Civil Law and Algorithmic Journalism Liability and Defamation Claims in Europe

1. Introduction

Algorithmic journalism liability concerns civil and fundamental-rights disputes arising when artificial intelligence or algorithmic systems are used to create, select, rank, summarise, distribute, personalise, moderate, or amplify journalistic content.

Examples include:

• AI-generated news articles containing false allegations;
• automated summaries that distort a person's statements;
• algorithmically generated headlines that become defamatory;
• AI systems incorrectly identifying a person as a criminal;
• automated recommendation systems repeatedly promoting defamatory material;
• search algorithms increasing the visibility of allegedly defamatory articles;
• AI-generated photographs or videos falsely portraying a person;
• automated moderation incorrectly removing legitimate journalism;
• news platforms being sued for defamatory comments generated or posted by users;
• algorithmic archives continuing to display outdated allegations.

European law therefore requires a balance between freedom of expression and journalism on one side and reputation, privacy and personality rights on the other.

There is currently limited European case law directly concerning generative-AI journalism. The established case law concerning internet journalism, search engines, intermediary liability, reputation and freedom of expression therefore provides the principal doctrinal foundation.

2. Main Legal Framework

Algorithmic journalism disputes can involve several overlapping bodies of law.

European Convention on Human Rights

Particularly:

• Article 8 — private life and reputation;
• Article 10 — freedom of expression;
• Article 6 — fair trial in appropriate procedural situations;
• Article 13 — effective remedy.

EU Charter of Fundamental Rights

Particularly:

• Article 7 — private and family life;
• Article 8 — personal-data protection;
• Article 11 — freedom of expression and information;
• Article 47 — effective remedy and fair trial.

GDPR

Potentially relevant where AI journalism processes personal data:

• Article 5 — data-processing principles;
• Article 6 — lawful processing;
• Article 9 — special-category data;
• Articles 12–15 — transparency and access;
• Article 16 — rectification;
• Article 17 — erasure;
• Article 21 — objection;
• Article 82 — compensation.

Digital Services Act

Particularly relevant to online platforms and very large online platforms concerning:

• content moderation;
• transparency;
• systemic risks;
• recommender systems;
• notice-and-action mechanisms.

National civil law

Depending on the country, claims can involve:

• defamation;
• personality rights;
• privacy;
• tort;
• negligence;
• unjust enrichment;
• injunctions;
• damages;
• correction or retraction.

3. What Is Algorithmic Journalism?

Traditional journalism involves a journalist researching, verifying and publishing information.

Algorithmic journalism can introduce additional stages:

Data collection

↓

Automated analysis

↓

AI-generated or AI-assisted text

↓

Automated headline

↓

Algorithmic ranking

↓

Recommendation

↓

Audience exposure

↓

Potential reputational harm

This creates additional questions concerning who is responsible for each stage.

4. Types of Algorithmic Journalism Liability

A. AI-generated false statement

An AI system generates:

“Person X was convicted of fraud.”

when Person X was never convicted.

Potential issues:

• defamation;
• privacy;
• data protection;
• negligence;
• personality rights.

B. AI-generated implication

The system does not expressly say someone committed a crime but creates an implication through:

• headline;
• image;
• summary;
• juxtaposition;
• automated caption.

The legal question can become whether the overall publication conveys a defamatory meaning.

C. Algorithmic amplification

A truthful or potentially defamatory article may be repeatedly recommended because an algorithm predicts high engagement.

The issue becomes:

Does algorithmic amplification create an additional responsibility beyond the original publication?

This is legally more difficult than ordinary publication liability.

D. Automated comments

Readers may post defamatory statements underneath a news article.

The platform may use:

• automatic moderation;
• keyword filters;
• AI toxicity detection;
• automatic removal;
• recommendation algorithms.

The liability question then concerns the relationship between the publisher, platform and user-generated content.

E. Search-engine amplification

A search engine may associate a person's name with an old allegation.

This can generate:

• privacy claims;
• data-protection claims;
• de-referencing claims;
• reputation-related claims.

The leading authority is Google Spain.

5. Case Law 1 — Delfi AS v Estonia

Delfi AS v Estonia, Application No. 64569/09, ECtHR Grand Chamber, 16 June 2015

This is one of the most important European authorities on online-news liability.

Delfi was an online news portal. Users posted highly offensive and threatening comments below one of its news articles.

