Competition Law and Future Regulation of Recommendation Architectures

Introduction

Recommendation architectures are the technological and commercial systems through which digital platforms decide what products, services, content, sellers, advertisements, applications, or information a user is shown. They include recommendation algorithms, ranking systems, personalised feeds, search suggestions, “people also bought” systems, default recommendations, content-ranking engines, and AI-driven decision systems.

From a competition-law perspective, the important issue is not merely that a platform recommends something. The concern arises when the architecture through which recommendations are generated becomes a source of market power, exclusion, discrimination, self-preferencing, foreclosure, or manipulation of competitive visibility.

Traditional competition law generally examines price, output, market share and contractual restrictions. Future competition law increasingly has to examine algorithmic visibility: who gets recommended, who is suppressed, what data determines the recommendation, whether rivals can obtain comparable visibility, and whether the platform's own products receive structural advantages.

1. Meaning of Recommendation Architectures

A recommendation architecture may consist of several interconnected layers:

A. Data layer

The platform collects:

• search history;
• purchases;
• clicks;
• browsing behaviour;
• location;
• engagement data;
• ratings;
• transaction history;
• social connections; and
• inferred preferences.

B. Algorithmic layer

Machine-learning models process the data and predict:

• what a consumer is likely to buy;
• what content the consumer will watch;
• which seller should appear first;
• which application should be installed;
• which advertisement should be displayed.

C. Ranking layer

The system converts predictions into an ordered list.

For example:

Seller A → Seller B → Seller C

The order itself may determine a substantial proportion of transactions.

D. Interface layer

The platform determines how recommendations are presented:

• “Top Pick”;
• “Recommended for You”;
• “Best Match”;
• “Sponsored”;
• “Popular Near You”;
• default application;
• preferred seller; or
• automatically generated playlist.

E. Feedback layer

Consumer behaviour then becomes new training data.

Thus:

Recommendation → consumer response → new data → algorithmic learning → improved recommendation → stronger platform position

This can create a data-feedback loop.

2. Why Recommendation Architectures Matter Under Competition Law

Recommendation systems can influence competition even when the underlying products remain legally substitutable.

A platform may not expressly prohibit a rival from competing. Instead, it may reduce the rival's algorithmic visibility.

For example:

Platform's own product receives 35% of recommendation impressions while comparable independent products receive 5%.

The competitive effect may therefore occur through visibility rather than exclusion from access.

3. Major Competition Concerns

A. Self-Preferencing

A vertically integrated platform may recommend its own products or services more prominently than competing products.

Examples include:

• search engines promoting their own services;
• marketplaces promoting private-label products;
• app stores promoting their own applications;
• streaming platforms promoting their own content;
• travel platforms favouring affiliated hotels;
• financial platforms recommending affiliated products.

The legal question is whether preferential treatment constitutes an exclusionary abuse or another prohibited conduct.

4. Algorithmic Discrimination

Recommendation systems can discriminate between otherwise comparable businesses.

Possible variables include:

• commission rates;
• advertising expenditure;
• platform relationships;
• historical sales;
• strategic importance;
• membership status;
• data-sharing arrangements.

The difficulty is that discriminatory treatment may be invisible because the platform can describe the outcome as an algorithmic prediction rather than an intentional decision.

5. Ranking as a Competitive Bottleneck

In traditional markets, competitors compete for shelf space.

In digital markets, they increasingly compete for:

algorithmic attention.

The first few recommendations may receive a disproportionately large number of clicks.

Consequently, recommendation architecture can function as a digital bottleneck facility.

A seller may technically remain on the platform but effectively lose access to consumers if its ranking deteriorates.

6. Data Advantage and Recommendation Power

Large platforms frequently possess large quantities of behavioural data.

A simplified competitive cycle is:

More users → more behavioural data → better predictions → better recommendations → greater consumer engagement → more users

This creates a potentially self-reinforcing network effect.

The competition-law concern becomes stronger where competitors cannot realistically reproduce the same volume, variety or quality of data.

