AI Assistant Aggregation Layers and Information Funneling Power

1. Meaning

AI Assistant Aggregation Layers refers to digital systems that collect, combine, rank, summarize, and present information from many underlying sources through a single AI interface.

Examples include:

AI search assistants

conversational search engines

shopping assistants

travel and booking assistants

workplace AI copilots

financial-information assistants

AI agents that search several websites and APIs

AI systems that select which sources the user ultimately sees

Information funneling power means the ability of such an intermediary to determine which information reaches the user, in what order, from which sources, and sometimes in what form.

A simplified structure is:

Thousands of information sources → APIs/search indexes → AI aggregation layer → filtering/ranking/summarisation → user

The competition-law issue arises when an AI assistant becomes an important gateway between users and competing information providers and a dominant undertaking uses that gateway to disadvantage rivals.

Importantly, aggregation itself is not unlawful. The legal issue depends on market power, conduct, effects, and objective justification.

2. How AI Information Funneling Works

An AI assistant may perform several stages:

Stage 1 – Information collection

The system obtains information from:

websites;

publishers;

databases;

APIs;

merchants;

maps;

social platforms;

government databases;

proprietary datasets.

Stage 2 – Aggregation

Information from different sources is brought together.

For example:

100 travel websites → one AI travel answer.

Stage 3 – Filtering

The AI decides which information is relevant.

Stage 4 – Ranking

The system decides which sources or products should receive prominence.

Stage 5 – Synthesis

Instead of displaying 20 competing links, the assistant may provide one synthesized answer.

Stage 6 – Transaction or action

Increasingly, the assistant may:

purchase a product;

book a hotel;

select a flight;

recommend financial products;

initiate a payment;

communicate with another service.

Therefore, the AI layer can move from being merely an information intermediary to becoming an economic gateway.

3. Why Funneling Power Matters Under Competition Law

Traditional search engines generally send users to competing websites.

AI assistants can potentially keep the user inside the AI ecosystem.

For example:

Traditional model

User → Search engine → 10 results → External websites

AI-assistant model

User → AI assistant → synthesized answer → recommendation → transaction

The second structure can create greater intermediary power because the AI may determine not merely which result appears first, but whether a source appears at all.

This creates several competition-law questions:

Can the AI operator favour its own services?

Can it systematically exclude competing sources?

Can it impose discriminatory access conditions?

Can it use information obtained from competitors to compete against them?

Can it tie the assistant to another product?

Can it prevent interoperability?

Can it exploit data generated through the AI interface?

Can it use contractual arrangements to prevent rival AI systems from accessing important information?

Can it manipulate recommendations?

Can the AI become an unavoidable gateway to customers?

4. Relevant Competition-Law Framework

In the EU, the principal traditional framework is:

Article 101 TFEU

Relevant where agreements, decisions or concerted practices between undertakings restrict competition.

Article 102 TFEU

Relevant where a dominant undertaking abuses its dominant position.

Potential theories include:

self-preferencing;

tying and bundling;

refusal to supply;

discriminatory access;

leveraging;

exclusionary contractual arrangements;

interoperability restrictions;

margin squeeze;

exploitative data practices where connected with dominance.

The important point is that being a powerful AI assistant is not by itself an infringement.

The authority would normally need to establish the relevant market, dominance, specific conduct, competitive effects or capability of producing them, and the absence of adequate objective justification.

5. Information Funneling as a Form of Intermediation

Suppose an AI assistant aggregates:

Amazon;

independent retailers;

specialist shopping sites;

manufacturer websites;

price-comparison services.

If the assistant consistently directs users toward its own affiliated marketplace, the concern resembles traditional self-preferencing.

The difference is that AI may make the preference less visible.

A traditional search engine might display:

Result 1
Result 2
Result 3

An AI assistant may instead say:

“The best option is X.”

The user may never know that ten alternative suppliers were considered but excluded.

Therefore, ranking power can become recommendation power, and recommendation power can become transactional power.

6. Major Competition Concerns

A. Self-Preferencing

An AI assistant could give preferential treatment to:

its own shopping service;

its own travel service;

its own cloud service;

its own advertising network;

its own financial products;

its own content.

The important precedent is Google Shopping.

7. Case Law

Case 1 – Google and Alphabet v Commission (Google Shopping), C-48/22 P

Court: Court of Justice of the European Union
Date: 10 September 2024

The case concerned Google's treatment of competing comparison-shopping services.

Google's general search results gave preferential positioning to its own comparison-shopping service, while competing services were subject to different ranking treatment. The Court of Justice upheld the finding of abuse and dismissed Google's appeal. (curia)

Principle

A dominant platform's conduct may constitute an abuse where it uses its position in an upstream gateway market to favour its own service in a related market and the conduct is capable of producing anticompetitive effects.