The Grand Chamber held that Estonia had not violated Article 10 by imposing liability on Delfi in the particular circumstances.

The Court considered factors including:

• nature of the comments;
• extreme nature of the speech;
• context;
• commercial and professional nature of the portal;
• measures available to the portal;
• steps taken to remove the comments;
• consequences of liability.

Importance for algorithmic journalism

The case provides a foundation for asking whether an online news organisation exercising substantial technological control over its platform can bear responsibility for harmful third-party material.

However, Delfi should not be treated as establishing automatic liability for every online platform or every AI-generated item.

Principle

Intermediary liability must be assessed in light of the nature of the content, the platform's role and the proportionality of the resulting restriction on expression.

6. Case Law 2 — MTE and Index.hu v Hungary

Magyar Tartalomszolgáltatók Egyesülete and Index.hu Zrt v Hungary, Application No. 22947/13, ECtHR, 2 February 2016

This case is particularly important because it demonstrates that Delfi does not create unlimited liability.

An online news portal hosted user comments that were offensive and vulgar toward commercial companies.

The ECtHR found a violation of Article 10 when the applicants were held liable in the circumstances of the case.

The Court distinguished the situation from Delfi, including because the comments were offensive and vulgar but were not comparable to the extreme threats involved in Delfi.

Algorithmic significance

AI moderation systems frequently have difficulty distinguishing:

• criticism;
• insult;
• satire;
• defamatory assertion;
• hate speech;
• threats.

This case supports a contextual approach rather than automatic liability.

Principle

Not every offensive online statement justifies imposing liability on the intermediary hosting it.

7. Case Law 3 — Sanchez v France

Sanchez v France, Application No. 45581/15, ECtHR Grand Chamber, 15 May 2023

This case concerned comments posted on a politician's public Facebook page.

The applicant was convicted because he failed to remove certain hateful comments posted by third parties.

The Grand Chamber examined whether the resulting responsibility was compatible with Article 10.

The Court upheld the interference in the particular circumstances.

Algorithmic significance

The case demonstrates that responsibility can depend on the person's:

• role;
• degree of control;
• context;
• relationship with the communication space;
• ability to respond to unlawful comments.

An AI-powered journalism platform therefore cannot be analysed solely by asking whether it technically generated or hosted the content.

Principle

Responsibility for third-party online speech is highly context-dependent.

8. Case Law 4 — Google Spain

Google Spain SL and Google Inc. v AEPD and Mario Costeja González, C-131/12, CJEU, 13 May 2014

This is a foundational European digital-reputation case.

The case concerned search-engine processing of personal information appearing on third-party webpages.

The CJEU recognised circumstances in which a search-engine operator could be required to remove links from results following a search based on a person's name.

The Court balanced:

• Article 7 Charter — privacy;
• Article 8 Charter — personal-data protection;
• freedom of information;
• interests of the public;
• role of the search engine.

Algorithmic journalism significance

The original journalist or publisher may not be the only technologically significant actor.

An algorithmic search system can substantially determine:

what information about a person becomes visible to the public.

Principle

Algorithmic indexing and ranking can create legally significant responsibility distinct from the original publication.

9. Case Law 5 — Axel Springer AG v Germany

Axel Springer AG v Germany, Application No. 39954/08, ECtHR Grand Chamber, 7 February 2012

This case concerned newspaper reporting about the arrest and conviction of a well-known television actor for drug-related offences.

German courts restricted publication, leading to litigation concerning Articles 8 and 10.

The ECtHR emphasised the important role of the press in democratic society while also recognising the protection of reputation and private life.

Factors relevant to the balancing exercise included:

• contribution to a debate of general interest;
• public status of the person;
• how the information was obtained;
• accuracy;
• consequences of publication;
• severity of the interference with private life.

Algorithmic significance

An AI-generated article should not be assessed solely according to whether it contains technically accurate individual sentences.

The broader question can include:

What did the publication communicate, in what context, and with what impact?

Principle

Journalistic freedom and reputation/privacy must be balanced according to the circumstances of the publication.

10. Case Law 6 — Glawischnig-Piesczek v Facebook Ireland

Glawischnig-Piesczek v Facebook Ireland, C-18/18, CJEU, 3 October 2019

The CJEU considered injunctions requiring Facebook to remove unlawful comments and, under certain conditions, equivalent content.

