LinkedIn, Harassment and the EU Digital Services Act

 What Five Transparency Reports

LinkedIn presents itself as a professional networking platform, yet many members especially women report being subjected to sexual ahrassment and bullying. How well does it protect users from these behaviours, and how consistently does it enforce its rules? Five transparency reports published by LinkedIn under the EU Digital Services Act (DSA) offer an opportunity to examine how the platform handles harassment, what its moderation systems achieve and where significant gaps remain.

The findings raise questions about the increase in harassment reports, the reliability of automated moderation, the availability of language expertise in the human-in-the loop role, and the absence of data showing how harassment affects women.

The reports provide valuable information. They do not, however, tell us how much harassment actually occurs or whether LinkedIn’s systems protect all users equally.

Why LinkedIn publishes transparency reports

The European Commission designated LinkedIn a Very Large Online Platform (VLOP) under the Digital Services Act. Article 42 requires designated platforms to publish transparency reports at least every six months, providing information about content moderation, including user reports, removals, appeals and the use of automated systems.

These reporting obligations are important because they create a degree of public accountability that does not depend solely on individual complaints reporting harm after the fact. Regulators, researchers and users can examine the information and question whether platforms are meeting their responsibilities.

Non-compliance with the DSA can attract fines of up to 6% of a provider’s global annual turnover.

Thanks to Guy Strijbosch for sharing the LinkedIn Transparency Community Report.

I examined five LinkedIn reports published in October 2023, April 2024, October 2024, February 2025 and August 2025, covering the period from October 2023 to August 2025. The analysis used AI to help compare and interpret the published tables.

The first two reports cover shorter periods and have a narrower scope. Where appropriate, figures were standardised to a 30-day period and adjusted for the number of logged-in EU users, allowing more meaningful comparisons over time. These adjusted rates are calculated estimates, not figures published directly by LinkedIn.

The analysis identifies several important trends.

1. Harassment reports are rising faster than the user base

During the first half of 2025, EU users filed 72,291 reports of harassment involving general content, including posts, comments and profiles.

After adjusting for reporting-period length and the number of logged-in users, the rate increased from 17.2 reports per 100,000 users per 30 days in the first half of 2024 to 21.9 in the first half of 2025. This represents an increase of approximately 27%, while the logged-in EU user base grew by around 5%, reaching 54.7 million.

This is a significant increase in the rate of harassment reporting relative to the user base. However, the figures cannot establish why reporting has increased. Possible explanations include a rise in harassment, greater willingness to report, changes to reporting tools or shifts in users’ expectations of acceptable behaviour. The data does not distinguish between these possibilities.

Harassment reports involving job posts show a different pattern. The adjusted rate has remained broadly stable at around 0.3 reports per 100,000 users per 30 days. There were 1,015 such reports in the first half of 2025.

The contrast raises questions about where harassment occurs on the platform and whether different forms of content require different approaches to prevention and moderation.

2. Few harassment reports result in a recorded violation

In the first half of 2025, LinkedIn recorded a violation in:

By comparison, 15.5% of reports concerning fake accounts resulted in a recorded violation. These figures need careful interpretation. A report that does not result in a recorded violation is not necessarily false, malicious or unreasonable. Users may report behaviour that they consider offensive but that does not breach LinkedIn’s rules. Conversely, genuine harassment may go unrecognised or may be difficult to assess from the available evidence.

Harassment can involve context, tone, repeated behaviour, power dynamics and the cumulative impact of apparently minor incidents. A single comment may look innocuous in isolation but form part of a pattern of intimidation.

The published figures do not tell us how accurately LinkedIn distinguishes between legitimate disagreement and harassment, or how consistently its systems recognise behaviour that breaches its rules.

This is an important limitation when interpreting the results.

3. LinkedIn identifies most harassment removals itself

User reports provide only part of the picture. LinkedIn also identifies and removes content proactively, largely through automated detection. In the first half of 2025, LinkedIn reported 7,038 proactive removals of harassment-related content, compared with 2,408 removals following user reports.

