Identity Ossification in Recommendation Systems

1. Introduction

Identity ossification in recommendation systems describes a situation in which a digital platform's recommendation algorithm increasingly fixes a user into a persistent profile or behavioural identity and then repeatedly recommends content, products, services, or information consistent with that profile.

The problem is not simply that recommendations are personalised. The competition concern arises when historical behavioural data becomes self-reinforcing:

past behaviour → inferred identity → personalised recommendations → constrained exposure → new behaviour → stronger identity inference.

Over time, the recommendation system may stop treating the user as capable of changing preferences. The user's algorithmic identity becomes “ossified”—stable, difficult to escape, and increasingly determinative of what the user can discover.

This has important implications under competition law, consumer protection, data governance, privacy law, and digital-platform regulation.

2. Meaning of Identity Ossification

Identity ossification occurs where a platform:

1. collects extensive behavioural data;
2. constructs a persistent user profile;
3. categorises the user into behavioural or commercial segments;
4. uses those categories to rank recommendations;
5. repeatedly exposes the user to similar options;
6. interprets subsequent behaviour as confirmation of the original profile; and
7. thereby makes the profile progressively more difficult to change.

For example, suppose a user watches several videos about conservative financial investments.

The platform may infer:

“Low-risk investor.”

It then recommends:

• fixed-income products;
• conservative investment channels;
• low-risk financial content;
• advertisements directed at risk-averse users.

Because the user receives little exposure to alternative investment content, the system may interpret the continuing behaviour as evidence that its original classification was correct.

The result is an algorithmic feedback loop.

3. Identity Ossification Versus Ordinary Personalisation

Personalisation is not inherently problematic.

Thus, the competition-law concern is not “personalisation = unlawful.”

The concern is:

personalisation + persistence + feedback loops + reduced choice + exclusionary effects.

4. How Identity Ossification Develops

Stage 1 — Data accumulation

The platform collects:

• clicks;
• searches;
• watch time;
• purchases;
• location information;
• device information;
• social connections;
• browsing behaviour;
• interaction history.

The resulting dataset provides the foundation for identity classification.

Stage 2 — Identity inference

The platform predicts attributes such as:

• interests;
• purchasing propensity;
• political or cultural interests;
• price sensitivity;
• financial behaviour;
• brand loyalty;
• susceptibility to particular advertising;
• preferred content genres.

The user may never have expressly chosen these classifications.

Stage 3 — Recommendation filtering

The recommendation engine ranks content according to the inferred identity.

A user classified as a particular “type” therefore receives fewer opportunities to encounter material outside that classification.

Stage 4 — Feedback

The user's available choices affect subsequent behaviour.

If the system predominantly displays one category of content, the user naturally interacts more with that category.

The platform then treats the interaction as evidence that its original classification was correct.

Stage 5 — Ossification

Eventually, the identity becomes difficult to change.

The system effectively says:

“You are the kind of user who likes X, therefore we will continue showing X because you like X.”

The circularity is important.

5. Why Identity Ossification Can Become a Competition Problem

Identity ossification becomes particularly significant where the platform possesses substantial market power.

A dominant platform may control:

• the user identity layer;
• recommendation infrastructure;
• advertising infrastructure;
• search ranking;
• application distribution;
• payment systems;
• data collection;
• interoperability;
• access to competing services.

The recommendation system can then become a gatekeeping mechanism.

The platform does not necessarily need to expressly prohibit competitors.

Instead, it can reduce competitors' visibility by systematically ranking them lower for users whose algorithmic identities make them appear less likely to engage with them.

6. Foreclosure Through Recommendation

Consider a dominant marketplace.

A consumer previously purchased Brand A.

The recommendation algorithm therefore predicts:

Brand A affinity = 92%.

Brand B offers a cheaper and potentially superior product, but the system predicts only:

Brand B affinity = 8%.

The algorithm repeatedly recommends Brand A.

Brand B consequently receives:

• fewer impressions;
• fewer clicks;
• fewer transactions;
• less behavioural data;
• weaker future recommendation scores.

This creates a recommendation disadvantage feedback loop.

