RTFP

LinkedIn’s content moderation governance framework incorporates cross‑functional oversight and collaboration, with the aim that moderation decisions are applied in an objective, non‑arbitrary, and timely manner. LinkedIn’s content policies are authored and maintained by LinkedIn’s Content Policy team. The Content Policy team is a global team of policy management professionals, and sits within LinkedIn’s Legal & Public Policy organisation. Content review is conducted and LinkedIn’s policies are applied to content by LinkedIn’s Trust Review Operations team. The Trust Review Operations team is part of LinkedIn’s Trust & Safety team, which sits within LinkedIn’s Engineering organization. When appropriate, content moderation decisions may be escalated to the Content Policy team for expert policy guidance and review. Member State orders for content moderation, Article 16 notices for illegal content, and disputes submitted to out-of-court dispute settlement bodies are handled by LinkedIn’s Legal Regulatory Operations team, which is part of LinkedIn’s Legal & Public Policy organisation.

As described above, LinkedIn applies a three-layer, multidimensional approach to moderate content on LinkedIn: - The first layer of protection is automated and proactive prevention. When a member attempts to create a piece of content on LinkedIn, various calls (or signals) are sent to LinkedIn’s machine learning services. These services aim to automatically filter out certain policy-violating content at the time of creation. - The second layer of protection is a combination of automated and human-led detection. LinkedIn’s second layer of moderation detects content that is likely to be violative but for which LinkedIn is not sufficiently confident to warrant automatic removal, and sends it for human review. - The third layer of protection is human-led detection. In addition to LinkedIn's Article 16 notice and action mechanism for illegal content, if users locate content that they believe violates LinkedIn’s policies, they are able to report it using LinkedIn’s in-product flagging functionality. During the reporting period, LinkedIn’s detection methods for violative content include at least the following: - LinkedIn employs automated means to attempt to detect violative content, discussed further in the indicators below. - LinkedIn investigations may manually identify violative content. - LinkedIn may receive user flags via its in-product flagging functionality regarding violative content. - LinkedIn may receive notices via its Article 16 notice and action mechanism regarding violative content. - LinkedIn may receive orders from Member State authorities regarding violative content. When LinkedIn detects potentially violative content, LinkedIn evaluates the content against its policies. Such content may be queued for review by human moderators or reviewed by LinkedIn’s automated system. When LinkedIn determines content violates its policies, it applies the actions outlined in the response to Indicator 1 above.

As discussed above, information regarding the number of internal and external content moderators for the reporting period is reported in Sheet 9 (9_human_resources). To calculate the number of internal and external moderators for the reporting period, LinkedIn queries its human resources systems for the number of moderators staffed as of the last day of the reporting period (that is, the number of moderators staffed as of 31 December 2025). As provided by Commission Implementing Regulation (EU) 2024/2835, counts are reported in full-time equivalent units (i.e., a full time moderator is counted as 1). With respect to linguistic expertise, LinkedIn reports the number of moderators that have sufficient linguistic expertise in the indicated language. Sufficient linguistic expertise is defined as at least level B2 or above according to the Common European Framework of Reference for Languages (CEFR). Language expertise for internal moderators is reported based on the attestation of the employee moderator. Language experience for external moderators is reported based on the attestation of the service provider.

Information regarding the number of internal and external content moderators for the reporting period is reported in Sheet 9 (9_human_resources). LinkedIn utilizes both internal employee moderators for content moderation and external contracted moderators via a service provider for content moderation. For internal moderators, LinkedIn requires that moderators have a bachelor’s degree or equivalent, with a master’s degree preferred. For moderator positions, LinkedIn requires one to two-plus years of relevant experience; for senior moderator positions, LinkedIn requires four-plus years of relevant experience. While LinkedIn does not reference the European Qualifications Framework in job posts, this may be considered to correspond to EQF level 67 (bachelor’s degree). For external moderators, hiring is performed by the service provider. LinkedIn requires that the service provider provide moderators with at least a bachelor’s degree, with a master’s degree preferred. For entry-level moderator positions, LinkedIn requires zero to two years relevant experience; for senior moderator positions, LinkedIn requires two or more years of relevant experience. While LinkedIn does not express requirements to the service provider in the form of the European Qualifications Framework, this may be considered to correspond to EQF level 6 (bachelor’s degree). All moderators, both internal and external, undergo training and quality assurance audits, described further in the indicator below.