Relevance to AI assistants

This is highly relevant by analogy.

An AI assistant could potentially:

General information gateway → AI answer → own shopping/travel/service product

If the assistant systematically favours its own downstream service, Google Shopping provides an important analytical framework.

However, an AI assistant is not automatically subject to the same conclusion merely because it ranks information.

8. Case 2 – Microsoft v Commission, T-201/04

Court: General Court
Date: 17 September 2007

Microsoft concerned two important forms of conduct:

refusal to provide interoperability information; and

tying Windows with Windows Media Player.

The General Court upheld the Commission's findings concerning Microsoft's refusal to provide interoperability information and its tying conduct. (Infocuria)

Principle

Control over an important technological interface can give a dominant undertaking significant power over complementary markets.

AI relevance

An AI assistant may become an interface between:

users ↔ information providers ↔ applications ↔ services.

If the operator restricts interoperability so that competing AI systems or information providers cannot effectively interact with the ecosystem, Microsoft provides an important analogy.

9. Case 3 – Bronner v Mediaprint, C-7/97

Court: Court of Justice
Date: 26 November 1998

The case concerned access to a newspaper home-delivery system.

The Court established strict conditions for treating a refusal to provide access as an abuse, including circumstances involving indispensability, elimination of effective competition, and lack of objective justification. The later Slovak Telekom judgment expressly recalled these principles. (Infocuria)

Principle

A dominant undertaking does not automatically have to provide competitors with access to infrastructure it has developed for its own business.

AI relevance

Suppose a dominant AI platform controls an information-access infrastructure that competitors allegedly need.

A competitor might argue:

“Without access to this AI aggregation layer, we cannot effectively compete.”

Bronner indicates that mere usefulness is insufficient. The access must satisfy demanding conditions.

Thus:

Important AI infrastructure ≠ automatically an essential facility.

10. Case 4 – Google Android, T-604/18

Court: General Court
Date: 14 September 2022

The Google Android case concerned Android, Google Play Store, Google Search and Chrome, together with agreements involving device manufacturers and mobile network operators.

The General Court examined the ecosystem as a multi-sided platform and considered product bundling, exclusivity payments and anti-fragmentation obligations and their exclusionary effects. (Infocuria)

Principle

Competition analysis can take account of relationships among interconnected products and markets within a digital ecosystem.

AI relevance

An AI assistant may sit inside a much larger ecosystem:

Operating system
↓
AI assistant
↓
Search
↓
Browser
↓
Shopping
↓
Advertising
↓
Cloud
↓
Payments

The greater the ecosystem integration, the greater the possibility that power from one layer can be leveraged into another.

11. Case 5 – Meta Platforms, C-252/21

Court: Court of Justice
Date: 4 July 2023

The case concerned Meta's processing and combination of user data, including off-Facebook data.

The Court held that a national competition authority may, in the context of examining abuse of dominance, take account of potential incompatibility with the GDPR, while respecting the powers and decisions of the competent data-protection authorities. (curia)

Principle

Data practices can become relevant to competition analysis when they form part of conduct by a dominant undertaking.

AI relevance

AI assistants can collect enormous quantities of:

questions;

preferences;

purchase intentions;

browsing information;

interaction history;

location-related information;

professional interests.

This can create a data feedback loop:

More users
↓
More interactions
↓
More behavioural data
↓
Better AI recommendations
↓
More users

Therefore, information funneling can generate both traffic power and data advantages.

12. Case 6 – Slovak Telekom v Commission, C-165/19 P

Court: Court of Justice
Date: 25 March 2021

The case concerned access conditions to telecommunications infrastructure and alleged margin squeeze.

The Court examined the relationship between access obligations, dominance, pricing and exclusionary effects. (Infocuria)

Principle

Control over an important upstream infrastructure can affect competition in downstream markets where competitors depend upon access to it.

AI relevance

Consider:

AI aggregation layer → access to users → downstream information/service providers.

If competing providers must access consumers through the dominant AI interface, discriminatory access or pricing could potentially create downstream foreclosure.

13. Case 7 – Eturas and Others, C-74/14

Court: Court of Justice
Date: 21 January 2016

Travel agencies used a common computerized booking system. The system administrator introduced an automatic restriction on online discounts, and the Court examined whether the conduct could constitute a concerted practice and how electronic evidence should be assessed. (Infocuria)

Principle

Digital systems can facilitate competition-restricting coordination, but the legal conclusion still depends on evidence establishing the relevant elements of the infringement.

AI relevance

AI systems could theoretically:

observe competitors' pricing;

recommend prices;

exchange information through common systems;

automatically respond to competitor behaviour.

Therefore, AI-mediated coordination may raise Article 101 issues.