The Court accepted that EU law could permit measures directed at identical or equivalent unlawful content while addressing the limits of general monitoring.

Algorithmic significance

This case becomes highly relevant where AI moderation is used.

A platform may use automated technology to:

• detect prohibited content;
• identify duplicates;
• identify substantially equivalent content;
• enforce court orders.

But automated filtering also creates risks of:

• false positives;
• over-removal;
• suppression of lawful journalism;
• contextual mistakes.

Principle

Content-removal obligations must be reconciled with freedom of expression and the prohibition on imposing an unlimited general monitoring obligation.

11. Case Law 7 — L'Oréal v eBay

L'Oréal SA v eBay International AG, C-324/09, CJEU, 12 July 2011

The case concerned intermediary responsibility for unlawful material and commercial activity on an online platform.

The CJEU considered the role played by the intermediary and the circumstances in which injunctions could be directed against it.

Algorithmic significance

The case helps establish an important methodological point:

The intermediary's actual role matters.

A passive technical intermediary and a platform that actively structures, promotes or controls content may raise different liability questions.

This reasoning can be relevant when analysing AI-driven journalism platforms.

12. Case Law 8 — Google LLC v CNIL

Google LLC v Commission nationale de l'informatique et des libertés (CNIL), C-507/17, CJEU, 24 September 2019

This case concerned the territorial scope of de-referencing.

The CJEU held that EU law did not generally require worldwide de-referencing merely because a person had obtained de-referencing within the EU, while requiring appropriate measures to prevent or seriously discourage access from the EU to the relevant results.

Algorithmic significance

AI journalism and search systems operate globally.

Therefore, a court may face the difficult question:

Should an algorithmic removal or restriction apply only within Europe or globally?

Principle

Territoriality matters when determining the scope of technological remedies affecting online information.

13. What Constitutes Defamation?

National European legal systems use different concepts and terminology, but an algorithmic defamation claim commonly requires examination of:

1. publication;
2. identification of the claimant;
3. defamatory meaning;
4. unlawfulness;
5. fault or other applicable liability standard;
6. absence of an applicable defence;
7. causation;
8. damage or legally relevant injury.

The precise elements differ considerably between European jurisdictions.

14. AI Hallucination and Defamation

AI hallucination creates a particularly serious problem.

Suppose an AI-generated article states:

“John Smith was arrested for corruption.”

but the underlying data contain no such event.

The claimant may argue:

• the statement is false;
• it identifies the claimant;
• it is defamatory;
• the publisher disseminated it;
• the publisher failed to verify the output;
• reputational damage followed.

The central civil-law question becomes whether the organisation deploying the AI had a duty to verify and control the generated content.

15. Who Is Responsible for an AI-Generated Article?

There may be several potential actors:

AI developer

Created the underlying model.

News organisation

Prompted or deployed the model.

Editor

Approved or failed to verify the article.

Journalist

Used or published the AI-generated material.

Platform

Distributed or recommended the article.

Search engine

Indexed and ranked it.

Hosting provider

Hosted the content.

User

Reposted the material.

Liability cannot automatically be transferred to the AI developer merely because the developer created the technology.

16. Publisher Liability

A traditional publisher may have responsibility for material it publishes.

With AI journalism, courts may examine:

• whether AI was merely a drafting tool;
• whether a journalist reviewed the output;
• whether factual verification occurred;
• whether editorial controls existed;
• whether warnings appeared;
• whether the publisher knew of inaccuracies;
• whether complaints were ignored.

The greater the publisher's editorial control, the stronger the argument that ordinary journalistic standards remain relevant.

17. AI as a Journalistic Tool

AI can be used for:

• transcription;
• translation;
• research;
• summarisation;
• headline generation;
• data analysis;
• article drafting;
• image generation.

Using AI does not necessarily change the fundamental legal obligations surrounding publication.

A useful principle is:

Automation of editorial work does not automatically eliminate editorial responsibility.

18. Algorithmic Headlines

A headline may cause greater reputational harm than the body of an article.

Example:

Body: explains that a person was merely questioned.

AI headline: “Businessman Arrested in Major Fraud Probe.”

The headline can materially alter public understanding.

Therefore, courts may need to examine:

• headline;
• article;
• photograph;
• caption;
• surrounding context;
• search snippet;
• social-media preview.