7. Tying Recommendation to Platform Services

A platform might condition favourable recommendation placement upon:

• purchasing advertising;
• using the platform's payment system;
• using its logistics service;
• accepting data-sharing conditions;
• subscribing to premium services.

This may transform recommendation power into a mechanism for leveraging dominance into adjacent markets.

8. Algorithmic Exclusion

An algorithm can potentially exclude a competitor without a traditional contractual restriction.

For example:

1. A platform changes its recommendation model.
2. Rival products fall substantially in ranking.
3. Platform-owned products rise.
4. Consumer traffic migrates toward the platform's products.
5. Rivals lose scale.
6. Their data quality declines.
7. Their future ranking becomes even weaker.

This can produce an algorithmic foreclosure cycle.

9. Important Case Laws

The following cases are particularly relevant to the development of competition law governing recommendation architectures. Some concern recommendation systems directly, while others establish principles that can be applied to algorithmic ranking and personalised recommendations.

Case 1: Google Search (Shopping) — European Commission / General Court

Google Search (Shopping), Case T-612/17

Facts

Google was found to have given prominent placement in its general search results to its own comparison-shopping service while competing comparison-shopping services were subject to less favourable positioning.

Legal significance

The case established an important principle concerning the competitive significance of visibility and ranking.

The General Court upheld the finding of abuse of dominance, although it modified aspects of the Commission's reasoning.

Relevance to recommendation architectures

This is one of the most important precedents for future recommendation regulation because:

• ranking can affect competition;
• prominence can determine consumer traffic;
• discrimination in algorithmic visibility can disadvantage rivals;
• technically available access does not necessarily mean effective competitive access.

Principle

Algorithmic prominence can itself have competitive significance.

10. Case 2: Google Android

Google Android, Case T-604/18

Facts

The European Commission examined Google's contractual arrangements concerning Android devices, including requirements relating to Google Search, Chrome and application distribution.

Competition significance

The case involved leveraging Google's position across interconnected digital markets.

The General Court largely upheld the Commission's findings while modifying the penalty assessment.

Relevance

Recommendation architectures frequently operate within ecosystems.

A platform controlling:

• operating system;
• search;
• app store;
• browser;
• voice assistant; and
• recommendation services

can potentially use one layer to influence competition at another.

Principle

Competition analysis must consider ecosystem-level leveraging, not merely isolated products.

11. Case 3: Google Android Auto / Enel — European Commission

The Google Android Auto matter concerning Enel X illustrates a different form of digital access problem.

Facts

The dispute concerned access to Google's Android Auto platform and compatibility with an application providing electric-vehicle charging services.

Competition significance

The matter demonstrated how control over an important digital interface can affect the ability of third-party applications to reach consumers.

Relevance to recommendation architecture

Recommendation systems increasingly depend upon:

• operating-system interfaces;
• APIs;
• app stores;
• voice assistants;
• connected-car systems.

If a platform controls the interface through which recommendations reach consumers, access to that interface may become a competitive necessity.

Principle

Control over a digital interface can become an important competition-law issue when it determines access to consumers.

12. Case 4: Bundeskartellamt v Facebook / Meta

Bundeskartellamt, Facebook, B6-22/16

Facts

The German competition authority investigated Facebook's collection and combination of user data from Facebook and other services.

The case ultimately reached the German courts and the European Court of Justice.

Competition significance

The case connected:

• market power;
• data collection;
• consumer conditions; and
• exploitation of data advantages.

Relevance to recommendation systems

Recommendation algorithms depend heavily on behavioural information.

A dominant platform with extensive cross-service data may have a substantial advantage in:

• personalization;
• targeting;
• prediction;
• ranking;
• recommendation accuracy.

Principle

Data-related conduct can have competition significance where it is connected with substantial market power and competitive conditions.

13. Case 5: Bundeskartellamt v Meta Platforms — C-252/21

Meta Platforms Inc. and Others v Bundeskartellamt, Case C-252/21

Facts

The Court of Justice considered the interaction between competition law and Meta's processing and combination of personal data.

Significance

The judgment demonstrated that competition authorities may have to consider the interaction between:

• market power;
• data processing;
• contractual conditions;
• consumer choice; and
• other regulatory frameworks.