But similar AI outputs alone do not establish a cartel. Evidence of communication, knowledge, coordination or other legally relevant elements remains important.

14. Information Funneling and Market Foreclosure

A central concern is foreclosure.

Imagine 1 million users normally visit 10,000 websites.

After widespread AI adoption:

1 million users → AI assistant → only 100 selected sources.

The AI intermediary may therefore control a large portion of user attention.

If competing suppliers depend upon visibility, the AI operator could potentially influence:

customer acquisition;

advertising revenues;

website traffic;

transaction volumes;

data generation;

brand visibility.

This is sometimes described economically as gateway power.

15. The "Answer Box" Problem

Traditional search:

User → query → many results → user chooses.

AI assistant:

User → query → AI evaluates sources → AI generates answer → user accepts answer.

The intermediary therefore potentially exercises several forms of control:

The higher the number of layers controlled by one undertaking, the greater the potential for vertical leverage.

16. Data Funneling

AI assistants may also become data funnels.

For example:

10,000 websites
↓
AI assistant
↓
millions of user queries
↓
purchase intentions
↓
behavioural information

The AI operator may learn:

what users want;

what products they compare;

what prices they accept;

which competitors they reject;

which information attracts attention.

This may create a strategic advantage over the underlying information providers.

17. Zero-Click Information and Traffic Diversion

One major concern is the movement from:

Search → website visit

to:

Search → AI answer → no website visit

This may reduce traffic to:

publishers;

specialist websites;

comparison services;

independent retailers;

review platforms.

However, reduced traffic alone does not establish an antitrust infringement.

An authority would need to examine:

market definition;

dominance;

conduct;

competitive effects;

causation;

efficiencies or objective justification.

18. Algorithmic Self-Preferencing

Suppose an AI assistant controls a recommendation algorithm.

It evaluates:

relevance;

price;

quality;

popularity;

user preferences.

But secretly gives an affiliated product an additional ranking advantage.

That creates a potential structure:

Dominant information gateway

affiliated downstream service

preferential algorithm
= potential leveraging concern.

Google Shopping is particularly relevant because the Court considered the treatment of Google's own specialised search service compared with competing services. (Infocuria)

19. Source Exclusion

AI systems may also create source invisibility.

A website might still technically exist and remain accessible.

But if the dominant AI assistant:

does not retrieve it;

does not cite it;

ranks it below competitors;

systematically excludes it;

the practical customer-access problem may be significant.

This raises an important distinction:

Technical accessibility ≠ commercial visibility.

Competition law may therefore need to consider how digital intermediaries affect actual access to customers.

20. Interoperability Risks

AI ecosystems can become closed systems.

For example:

AI Assistant A
↓
proprietary API
↓
proprietary identity system
↓
proprietary payment system
↓
proprietary marketplace

If competitors cannot interoperate effectively, switching costs can increase.

Potential concerns include:

API restrictions;

discriminatory access;

technical incompatibility;

data portability barriers;

identity lock-in;

restrictions on third-party agents.

Microsoft's interoperability case provides an important legal analogy. (Infocuria)

21. Tying and Bundling

An AI assistant could potentially be tied to:

operating systems;

browsers;

cloud services;

office software;

search engines;

advertising services;

payment systems.

For example:

Dominant operating system

mandatory AI assistant

preferential access to system data

restrictions on rival assistants.

The Android and Microsoft cases illustrate why interconnected products and technological integration can be relevant in Article 102 analysis. (Infocuria)

22. Network Effects

AI assistants can benefit from strong network effects.

A simplified cycle is:

More users

↓

More queries

↓

More behavioural information

↓

Better recommendations

↓

More developers and service providers

↓

More functionality

↓

More users

This can produce a self-reinforcing ecosystem.

The competition-law question is not whether network effects are inherently unlawful. They are generally legitimate economic characteristics.

The issue is whether a dominant firm uses exclusionary conduct to reinforce or extend its position.

23. Switching Costs

Users may become dependent on an AI assistant because it remembers:

preferences;

conversations;

contacts;

documents;

workflows;

purchases;

personal settings;

applications.

Switching to another assistant could therefore involve significant costs.

This may increase user lock-in.

24. Information Asymmetry

The AI operator may know:

which sources it considered;

which sources it rejected;

how ranking works;

how often a source is displayed;

user engagement;

conversion rates.

The information provider may know none of these things.

This creates an information asymmetry between the gateway and dependent businesses.

25. AI Aggregator as a "Digital Bottleneck"

The economic structure can be represented as:

Many Information Providers          ↓      Data/API Layer          ↓   AI Aggregation Layer          ↓   Ranking & Filtering          ↓    AI Recommendation          ↓        Users          ↓  Purchases / Attention

If one undertaking controls the central layer, it may become a bottleneck.