Information regarding the indicators of accuracy and possible error rate of the automated means is available in Sheet 8 (8_automated_means). Specifically, for the reporting period, the precision for moderation decisions made by LinkedIn’s own initiative automated means is estimated to be 98 percent. The accuracy for LinkedIn’s own initiative automated means is estimated to be 99 percent or above. The recall for LinkedIn’s own initiative automated means is estimated to be 87 percent. Precision is a measure of the proportion of correct enforcement actions applied by LinkedIn's automated system (e.g. when the automated system decides to remove a piece of violative content or restrict an account). A high precision means fewer false positives. Estimated precision rates are based on the number of enforcement actions made by the automated system that are overturned following appeal (i.e. the automated system made an error). To calculate precision, LinkedIn takes one minus the ratio of the number of overturned automated enforcement actions to the total number of appealable automated enforcement actions during the reporting period. Accuracy is a measure of the proportion of correct moderation decisions made by LinkedIn's automated system -- including both decisions that a piece of content is violative and decisions that a piece of content is not violative. A high accuracy means fewer false positives and false negatives. To calculate accuracy, LinkedIn takes the number of correct moderation decisions by the automated system (both that content is violative and non-violative), divided by the number of decisions made by the automated system during the reporting period. A moderation decision is incorrect if the automated system's moderation decision is overturned or changed as of the end of the reporting period. (For example, if the automated system evaluates a piece of content and determines it does not violate LinkedIn's policies, and the content is later restricted (e.g., because it is reported by a user and found to be violative, or otherwise restricted by LinkedIn), that counts against the system's accuracy. Likewise, if the automated system restricts a piece of content and the content is later reinstated following appeal, that also counts against the system's accuracy.) Recall is a measure of the proportion of violative content found and actioned by LinkedIn's automated system, as opposed to other methods. A high recall means fewer false negatives (missed positives). To calculate recall, LinkedIn takes the number of correct enforcement actions by the automated system, divided by the total number of correct enforcement actions by any method (whether human, automated system, or otherwise). Correct enforcement actions are computed by multiplying all machine enforcement actions with the precision computed as described above. LinkedIn also reports estimates of precision, accuracy, and recall by official languages of the EU in Sheet 8.

LinkedIn is a global platform with over 1 billion members worldwide and is committed to keeping its platform safe, trusted, and professional. In service of that commitment, LinkedIn utilizes both human content moderation and automated systems to moderate content. As relevant to this report, LinkedIn’s automated systems are used for content moderation in two relevant ways. First, LinkedIn uses an automated system to identify and remove policy-violating content. Second, LinkedIn uses an automated system to identify and restrict policy-violating accounts (e.g. fake accounts). These automated systems are based in part on past decisions human reviewers have made regarding whether content and accounts violate LinkedIn’s policies. In setting parameters for its automated systems, LinkedIn considers a number of factors including: - Precision: LinkedIn considers precision of the system and its ability to correctly identify violative content without incorrectly actioning non-violative content, or false positives. Where appropriate, content that does not meet confidence thresholds for automated action is instead queued for human review or cleared. - Recall: LinkedIn considers the recall of the system and its ability to correctly distinguish between violative and non-violative content without missing violative content, or false negatives, particularly where failure to act could result in harm to members or the platform. - Content sensitivity: For higher-risk content, LinkedIn prioritizes recall with lower thresholds to mitigate platform risk. For lower-risk content, LinkedIn prioritizes precision to reduce false positives. - Error correction and feedback loops: LinkedIn considers the availability of error correction and feedback loop mechanisms – such as quality assurance, appeals, ongoing human review, and performance monitoring – and incorporates them where appropriate to help identify errors and continuously improve system performance.