The incumbent becomes stronger precisely because it is already strong.

7. Identity Ossification and Network Effects

Identity ossification can amplify network effects.

A platform with more users generates more behavioural data.

More data improves recommendation accuracy.

Improved recommendations increase user engagement.

Higher engagement produces additional data.

This creates:

Users → Data → Identity inference → Recommendations → Engagement → More data

The resulting data advantage can become a barrier to entry.

A new competitor may have an equally good product but lack the historical behavioural data necessary to compete with the incumbent's recommendation engine.

8. Identity Ossification and Switching Costs

Ossified identities can also create algorithmic switching costs.

When a consumer moves to a competing platform, the new platform may not know:

• what the consumer likes;
• what the consumer dislikes;
• preferred price ranges;
• purchasing patterns;
• content preferences;
• social relationships;
• previous interactions.

The consumer therefore experiences a less personalised service.

This can discourage switching.

The incumbent effectively possesses a valuable behavioural memory that competitors cannot easily reproduce.

9. Data Portability and Identity Portability

A particularly important issue is whether users can transfer their algorithmic identity.

Traditional data portability may allow users to transfer:

• account information;
• contacts;
• photographs;
• documents.

But the recommendation profile may remain behind.

For example:

“User prefers independent films, purchases environmentally sustainable products, usually responds to discounts, and dislikes long-form advertisements.”

Such an inferred profile may not be portable.

Consequently, formal data portability may exist while functional identity portability remains absent.

10. Competition-Law Theories

Identity ossification may potentially be analysed through several competition-law theories.

A. Abuse of dominance

A dominant platform may use its recommendation infrastructure to disadvantage competing suppliers.

B. Self-preferencing

The platform may systematically recommend its own products or services to users whose profiles predict high conversion rates.

C. Tying and leveraging

Control over identity and recommendation services may be leveraged into adjacent markets.

D. Refusal to provide access

In exceptional circumstances, refusal to provide access to essential data or interoperability may raise access-related concerns.

E. Exclusionary data advantages

Persistent behavioural data may create an advantage that competitors cannot realistically replicate.

F. Exploitative conduct

Ossified identities may facilitate highly targeted commercial exploitation, particularly where users have limited awareness or control over profiling.

11. Important Case Laws

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

The Google Shopping litigation is highly relevant because it established that a dominant search platform can potentially distort competition through the manner in which it positions and displays results.

The fundamental lesson is that ranking and visibility can themselves constitute an important competitive parameter.

For identity ossification, the analogy is significant.

A recommendation platform may not formally exclude competing products. Instead, it can determine which products are visible to particular categories of users.

Principle: Algorithmic ranking can have exclusionary competitive consequences even without an express prohibition on competitors.

2. Google Android — European Commission / General Court

The Google Android case concerned Google's conduct surrounding Android and related services, including restrictions affecting competing search and distribution opportunities.

The case demonstrates how control over one important digital layer can be leveraged into adjacent markets.

For identity ossification, the relevant lesson is that:

Control over an ecosystem layer can reinforce market power in another layer where user access and defaults are important.

If identity and recommendation functions become embedded across an ecosystem, the same leverage theory may arise.

3. Google Search (AdSense) — European Commission

The Google AdSense case concerned contractual restrictions that allegedly limited the ability of third-party websites to display competing search advertisements.

The case is relevant to identity ossification because it illustrates how contractual and technological arrangements can reinforce an incumbent's position in an adjacent digital market.

Where a dominant platform controls the recommendation interface and simultaneously restricts alternative recommendation or discovery mechanisms, similar foreclosure questions can arise.

4. Amazon Marketplace — European Commission

The European Commission's Amazon investigations concerning marketplace data and the use of non-public seller information are highly relevant to data-driven platform competition.

The underlying concern illustrates an important structural problem:

the platform operator may simultaneously operate infrastructure used by competitors while possessing extensive information generated through that infrastructure.

For recommendation systems, this can become particularly powerful.

The platform can combine:

• seller information;
• consumer behaviour;
• transaction history;
• recommendation data;
• conversion information.