The publication should generally be evaluated as a whole.

19. AI-Generated Images and Deepfakes

Algorithmic journalism can generate:

• synthetic photographs;
• manipulated images;
• AI video;
• voice cloning;
• fabricated interviews.

Potential claims include:

• defamation;
• privacy;
• personality rights;
• image rights;
• copyright;
• data protection;
• misuse of private information.

The problem becomes especially serious where a synthetic image creates a false factual impression.

20. Algorithmic Recommendation Liability

Suppose an AI recommendation system repeatedly recommends a defamatory article.

The legal question becomes:

Is recommending content equivalent to publishing content?

European law does not provide a single universal answer for every context.

Relevant factors may include:

• degree of editorial control;
• knowledge;
• active participation;
• algorithmic selection;
• monetisation;
• notice;
• speed of response;
• nature of the content.

The Delfi, MTE, Sanchez and Google Spain lines of authority demonstrate why the role of the intermediary must be examined carefully.

21. Defamation Versus Opinion

Algorithmic journalism must distinguish:

Factual allegation

“Company X committed fraud.”

This can potentially be verified as true or false.

Opinion

“Company X's conduct was irresponsible.”

This involves evaluative judgment.

The distinction is important because Article 10 protection is generally stronger for value judgments, although an opinion can still be unlawful depending on its factual basis and context.

22. Public Interest Journalism

European human-rights law gives particularly important protection to journalism concerning matters of public interest.

Relevant considerations include:

• political debate;
• public administration;
• corruption;
• public health;
• corporate wrongdoing;
• criminal justice;
• public officials.

However:

Public interest does not create an automatic defence for false factual allegations.

Accuracy and responsible journalistic practice remain important.

23. Reputation Under Article 8 ECHR

Reputation can fall within the scope of Article 8 where an attack on reputation reaches a sufficient level of seriousness and affects personal identity and private life.

This creates a balancing relationship:

Article 8

Protection of reputation/private life

versus

Article 10

Freedom of expression and journalism.

The Axel Springer and related Strasbourg case law demonstrates the importance of balancing these rights rather than automatically treating one as superior.

24. GDPR and AI Journalism

AI journalism frequently processes personal information.

For example:

• names;
• photographs;
• criminal allegations;
• employment information;
• political activity;
• health information.

GDPR Article 5 requires lawful, fair and transparent processing and includes requirements concerning:

• purpose limitation;
• data minimisation;
• accuracy;
• storage limitation.

The accuracy principle can be particularly significant where AI generates factual claims.

25. Right to Rectification

Where an AI-generated article contains inaccurate personal information, Article 16 GDPR may become relevant.

The individual may seek correction of inaccurate personal data.

This is particularly important because AI errors can be:

• copied;
• indexed;
• republished;
• translated;
• summarised;
• incorporated into other datasets.

One inaccurate AI output can therefore create a reputational feedback loop.

26. Right to Erasure

Article 17 GDPR may potentially provide a route for deletion in appropriate circumstances.

However, erasure is not absolute.

Freedom of expression and information must be considered, and journalistic processing can receive special treatment under national implementations of GDPR Article 85.

Therefore:

A person cannot automatically demand deletion of every negative journalistic reference.

27. The Public-Interest Exception

European data-protection law recognises the need to reconcile data protection with:

• freedom of expression;
• freedom of information;
• journalism;
• academic expression;
• artistic and literary expression.

National law plays an important role in establishing appropriate exemptions and derogations.

This is crucial because otherwise data-protection rights could potentially become a mechanism for suppressing legitimate journalism.

28. Algorithmic Journalism and the Digital Services Act

The DSA adds an important regulatory layer for online platforms.

Relevant issues include:

• content moderation;
• notice-and-action procedures;
• reasons for moderation decisions;
• recommender systems;
• systemic-risk assessments for VLOPs/VLOSEs;
• mitigation of systemic risks.

The DSA should not be confused with a universal civil-law defamation regime.

A regulatory violation does not automatically establish every element of a national damages claim.

29. Causation

Causation can become unusually complicated in algorithmic journalism.

Possible chain:

AI hallucination

↓

News publication

↓

Algorithmic recommendation

↓

Search indexing

↓

Social-media sharing

↓

Public exposure

↓

Employment/business consequences

↓

Reputational damage

A defendant may argue that:

• another publisher caused the damage;
• the claimant was already publicly associated with the allegation;
• users independently shared the information;
• the algorithm merely ranked existing content.