This means that the platform’s own detection systems account for most of the harassment removals recorded in this category. Users therefore cannot be expected to identify or report every incident for the platform to take action.

Proactive moderation is important, but the figures also make the quality of detection systems central to the effectiveness of enforcement. Automated systems must identify harmful content while avoiding the removal of legitimate contributions. The reports do not provide enough information to assess how accurately LinkedIn achieves this balance specifically for harassment.

One particularly striking finding concerns manual investigations. The number of harassment items identified through manual investigation varied considerably across the five reporting periods: 4, 19, 482, 48 and 1,508.

The reports do not explain these fluctuations. They may reflect changes in investigative activity, reporting scope or classification, but the available figures do not establish the reason.

LinkedIn should clarify what accounts for these variations and how manual investigations complement automated moderation.

4. Language coverage raises questions about consistency

Content moderation depends on more than identifying individual words. Harassment can involve irony, coded language, cultural references, slang, implied threats and patterns of behaviour that are difficult to interpret without context.

The transparency reports raise questions about whether LinkedIn’s moderation systems operate consistently across languages. Globally, the proportion of appealed automated removal decisions that were overturned remained around 4%. However, the picture differed by language. Between the first halves of 2024 and 2025, the proportion of decisions overturned increased in 12 of the 14 languages reported.

For example:

Nevertheless, the pattern warrants further investigation. It raises questions about whether automated moderation performs consistently across languages and whether some users face a greater risk of incorrect decisions.

5. Moderator capacity is uneven

The reports also indicate that LinkedIn’s global moderator headcount more than doubled between August 2023 and June 2025, reaching 1,757. Only around 180 were based in the EU.

Of the 15 EU languages supported by LinkedIn, six had no moderator with advanced proficiency above B1. : Czech, Danish, Greek, Hungarian, Romanian and Swedish. This does not, by itself, prove that users of these languages receive poorer moderation. However, it raises a legitimate question about how LinkedIn ensures that culturally and linguistically complex cases are assessed appropriately.

Automated tools can help platforms manage large volumes of content, but language proficiency and contextual understanding remain important when assessing harassment.

6. Appeals are taking longer

The reports indicate that approximately 29% of appeals were granted in the first half of 2025. The median decision time increased from 49 minutes in 2023 to approximately 17 hours. The figures suggest that appeals remain an important part of the moderation process, while the increase in decision time raises questions about the capacity and efficiency of that process.

The overall appeal figures should not be interpreted as harassment-specific results unless LinkedIn provides a separate breakdown. The reports do not establish whether appeals concerning harassment follow the same pattern.

What the reports don’t tell us

The most significant limitations concern the information that is missing.

1. There is no gender breakdown

LinkedIn has reported that women account for approximately 43% of its EU users,  yet the reports do not provide a gender breakdown of harassment reports, moderation outcomes or appeals. Without this information, it is impossible to determine from these reports whether women are disproportionately targeted, whether they experience different forms of harassment or whether their reports are handled differently.

Research on other online platforms has documented gendered patterns of abuse. Those findings cannot automatically be transferred to LinkedIn, but they underline why platform-specific data matters. A gender breakdown would allow researchers and regulators to examine whether professional networking platforms reproduce or amplify existing inequalities.

2. Sexual harassment is not separately identified

The published categories do not provide a distinct breakdown for sexual harassment that allows it to be examined separately from other forms of harassment. This limits the ability to understand the scale and nature of sexual harassment on LinkedIn, how it is reported and whether moderation outcomes differ by type of behaviour.

A more detailed classification would make the data more useful for prevention and accountability.

3. There is no breakdown by EU member state

The reports do not provide the necessary country-level breakdown to establish whether reporting rates or enforcement outcomes vary across EU member states. National differences in language, culture, reporting practices and moderation requirements could all be relevant. Without a country-level view, these differences cannot be properly investigated.