The legal significance depends on whether that bottleneck is merely efficient intermediation or is being used to exclude competitors.

26. Possible Article 102 Theories

These are potential theories, not automatic findings of illegality.

27. Objective Justifications

An AI provider may have legitimate reasons for filtering information.

For example:

accuracy;

cybersecurity;

privacy;

copyright compliance;

fraud prevention;

child safety;

prevention of malicious content;

reliability;

technical limitations;

protection against prompt injection;

protection against manipulated websites.

Therefore, an authority must distinguish:

legitimate quality control

from

strategic exclusion of competitors.

28. Essential-Facility Question

The most difficult argument may be:

"This AI aggregation layer has become indispensable."

Bronner demonstrates that European competition law traditionally applies stringent conditions before compelling a dominant undertaking to provide access to its infrastructure. (Infocuria)

Therefore, a claimant should not assume:

“The AI platform is important, therefore access must be mandatory.”

Instead, questions include:

Is the service indispensable?

Is there a realistic substitute?

Would refusal eliminate effective competition?

Can the infrastructure reasonably be duplicated?

Is the refusal objectively justified?

29. Data and Privacy Interaction

AI aggregation can combine data from multiple environments.

For example:

Search data + shopping data + social data + location data + AI conversations.

Meta demonstrates that competition authorities may encounter situations where data-protection issues interact with abuse-of-dominance analysis. (curia)

This does not mean every privacy violation is automatically an antitrust violation.

Rather, the legal systems can interact where the data practice forms part of the competitive conduct under examination.

30. AI Funneling and Article 101

Article 101 may become relevant where multiple companies use common AI systems or information platforms.

Potential examples:

common algorithmic pricing platform;

common AI recommendation infrastructure;

shared information exchange;

algorithmically coordinated prices;

common restrictions on discounts.

Eturas is particularly useful because it shows how a computerized system can become relevant to the assessment of concerted practices and evidence. (Infocuria)

31. Evidence Required in an AI Funneling Case

Competition authorities would potentially examine:

Technical evidence

algorithm architecture;

ranking models;

API logs;

retrieval logs;

source-selection records;

recommendation outputs.

Economic evidence

market shares;

user numbers;

switching costs;

traffic diversion;

conversion rates;

entry barriers.

Commercial evidence

contracts;

exclusivity agreements;

data-sharing arrangements;

pricing;

access conditions.

Internal evidence

strategy documents;

emails;

product plans;

engineering documents;

A/B testing records.

Counterfactual evidence

A particularly important question is:

What would have happened if the allegedly discriminatory ranking or filtering rule had not existed?

Google Shopping specifically involved analysis of potential anticompetitive effects, causation and counterfactual issues. (Infocuria)

32. Six Core Cases to Memorise

33. Key Legal Distinction

The most important distinction is:

AI aggregation ≠ antitrust violation

An AI assistant is allowed to:

aggregate information;

rank information;

remove spam;

recommend products;

summarize sources;

improve user experience.

The competition concern becomes stronger where there is:

Dominance + gateway power + exclusionary conduct + competitive harm + causal connection

Thus:

Information funneling power by itself is not unlawful.

The legal problem arises when the funnel becomes a mechanism for foreclosure, self-preferencing, tying, discriminatory access, or other abusive conduct.

34. Future Competition-Law Problem

AI assistants may change the meaning of market access.

Previously:

Website visibility = customer access

Increasingly:

AI recommendation = customer access

Therefore, the competitive asset may shift from search ranking to inclusion in the AI-generated answer.

This creates a new economic bottleneck:

Source → AI selection → AI answer → consumer decision

The party controlling the middle layer may influence the competitive opportunities of everyone on both sides.

35. Exam-Friendly Formula

AI Aggregation Layer + Large User Base + Data/Network Effects + Source Filtering + Ranking Power + Vertical Integration = Information Funneling Power

Where:

Information Funneling Power + Dominance + Exclusionary Conduct → Potential Article 102 Concern

But:

Funneling Power Alone ≠ Abuse

The authority must still establish the legally relevant elements of the particular theory of harm.

Conclusion

AI assistant aggregation layers represent a new form of digital intermediation in which an AI system can collect information from many sources and reduce it to a single recommendation or answer. This can create substantial information funneling power because users may no longer independently visit or compare all underlying providers.

The most relevant European competition-law precedents are Google Shopping, Microsoft, Bronner, Google Android, Meta Platforms, Slovak Telekom and Eturas. Together, they provide analogies for self-preferencing, interoperability, access to infrastructure, ecosystem leverage, data advantages, foreclosure and algorithmic coordination. (curia)

One-line revision:
“AI information funneling becomes a competition-law issue when a powerful AI gateway uses control over aggregation, ranking, data or recommendations to exclude or disadvantage competing information or service providers.”