Such informational advantages can reinforce the platform's recommendation position.

5. Meta Platforms / Facebook — European Commission

The European Commission's competition-law proceedings involving Meta illustrate the relationship between data-intensive platform ecosystems and competition.

The broader relevance is that a platform's competitive position can depend upon its ability to collect and combine data across services.

Identity ossification adds another dimension:

cross-service data integration → richer identity profile → stronger personalisation → greater user dependence → stronger ecosystem position.

Thus, identity is potentially an important competitive asset rather than merely a privacy issue.

6. Bundeskartellamt — Facebook / Meta (2019)

The German Facebook case is particularly important because the Bundeskartellamt examined the relationship between market power and extensive data collection/combination.

The case demonstrated that competition authorities may consider data practices in assessing the exercise of market power by a dominant digital platform.

Its relevance to identity ossification is substantial.

A platform that combines information from multiple sources may construct an increasingly comprehensive behavioural identity.

That identity can then be used to optimise recommendations, advertising and engagement.

Principle: Data collection and combination can be relevant to the competitive assessment of a powerful digital platform.

7. FTC v. Meta Platforms

The United States litigation concerning Meta's acquisitions of Instagram and WhatsApp is relevant to the broader question of how control over social and communication ecosystems can reinforce platform power.

Although the case does not directly establish a doctrine of “identity ossification,” it demonstrates the importance of analysing digital markets where:

• user relationships;
• network effects;
• data;
• engagement;
• platform ecosystems

interact.

Identity-based recommendation systems can strengthen those same network effects by making users' behavioural histories increasingly valuable and difficult to reproduce elsewhere.

8. United States v. Google

The U.S. Google search litigation is also highly relevant because it concerns the preservation of dominance in search through distribution and default arrangements.

The broader lesson for recommendation systems is that access and visibility are central competitive resources.

An identity-ossified recommendation system may create a similar problem at the discovery layer:

competitors technically remain available, but algorithmic exposure to them becomes substantially diminished.

12. Theoretical Competition Model

Identity ossification can be represented as:

It+1=f(It,Bt,Dt)I_{t+1}=f(I_t,B_t,D_t)

where:

• ItI_t = user's inferred identity at time tt;
• BtB_t = observed behaviour;
• DtD_t = available recommendation/discovery data.

The recommendation system then determines:

Rt=g(It,Pt)R_t=g(I_t,P_t)

where:

• RtR_t = recommendations;
• PtP_t = available products/content.

The user's subsequent behaviour depends partly upon those recommendations:

Bt+1=h(Rt,Ut)B_{t+1}=h(R_t,U_t)

Thus:

It→Rt→Bt+1→It+1I_t \rightarrow R_t \rightarrow B_{t+1} \rightarrow I_{t+1}

The system therefore creates a self-reinforcing identity loop.

13. The “Discovery Collapse” Problem

A particularly important competition concern is discovery collapse.

In a competitive market, consumers should have opportunities to discover new suppliers.

Ossified recommendation systems can reduce:

• serendipitous discovery;
• cross-brand comparison;
• exposure to new entrants;
• exposure to alternative business models;
• switching opportunities.

The consequence may be a reduction in contestability even if the user technically remains free to choose another product.

14. Entrant Disadvantage

New entrants face a particularly difficult problem.

An incumbent may possess years of behavioural data.

A new platform may possess:

• better technology;
• lower prices;
• better products;
• greater privacy;
• innovative business models.

But its recommendation engine lacks sufficient historical information.

Consequently:

Incumbent data advantage→better prediction→higher engagement→more data\text{Incumbent data advantage} \rightarrow \text{better prediction} \rightarrow \text{higher engagement} \rightarrow \text{more data}

This can produce a data-driven entry barrier.

15. Identity Ossification and Self-Preferencing

Suppose a platform operates its own marketplace.

It classifies users into identity groups and then optimises recommendations.

The platform could theoretically manipulate the recommendation function so that its own products receive favourable exposure.

The problematic mechanism becomes:

dominant identity database + recommendation control + own downstream products

This may produce a particularly powerful form of self-preferencing because the platform can tailor the preference not merely globally, but user-by-user.