4. We cannot identify who reports, who is targeted or who is responsible

The published data does not allow us to establish the characteristics of those reporting harassment, those targeted or those responsible for the behaviour. This limits our understanding of patterns of repeated harassment, the groups most affected and the circumstances in which incidents occur.

5. Direct messages remain an unanswered question

The first two reports explicitly excluded private messages from their scope and anecdotally my observation is this is where the most abuse takes place. Later reports don’t clearly indicate whether direct messages are included in the relevant figures.

Private messages are an important area to examine because harassment may occur away from public posts and comments, including through unwanted sexual remarks, intimidation and repeated unwanted contact.

The reports do not establish how prevalent such behaviour is on LinkedIn. Nor does the absence of a clear statement establish that private messages are excluded from every later report. LinkedIn should clarify what is covered, what is excluded and how users can report harassment occurring in private communications.

Reporting data is not prevalence data

The reports measure reported incidents and enforcement activity, not the total amount of harassment experienced by users which is a fundamental limitation. Some incidents may never be reported because people fear retaliation, doubt that action will be taken or do not want to draw attention to the behaviour.

Others may be reported but not result in a recorded violation. Still others may be detected proactively without a user complaint. The figures provide a useful view of how LinkedIn processes and responds to reported content. They cannot, on their own, establish the prevalence of harassment or the extent to which users feel safe on the platform.

Questions LinkedIn and the European Commission should address

The five reports provide a basis for more specific questions about transparency and accountability.

Gender and harassment

Direct messages and reporting scope

Moderation and language coverage

Resources and accountability

These are not questions that the current reports can answer. They are questions that better transparency would allow regulators, researchers and users to investigate.

Other Markets

The Digital Services Act (DSA) provides a common framework for online platform oversight across the EU, including transparency reporting, content moderation and mechanisms for users to challenge certain decisions.

Howevr, other countries have introduced their own online safety and platform regulations, but the rules, reporting requirements, definitions of harmful content and enforcement mechanisms vary and are not necessarily equivalent to the DSA. Global platforms such as LinkedIn may operate under several regulatory frameworks simultaneously.

The reports analysed here relate to LinkedIn’s EU operations and do not offer a picture of how the platform handles harassment in other markets worldwide.

What LinkedIn users in Europe can do

Transparency reporting is important, but users also have practical options when they experience or witness harassment.

1. Report the behaviour accurately

Use LinkedIn’s reporting function and select the reason that most accurately describes the behaviour. The reason selected affects how a report is classified, so accurate reporting helps the platform record the nature of the complaint.

Where possible, retain the original URL, date, relevant context and screenshots before reporting. Once a comment is reported it’s no longer visible to you and it may subsequently be removed.

2. Challenge a decision you believe is wrong

If LinkedIn rejects a report or makes another moderation decision you believe is incorrect, check the decision notice for the available review or complaint procedure.

The Digital Services Act provides mechanisms for users to challenge certain content-moderation decisions, including internal complaint handling and eligible cases referred to certified out-of-court dispute settlement bodies.

The transparency reports record appeals and dispute outcomes, but overall figures should not be treated as the success rate for harassment complaints specifically.

3. Contact the relevant national regulator

Each EU member state has a Digital Services Coordinator responsible for aspects of DSA implementation and enforcement. Users can find information about the relevant national authority through the European Commission’s Digital Services Coordinators directory.

The appropriate route depends on the nature of the concern. A platform moderation complaint and a report of potentially unlawful conduct may require different procedures.

4. Ask for better data

Users, researchers, employers and professional associations can ask LinkedIn to publish more detailed information about harassment, including gender-disaggregated data, sexual harassment categories, language-specific enforcement outcomes and the treatment of private messages.

They can also raise these questions with their elected representatives and relevant regulators.

Better data would help establish whether the platform’s systems are working consistently and where improvements are needed.

Many people especially men underestimate the impact of online sexual harrassment and the feeling of betrayal someone experiences when an organisations fails to offer the protection you feel you need.

If you are struggling get in touch for an adhoc coaching session to create a simple stategy to deal with it as a starting point.