The claimant may therefore need evidence showing the causal contribution of the defendant's system.

30. Damages

Potential damages can include:

Economic damage

• lost employment;
• lost contracts;
• lost customers;
• business losses.

Non-economic damage

• injury to reputation;
• humiliation;
• distress;
• interference with personality rights;
• loss of privacy.

The availability and calculation of damages depend substantially on national law.

31. Injunctions

Courts may consider remedies such as:

• removal;
• correction;
• right of reply;
• de-indexing;
• de-referencing;
• prohibition on republication;
• alteration of a headline;
• deletion of specific unlawful content.

However, injunctions must also respect freedom of expression.

The Glawischnig-Piesczek and Google Spain cases illustrate how European law approaches technological removal and de-referencing remedies.

32. AI Verification Duty

One of the emerging questions is:

Does using AI create an enhanced duty to verify?

A court could potentially examine:

• known hallucination risks;
• reliability of the particular system;
• verification procedures;
• availability of primary sources;
• human editorial review;
• correction mechanisms;
• urgency of publication.

There is currently no single European case establishing a universal rule that every AI-generated journalistic statement must satisfy a particular verification protocol.

Therefore, this should be analysed through existing principles of journalistic responsibility, negligence, defamation, data accuracy and proportionality rather than through an invented AI-specific precedent.

33. AI Journalism and Negligence

A civil negligence claim may ask:

1. Was there a duty of care?
2. Was the AI system deployed reasonably?
3. Were foreseeable risks identified?
4. Was adequate human verification performed?
5. Was the publication foreseeable as harmful?
6. Was the failure causally connected to the injury?

The exact test depends upon the national legal system.

34. Algorithmic Content Moderation

AI moderation can produce two opposite forms of liability risk.

Under-removal

The system fails to detect:

• defamatory comments;
• threats;
• unlawful hate speech.

Over-removal

The system incorrectly removes:

• investigative journalism;
• criticism;
• political speech;
• satire;
• legitimate reporting.

Therefore:

A moderation system can protect reputation while simultaneously threatening freedom of expression.

35. Deepfake Journalism

Deepfake journalism presents a particularly difficult situation.

Suppose AI creates a video showing a public figure apparently making a false statement.

Possible legal questions:

• Is the video presented as authentic?
• Was the synthetic nature disclosed?
• Is the content defamatory?
• Is there an identifiable victim?
• Was there journalistic/public-interest justification?
• Was the content materially altered?
• What damage resulted?

The traditional reputation-versus-expression framework remains relevant.

36. Evidentiary Issues

Algorithmic defamation litigation may require evidence concerning:

• prompts;
• model outputs;
• training or retrieval sources;
• system logs;
• editorial records;
• fact-checking records;
• metadata;
• publication timestamps;
• recommendation data;
• ranking systems;
• moderation logs;
• correction records.

An important issue is preserving the original AI output.

If the article is later corrected, the original generated version may become important evidence.

37. Liability of AI Developers

The AI developer should not automatically be treated as the publisher.

The legal analysis may depend on:

• contractual relationship;
• control over output;
• intended use;
• knowledge of the specific publication;
• foreseeability;
• product liability rules;
• negligence;
• involvement in publication.

For example:

Model developer → News organisation → Journalist → Published article

The news organisation may occupy a very different legal position from the model developer.

38. Liability of News Organisations

The strongest potential liability issues generally arise where the news organisation:

• publishes the AI output;
• presents it as verified journalism;
• fails to conduct reasonable checks;
• knows of an error;
• refuses to correct an established falsehood;
• algorithmically amplifies the material;
• profits from its distribution.

The use of AI does not necessarily provide a defence.

39. Six-Case Analytical Framework

The most useful authorities can be organised as follows:

40. Practical Legal Test for Algorithmic Journalism

A court considering an algorithmic journalism claim can be analysed through the following sequence:

Step 1 — Identify the publication

What exactly was generated or distributed?

Step 2 — Identify the responsible actor

Who:

• generated;
• edited;
• published;
• hosted;
• recommended;
• indexed?

Step 3 — Identify the legal wrong

Is the claim:

• defamation;
• privacy;
• data protection;
• personality rights;
• negligence;
• unlawful processing?