LinkedIn’s approach to content moderation is risk-based and carefully weighs safety against freedom of expression, erring toward freedom of expression whenever possible. LinkedIn employs the following safeguards, among others, to its automated system for content moderation: LinkedIn’s monitors the aggregate performance and accuracy of the automated system, and sets minimum thresholds for performance; LinkedIn sets thresholds for confidence of decisions and tunes the automated system to take action only above those confidence levels; LinkedIn limits the types of violating content the system will act on; LinkedIn generally allows authors to appeal a decision if they believe the decision is incorrect; and LinkedIn retrains its system as appropriate to account for, e.g., changes in human-reviewer decisions, technological innovation, and content trends over time.

As discussed above, LinkedIn is a global platform with over 1 billion members worldwide and is committed to keeping its platform safe, trusted, and professional. In service of that commitment, LinkedIn utilizes both human content moderation and automated systems to moderate content. As relevant to this report, LinkedIn’s automated systems are used for content moderation in two relevant ways. First, LinkedIn uses an automated system to identify and remove policy-violating content. Second, LinkedIn uses an automated system to identify and restrict policy-violating accounts (e.g. fake accounts). As addressed further in LinkedIn’s DSA Systemic Risk Assessment, LinkedIn’s automated moderation systems aide LinkedIn in enforcing its policies and addressing essentially all risk areas – consumer protection and fraud, human dignity, mental and physical wellbeing, discrimination and hate, freedom of expression and information, protection of personal data, civic discourse and electoral processes, public health, illegal content and activities, private and family life, public security, rights and protection of minors. LinkedIn does not used automated means for certain moderation decisions. For example, LinkedIn does not use automated means to resolve Member State orders, Article 16 notices, Article 16 notices from Trusted Flaggers, or certain high-risk reports (e.g. terrorist content).

LinkedIn is a real-identity online service for professionals to connect and interact with other professionals, learn, hire, and find jobs. LinkedIn’s vision is to create economic opportunity for every member of the global workforce. Its mission is to connect the world’s professionals to make them more productive and successful. As part of that mission, LinkedIn is committed to keeping its platform and services safe, trusted, and professional, and to providing transparency to its members, the public, and to regulators. All LinkedIn users are bound by the LinkedIn User Agreement (https://www.linkedin.com/legal/user-agreement). All content on LinkedIn must comply with the LinkedIn Professional Community Policies (https://www.linkedin.com/legal/professional-community-policies), which set out in detail the content LinkedIn permits and does not permit to keep its platform safe, trusted, and professional. In addition to the Professional Community Policies, job posts on LinkedIn must also comply with the LinkedIn Jobs Policies (https://www.linkedin.com/legal/l/jobs-policies), and ads must comply with the LinkedIn Advertising Policies (https://www.linkedin.com/legal/ads-policy). LinkedIn applies a three-layer, multidimensional approach to moderate content on LinkedIn: - The first layer of protection is automated and proactive prevention. When a member attempts to create a piece of content on LinkedIn, various calls (or signals) are sent to LinkedIn’s machine learning services. These services aim to automatically filter out certain policy-violating content at the time of creation. - The second layer of protection is a combination of automated and human-led detection. LinkedIn’s second layer of moderation detects content that is likely to be violative but for which LinkedIn is not sufficiently confident to warrant automatic removal, and sends it for human review. - The third layer of protection is human-led detection. In addition to LinkedIn's Article 16 notice and action mechanism for illegal content, if users locate content that they believe violates LinkedIn’s policies, they are able to report it using LinkedIn’s in-product flagging functionality. As relevant to this report, LinkedIn applied the following enforcement actions to content during the reporting period because it violated LinkedIn’s policies: - LinkedIn removed content because it violated its policies, reported under own initiative measure 'Visibility restriction Removal'. - LinkedIn limited the visibility of content because it violated its policies, reported under own initiative measure 'Visibility restriction Demoted'. - LinkedIn applied sensitive content warnings and limited the visibility of content because it violated its policies, reported under own initiative measure 'Visibility restriction Other'. A sensitive content warning obscures a post until a member clicks to view the post. - LinkedIn applied time-bounded restrictions to accounts’ ability to post content or to post jobs (without restricting the accounts in total) because they violated LinkedIn's policies, reported under own initiative measure 'Provision of the service Suspension'. - LinkedIn applied indefinite restrictions to accounts’ ability to post content or to post jobs (without restricting the accounts in total) because they violated LinkedIn's policies, reported under own initiative measure 'Provision of the service Termination'. - LinkedIn applied time-bounded restrictions to accounts in total, reported under own initiative measure 'Account restriction Suspension'. - LinkedIn applied indefinite restrictions to accounts in total, reported under own initiative measure 'Account restriction Termination'. LinkedIn did not apply own initiative measures under 'Visibility restriction Disable,' 'Visibility restriction Age restricted,' 'Visibility restriction Interaction restricted,' 'Visibility restriction Labelled,' 'Monetary restriction Suspension,' 'Monetary restriction Termination,' or 'Monetary restriction Other' during the reporting period. LinkedIn’s policy regarding the application of sensitive content warnings is provided within the LinkedIn Professional Community Policies, linked above. Specifically: “We may allow certain content that would otherwise violate these Professional Community Policies to remain on the platform if it is educational or newsworthy. Examples may include graphic depictions of an occupation, such as video of a surgeon operating, content shared about a newsworthy event like war or armed conflict, or inherently newsworthy content like statements or actions by certain public officials. When making a determination of newsworthiness, we balance the potential harm from the content staying on the platform against the value to our members and the public of it staying accessible. Graphic or disturbing newsworthy or educational content will include a warning screen.”