Relevance to recommendation architectures

Personalised recommendation engines depend upon extensive behavioural profiles.

Consequently, future competition analysis may examine whether a dominant platform:

• combines data across services;
• restricts meaningful alternatives;
• conditions access upon extensive data processing; or
• obtains a data advantage unavailable to rivals.

Principle

Competition law can intersect with data governance where data practices are connected to the exercise of market power.

14. Case 6: Epic Games v Google

Epic Games, Inc. v Google LLC

Facts

Epic challenged Google's practices concerning the Android app ecosystem, including distribution and payment arrangements.

The US proceedings examined Google's control over important aspects of the Android application ecosystem.

Competition significance

The case illustrates how platform architecture can influence:

• app distribution;
• consumer discovery;
• payments;
• developer access;
• platform dependency.

Relevance to recommendation architectures

Recommendation systems are frequently embedded within app stores and digital ecosystems.

A platform controlling:

distribution + ranking + payment + data

may have multiple opportunities to influence competitive outcomes.

Principle

Competition analysis increasingly needs to examine the architecture of the ecosystem, rather than a single contractual restriction.

15. Case 7: Epic Games v Apple

Epic Games, Inc. v Apple Inc.

Facts

Epic challenged Apple's App Store practices, particularly concerning distribution and payment restrictions.

The litigation examined Apple's control over the iOS app distribution ecosystem.

Relevance to recommendations

Although the principal dispute was not a pure recommendation case, app stores demonstrate how:

• search ranking;
• discovery;
• featured placement;
• default presentation;
• payment infrastructure; and
• platform rules

can collectively determine the commercial visibility of applications.

Principle

Digital-platform competition can depend heavily upon control of consumer discovery and access mechanisms.

16. Case 8: Amazon Marketplace Competition Proceedings

Competition authorities in Europe and elsewhere have examined Amazon's dual role as:

1. marketplace operator; and
2. competing retailer.

Competition concern

The central structural problem is that the platform may simultaneously:

• collect seller data;
• control marketplace rules;
• determine visibility;
• operate logistics;
• advertise products; and
• sell competing products.

Recommendation relevance

If Amazon or another marketplace uses marketplace information to influence recommendation or ranking decisions favouring its own products, the platform could potentially convert informational advantages into algorithmic advantages.

Principle

Vertical integration combined with control over ranking and marketplace data creates distinctive competition concerns.

17. Case 9: Booking.com v Bundeskartellamt

Booking.com BV v Bundeskartellamt, Case C-264/23

This litigation concerning hotel-platform parity clauses is important for understanding how online platforms can structure competition among suppliers.

Relevance

Online travel platforms control several mechanisms simultaneously:

• ranking;
• visibility;
• booking access;
• pricing presentation;
• consumer reviews;
• recommendation.

Even where the specific legal issue concerns parity clauses rather than recommendation algorithms, the case demonstrates why platform architecture must be analysed in the context of how suppliers reach consumers.

18. Lessons From the Case Law

Taken together, these cases indicate several emerging principles.

1. Visibility can be economically significant

A competitor need not be completely removed from a platform to suffer competitive harm.

2. Ranking can be a competitive variable

Algorithmic placement can influence consumer behaviour in a manner comparable to physical shelf positioning.

3. Data can reinforce market power

Data accumulated through one service may improve recommendation capabilities in another.

4. Ecosystems matter

Competition authorities increasingly examine interconnected services rather than isolated products.

5. Interface control matters

APIs, operating systems, app stores and search interfaces may determine whether competitors can effectively reach consumers.

19. Future Regulation of Recommendation Architectures

Future competition regulation is likely to focus on several areas.

A. Algorithmic Transparency

Dominant platforms may increasingly be required to provide regulators with information about:

• ranking criteria;
• recommendation objectives;
• important model changes;
• data sources;
• treatment of affiliated products.

This does not necessarily mean disclosure of the complete source code.

The regulatory objective would instead be contestability and accountability.