Step 4 — Determine whether the statement is factual or opinion

This affects the Article 10 analysis.

Step 5 — Assess accuracy

Was the statement true, false, misleading or materially incomplete?

Step 6 — Examine public interest

Was there a legitimate contribution to public debate?

Step 7 — Examine journalistic responsibility

Were reasonable verification measures taken?

Step 8 — Examine algorithmic amplification

Did the algorithm materially increase dissemination?

Step 9 — Examine notice

Was the organisation informed about the alleged falsity?

Step 10 — Examine response

Was the content:

• corrected;
• removed;
• amended;
• de-ranked;
• retained?

Step 11 — Balance Articles 8 and 10

Reputation/privacy versus expression/information.

Step 12 — Establish causation and damage

What actual legal harm resulted?

Step 13 — Determine remedy

Possible remedies include:

• damages;
• correction;
• removal;
• de-indexing;
• right of reply;
• injunction.

41. Key Legal Distinction

A particularly important distinction is:

AI-generated content

Who created the statement?

AI-assisted journalism

Who made the editorial decision?

Algorithmic distribution

Who determined its audience?

Algorithmic amplification

Who increased its visibility?

Search indexing

Who made it discoverable?

These may involve different legal actors and different causes of action.

42. Important Principles

1. Using AI does not automatically eliminate traditional journalistic responsibility.
2. AI hallucinations can potentially generate ordinary defamation problems.
3. The publisher, developer, platform and search engine occupy different legal positions.
4. Algorithmic amplification is legally more complex than simple publication.
5. Freedom of expression protects legitimate journalism but is not unlimited.
6. Reputation can receive protection under Article 8 ECHR.
7. Public-interest journalism receives strong Article 10 protection.
8. Context is essential in intermediary-liability cases.
9. Delfi does not establish automatic liability for every online intermediary.
10. MTE demonstrates the importance of distinguishing serious unlawful content from merely offensive material.
11. Google Spain demonstrates that search-engine algorithms can independently affect privacy and reputation.
12. AI moderation creates both under-removal and over-removal risks.
13. GDPR accuracy requirements can become important when AI generates personal information.
14. An AI Act or DSA violation does not automatically equal a private defamation damages award.
15. Causation and damage remain essential to civil liability.

43. Six Cases to Memorise

1. Delfi AS v Estonia

Rule: An online news portal may, in particular circumstances, bear responsibility for unlawful user comments without violating Article 10.

2. MTE and Index.hu v Hungary

Rule: Intermediary liability must be proportionate; offensive or vulgar comments do not automatically justify imposing liability on a portal.

3. Sanchez v France

Rule: Responsibility for third-party online speech depends significantly on the person's role, control and circumstances.

4. Google Spain, C-131/12

Rule: Search engines can have independent responsibilities concerning personal-data processing and de-referencing.

5. Axel Springer AG v Germany

Rule: Journalism, reputation and privacy require a contextual Article 8/Article 10 balancing exercise.

6. Glawischnig-Piesczek, C-18/18

Rule: Removal of unlawful online content can be required in appropriate circumstances, but EU law does not simply authorise unlimited general monitoring.

44. Exam-Ready Conclusion

Algorithmic journalism liability in Europe is developing at the intersection of defamation law, personality rights, data protection, intermediary liability and freedom of expression. There is not yet a comprehensive body of European case law specifically deciding when a generative-AI news article itself creates civil liability. Existing European jurisprudence therefore supplies the principal legal framework.

The Delfi, MTE, and Sanchez cases establish important principles concerning intermediary responsibility for online speech; Google Spain demonstrates the independent legal significance of algorithmic search and indexing; Axel Springer provides the central reputation-versus-journalism balancing framework; and Glawischnig-Piesczek addresses technological removal of unlawful online content.

The central legal formula is:

AI generation → editorial control → factual accuracy → defamatory meaning → public interest → algorithmic amplification → notice → correction/removal → freedom of expression → reputation/privacy → causation → damage → remedy.

The most important point is that AI should not be treated as a legal substitute for the human or organisation responsible for publication. Courts are likely to focus on the actual roles of the AI developer, journalist, publisher, platform and search/distribution system, together with the accuracy of the content, the degree of editorial control, the public interest involved and the proportionality of any restriction on freedom of expression.