LinkedIn Trust and Safety has a dedicated position focused on wellness and has established wellness programs in place globally to provide on demand – linguistically- and culturally- aligned – support to moderators who may need assistance after reviewing abusive or extreme content. With respect to employees, LinkedIn has numerous resources available that support various dimensions of wellness – from top notch medical benefits to a robust Employee Assistance Program (EAP) service, as well as comprehensive fitness and mental health programs on site and virtually. LinkedIn also has vendors in each region to support the mental health of moderators. These vendors provide group education and 1:1 wellness coaching sessions with a trauma-informed lens that is geared towards the most at-risk employees. LinkedIn’s Trust Review Operations team also regularly conducts "pulse checks" and surveys to understand the wellness needs of the team and shift resources as necessary to meet those needs. Separately, content moderation tooling enhancement requests are also collected and implemented on a regular basis to improve the moderator’s experience, both in terms of efficacy and wellness. LinkedIn also takes into consideration the physical environment of content moderators to enhance design of working locations to promote mental wellbeing, tracks turnover and attrition as an additional indicator of wellness, and creates opportunities for team members to offboard to other roles within LinkedIn as needed. With respect to external moderators, LinkedIn requires that the service provider provide a wellness program consistent with industry best practices, including, among other things, regular resilience and wellness training, frequent and periodic check-ins, access to wellness counselors, and time away from work to participate in wellness activities. LinkedIn also requires that team leads be trained on wellness guidelines and that the service provider provide regular team bonding activities. LinkedIn conducts regular check-ins with the service provider’s wellness team to discuss team wellness activities as well as any moderator topics of concern.

LinkedIn has implemented a robust training and Quality Assurance (QA) program designed to help improve reviewer performance and consistency. Regular QA audits are performed on a sample of all content items reviewed by LinkedIn’s moderation teams. Results are shared weekly with more detailed summaries and reports provided monthly. Moderators who underperform may receive one-on-one coaching, common error trends are addressed during group calibration sessions, and policy refresher trainings are delivered as needed. LinkedIn has an intensive onboarding and training program for enforcement personnel, including topical training on key risk areas like elections. LinkedIn's dedicated team of trainers and quality assurance analysts are tasked with onboarding new content moderators, training content moderators on new policies and policy changes, and monitoring and improving moderator accuracy and consistency. Each reviewer goes through 3 months of training across content policies, as well as additional training as necessary to stay up-to-date or resolve areas of confusion. Content moderators also have direct access to content policy managers through regular office hours and dedicated escalation pathways. For particularly complex decisions, content policy managers also have access to in-house lawyers who can consult regional legal experts as needed. Though it can vary based on moderators’ area of work and the number of policy updates, a moderator will typically be required to complete ten to thirty e-learning training modules on policy updates and new policies over the course of a year. Additionally, LinkedIn requires all employees to complete an annual mandatory compliance training that, among other things, is designed to ensure that all employees are trained in the service’s approach to protecting users from harmful content or activities.