20. Algorithmic Auditing

Competition authorities could require independent audits examining:

• self-preferencing;
• discriminatory ranking;
• exclusionary effects;
• changes in visibility;
• treatment of rivals;
• effects of model updates.

The emphasis would shift from:

“What rule did the platform write?”

to:

“What competitive effect does the recommendation architecture systematically produce?”

21. Non-Discrimination Obligations

Dominant platforms could face obligations requiring comparable treatment of:

• affiliated and independent sellers;
• first-party and third-party applications;
• competing content;
• competing service providers.

However, legitimate quality differences would still need to be accommodated.

A platform should generally be able to recommend a product because it genuinely performs better according to an objective criterion.

The competition-law issue arises when supposedly neutral criteria are used to systematically disadvantage rivals.

22. Self-Preferencing Controls

Future regulation may require platforms to distinguish clearly between:

• neutral recommendations;
• sponsored recommendations;
• affiliated recommendations;
• personalised recommendations.

This would reduce the possibility that consumers mistake commercially motivated placement for neutral algorithmic advice.

23. Data Portability and Interoperability

Competition authorities may increasingly support:

• portability of user-generated data;
• interoperability;
• API access;
• transfer of recommendation histories;
• portability of preference profiles.

This could reduce switching costs.

For example:

Consumer changes platform → recommendation history moves with the consumer → new platform can personalise recommendations → switching becomes easier.

24. Consumer-Controlled Recommendation Systems

An important future model is user-selectable recommendation parameters.

Consumers might be permitted to choose:

• chronological order;
• price-based recommendations;
• independent-seller recommendations;
• non-personalised results;
• privacy-oriented recommendations;
• diversity-oriented recommendations.

This can reduce dependence upon the platform's proprietary ranking logic.

25. Recommendation Neutrality

Future regulation could develop a concept of recommendation neutrality.

This would not necessarily require every product to receive equal ranking.

Instead, the platform could be required to avoid unjustified discriminatory treatment based on:

• ownership;
• affiliation;
• strategic interests;
• retaliation against competitors.

26. Algorithmic Collusion

Recommendation architectures may also facilitate coordination between competitors.

Algorithms can process enormous quantities of:

• prices;
• inventory;
• demand;
• competitor behaviour;
• consumer responses.

If competing systems continuously react to each other, they may potentially produce coordinated outcomes without conventional human communication.

Competition law will therefore have to distinguish:

legitimate autonomous optimisation

from

algorithmically facilitated coordination.

27. Dynamic Market Definition

Traditional market definition may become difficult where a single recommendation system simultaneously influences several markets.

For example:

Search → recommendation → advertising → transaction → payment → logistics

may constitute an interconnected competitive ecosystem.

Future enforcement may therefore use:

• multi-sided market analysis;
• ecosystem analysis;
• data-flow analysis;
• attention-market analysis;
• user-dependency analysis.

28. Recommendation Data as an Essential Competitive Input

A future question could be whether certain categories of recommendation data constitute an indispensable competitive input.

Examples include:

• historical consumer preferences;
• product-performance data;
• user reviews;
• transaction data;
• behavioural signals.

Where access to such information becomes indispensable, competition law may have to examine whether refusal to provide access creates foreclosure.

29. AI-Based Recommendation Engines

Generative AI will substantially increase the importance of recommendation regulation.

Instead of showing:

Product A
Product B
Product C

an AI assistant may simply tell the consumer:

“I recommend Product A.”

This creates a major shift.

The intermediary is no longer merely ranking options. It may be making the purchasing decision easier—and potentially decisive—for the consumer.

Therefore, future competition law may scrutinise:

• AI training data;
• commercial affiliations;
• sponsored recommendations;
• model bias;
• hidden self-preferencing;
• default commercial partners;
• exclusive AI distribution agreements.

30. Recommendation Architecture and Gatekeeper Regulation

Modern digital competition regulation increasingly treats certain platforms as gatekeepers or systemically important intermediaries.

Future regulatory frameworks may impose obligations concerning:

• self-preferencing;
• interoperability;
• access;
• data combination;
• transparency;
• user choice;
• switching;
• business-user fairness.