16. Identity Ossification and Dark Patterns

The problem can become stronger where recommendation systems use:

• infinite scrolling;
• autoplay;
• personalised notifications;
• personalised discounts;
• urgency messages;
• behavioural nudging;
• default selections.

The user may believe they are freely exploring the market while the platform is progressively narrowing the commercially relevant universe presented to them.

17. Privacy and Competition Intersection

Identity ossification illustrates why privacy and competition cannot always be treated as completely separate regulatory fields.

A privacy concern asks:

What information does the platform collect and how does it use it?

A competition question asks:

Does control over that information enable the platform to exclude rivals or entrench market power?

An identity-ossification problem can involve both.

18. Remedies

Possible remedies include:

1. Profile transparency

Platforms could disclose the principal factors influencing recommendation classifications.

2. Identity reset

Users could be permitted to reset recommendation histories.

3. Profile correction

Users could challenge inaccurate inferred characteristics.

4. Algorithmic diversity

Platforms could introduce minimum levels of recommendation diversity.

5. Interoperability

Users could potentially transfer relevant recommendation information to competing services.

6. Data portability

Portability could include meaningful behavioural information rather than merely raw account data.

7. Non-discrimination

Dominant platforms could be restricted from manipulating recommendation systems to disadvantage competing products.

8. Independent auditing

Competition authorities could require auditing of recommendation systems where there is evidence of systematic foreclosure.

9. Structural separation

In extreme cases, separation between infrastructure and downstream commercial operations could be considered.

19. Test for Competition Authorities

A useful analytical framework is:

Step 1 — Identify the relevant market

Is the market:

• social media;
• search;
• online marketplace;
• app distribution;
• video;
• music;
• advertising;
• travel;
• financial services?

Step 2 — Identify identity control

Does the platform control a substantial amount of behavioural identity data?

Step 3 — Examine recommendation dependence

How heavily do consumers rely upon the algorithm for discovery?

Step 4 — Measure persistence

Can users realistically change or reset their profiles?

Step 5 — Examine foreclosure

Are competing suppliers systematically disadvantaged?

Step 6 — Examine feedback effects

Does reduced exposure produce poorer performance data for rivals?

Step 7 — Examine entry barriers

Can a new competitor realistically reproduce the incumbent's behavioural dataset?

Step 8 — Examine efficiencies

Does the recommendation system genuinely improve consumer welfare?

Step 9 — Examine less restrictive alternatives

Could similar recommendation benefits be achieved through a less exclusionary design?

20. Key Legal Principle

The central legal insight can be expressed as follows:

A recommendation system becomes a potential competition concern when personalisation ceases merely to reflect consumer preferences and instead begins to shape, restrict, and permanently reinforce those preferences in a manner that protects platform market power.

The strongest cases will therefore not arise merely from the existence of algorithmic profiling.

They will arise where there is evidence of:

dominance + persistent profiling + recommendation control + reduced discovery + feedback loops + foreclosure or exploitation.

21. Conclusion

Identity ossification in recommendation systems represents a sophisticated form of digital market entrenchment.

The problem is not simply that an algorithm knows what a consumer likes. The deeper concern is that the algorithm may gradually construct the consumer it claims merely to observe.

Once the inferred identity becomes persistent, recommendation systems can:

• reduce consumer discovery;
• increase switching costs;
• reinforce incumbents;
• disadvantage new entrants;
• strengthen network effects;
• create data-based entry barriers;
• facilitate self-preferencing;
• increase ecosystem dependence.

The most relevant jurisprudential lessons can be drawn from Google Shopping, Google Android, Google AdSense, Amazon Marketplace, Facebook/Meta data cases, and U.S. Google/Meta litigation. None creates a standalone doctrine called “identity ossification,” but together they provide the legal building blocks for analysing ranking power, data advantages, ecosystem leverage, self-preferencing, entry barriers, and digital entrenchment.

Accordingly, identity ossification should be understood as a potential competition-law mechanism through which control over behavioural identity becomes control over market discovery itself.