A dominant recommendation platform can affect market entry in two opposite ways.

Positive effect

Recommendations can help new businesses obtain visibility without establishing their own distribution networks.

Negative effect

If the platform changes its ranking system to favour established or affiliated products, new entrants may struggle to reach customers.

Thus, recommendation engines can either:

reduce entry barriers

or

increase entry barriers.

The competition assessment depends on actual market conditions.

16. Recommendation Engines in E-Commerce

Common competition concerns include:

• preferential ranking of private-label goods;
• manipulation of "best match";
• exclusion of competing sellers;
• discriminatory product recommendations;
• use of seller data;
• paid ranking;
• loyalty-based ranking;
• bundling;
• platform commission discrimination;
• suppression of rival products; and
• manipulation of consumer reviews.

The distinction between organic ranking and sponsored placement can be especially important.

17. Recommendation Engines in Digital Advertising

Ad-tech recommendation systems may create concerns through:

• preferential access to advertising inventory;
• discriminatory auction rules;
• self-preferencing;
• use of competitor data;
• tying advertising services together;
• exclusion of rival ad exchanges; and
• algorithmic optimisation that disadvantages competing intermediaries.

Where the platform operates both the advertising marketplace and competing advertising services, conflicts of interest can become particularly significant.

18. Recommendation Engines in Streaming Services

Streaming platforms can use recommendation engines to determine:

• which films appear first;
• which songs are suggested;
• which creators receive exposure;
• autoplay sequences; and
• personalised homepages.

Competition issues can arise if the platform:

• systematically favours its own content;
• discriminates against independent producers;
• conditions visibility on exclusivity;
• uses data from independent producers to compete against them; or
• imposes discriminatory access conditions.

However, editorial or consumer-preference choices should not automatically be treated as anticompetitive conduct.

19. Recommendation Engines and Network Effects

Recommendation systems can strengthen network effects because:

1. more users generate more data;
2. more data improves recommendations;
3. better recommendations attract more users;
4. more users generate additional data.

This creates a data-feedback loop.

A large platform can therefore obtain a competitive advantage that is difficult for smaller rivals to replicate.

Competition analysis may consequently consider:

• data accumulation;
• data portability;
• interoperability;
• switching costs;
• access to users; and
• contestability.

20. Remedies

Where competition authorities establish an infringement, possible remedies may include:

Structural remedies

• divestiture;
• separation of business units;
• restrictions on acquisitions.

Behavioural remedies

• non-discrimination obligations;
• ranking transparency;
• algorithmic auditing;
• access obligations;
• data-use restrictions;
• prohibition of self-preferencing;
• interoperability;
• data portability.

Procedural safeguards

Platforms may be required to:

• document ranking criteria;
• maintain audit trails;
• preserve algorithmic records;
• explain material ranking changes;
• establish compliance procedures; and
• provide regulator access to relevant information.

21. Compliance Framework for Businesses Operating Recommendation Engines

A competition-compliance programme should include:

1. Algorithm inventory

Maintain a record of algorithms affecting:

• ranking;
• recommendations;
• search;
• advertising;
• pricing; and
• product visibility.

2. Conflict-of-interest assessment

Identify whether the platform also competes with businesses whose products it recommends.

3. Ranking governance

Document legitimate ranking criteria.

4. Audit mechanisms

Regularly test for:

• affiliate bias;
• discriminatory treatment;
• unexplained demotion;
• exclusionary effects.

5. Data governance

Separate competitively sensitive seller information from internal competitive decision-making where appropriate.

6. Human oversight

High-impact algorithmic changes should receive appropriate legal and competition review.

7. Record keeping

Maintain:

• algorithm versions;
• ranking changes;
• testing results;
• governance approvals;
• complaints; and
• explanations for material changes.

22. Key Competition-Law Questions

23. Distinction Between Legitimate and Potentially Problematic Recommendations

Generally legitimate objectives

• improving relevance;
• reducing search costs;
• preventing fraud;
• improving delivery reliability;
• matching consumer preferences;
• improving security;
• recommending products based on genuine user demand.

Potential competition concerns

• artificially promoting affiliated products;
• systematically demoting rivals;
• secretly changing ranking criteria;
• conditioning visibility on unrelated purchases;
• exploiting confidential competitor data;
• excluding rival platforms;
• using algorithms to implement an unlawful agreement;
• manipulating recommendations to prevent switching.

The economic effect and legal context are therefore more important than the mere existence of an algorithm.

24. Conclusion

Recommendation engines are becoming an important competitive infrastructure in digital markets. Their importance arises because consumers frequently encounter products and services through algorithmically selected rankings rather than through traditional market search.

The principal competition concerns are:

1. self-preferencing;
2. algorithmic discrimination;
3. foreclosure of rival businesses;
4. data exploitation;
5. algorithmic collusion;
6. consumer lock-in;
7. barriers to entry;
8. leveraging of ecosystem power;
9. manipulation of advertising and sponsored placement; and
10. lack of effective access or interoperability.

The Google Shopping litigation is particularly significant because it demonstrates how ranking and preferential treatment can become central to abuse-of-dominance analysis. The Amazon and Naver proceedings further illustrate the importance of examining the relationship between platform control, seller data, affiliated services and ranking mechanisms.

Ultimately, competi