Recommendation architecture is likely to become one of the principal technical mechanisms through which such obligations are implemented.

31. Proposed Regulatory Framework

A comprehensive future framework could operate through six stages:

Stage 1 — Identify the recommendation system

Determine:

• what is being recommended;
• to whom;
• using what data;
• through which interface.

Stage 2 — Identify market power

Examine:

• market share;
• network effects;
• switching costs;
• data advantages;
• user dependency;
• ecosystem control.

Stage 3 — Audit the algorithm

Investigate:

• ranking criteria;
• affiliated-party treatment;
• model objectives;
• data inputs;
• changes over time.

Stage 4 — Measure competitive effects

Examine:

• traffic diversion;
• rival visibility;
• entry barriers;
• foreclosure;
• consumer switching;
• innovation.

Stage 5 — Apply proportional remedies

Possible remedies include:

• transparency;
• non-discrimination;
• interoperability;
• data portability;
• algorithmic auditing;
• choice screens;
• restrictions on self-preferencing.

Stage 6 — Continuous monitoring

Because recommendation models continuously evolve, one-time regulatory approval may be insufficient.

32. Key Future Competition-Law Questions

Courts and regulators are likely to confront questions such as:

1. Can algorithmic demotion constitute exclusionary conduct?
2. When does self-preferencing become abusive?
3. Can recommendation data constitute an essential facility?
4. Should dominant platforms disclose ranking criteria?
5. Can AI-generated recommendations constitute a form of commercial discrimination?
6. How should algorithmic collusion be distinguished from independent optimisation?
7. Should users have a right to choose alternative recommendation systems?
8. Should recommendation histories be portable?
9. Can a platform use data obtained from rivals to compete against those same rivals?
10. Should regulators audit high-impact recommendation models?
11. Can an AI assistant be required to disclose commercial relationships?
12. How should competition authorities evaluate algorithmic foreclosure where causation is probabilistic rather than contractual?

33. Compliance Framework for Platforms

A platform operating a major recommendation system should maintain:

Governance

• algorithmic governance committee;
• competition-law review;
• documented model objectives.

Data governance

• documented data sources;
• data-access controls;
• restrictions on cross-use of competitor information.

Ranking governance

• objective ranking criteria;
• records of material model changes;
• safeguards against unexplained self-preferencing.

Audit

• periodic competition audits;
• bias testing;
• foreclosure testing;
• affiliated-product testing.

Documentation

• model cards;
• version histories;
• decision logs;
• impact assessments.

Complaint mechanism

Businesses should have a process for challenging:

• unexplained demotion;
• discriminatory ranking;
• removal;
• recommendation suppression.

34. Competition-Law Test for Recommendation Architectures

A useful analytical framework is:

Market Power
↓
Control over Recommendation Channel
↓
Algorithmic Advantage / Discrimination
↓
Reduced Rival Visibility
↓
Traffic or Demand Diversion
↓
Foreclosure / Entry Barrier
↓
Consumer and Innovation Effects
↓
Competition-Law Remedy

This framework is particularly useful for analysing AI assistants, marketplaces, search engines, social-media feeds, app stores and digital advertising systems.

Conclusion

The future of competition law will increasingly involve competition inside algorithms rather than merely between companies.

Recommendation architectures can determine which competitors consumers see, which products receive attention, which sellers obtain transactions and which innovations achieve scale. The important legal development is therefore a movement from analysing only market access toward analysing algorithmic access to consumer attention.

The major lessons from cases such as Google Shopping, Google Android, Meta/Facebook, Epic Games v Google, Epic Games v Apple and Booking.com are that digital competition can be affected by ranking, data, ecosystem control, platform interfaces and contractual architecture.

Future competition regulation is consequently likely to focus on algorithmic transparency, self-preferencing, non-discrimination, data advantages, interoperability, portability, algorithmic auditing, AI recommendations and continuous monitoring.

The central question for future competition law will not simply be:

“Can a rival enter the market?”

It will increasingly be:

“Can a rival obtain meaningful visibility and compete fairly within the recommendation architecture through which consumers reach the market?”