RTFP

Snap has established a cross-functional platform governance body with representation from Trust & Safety, Content Review, Product, Safety Legal, Policy and Compliance teams. The objective of the platform governance team is to manage, oversee, monitor, assess and adjust Snap's DSA compliance programme. The platform governance team meets on a regular basis to 1) assess and ensure ongoing and consistent compliance with DSA obligations through internal reviews, assessments; and 2) monitor relevant content moderation metrics to ensure ongoing compliance and to early detect any anomalies which would then be addressed.

Across Snapchat, we set clear Community Guidelines for what is not permitted on our platform to maintain a positive experience for all Snapchatters. These guidelines apply to all content and all Snapchatters. We also set clear Content Guidelines for Recommendation Eligibility to outline the additional and stricter standards that must be met in order for Public Content to be eligible for algorithmic recommendation. We use a combination of in-app reporting, automation tools (such as abusive language detection, image recognition models, and account history), and human review to detect and moderate harms on the platform. We assess content reactively (upon receipt of a report) and proactively (when flagged by our automated detection tools) using a combination of automated review tools and/or human review. Proactive detection mechanisms trigger a review, at which point, our tooling systems process the request, gather relevant metadata, and may route the relevant task to the moderation teams via a structured user interface that is designed to facilitate effective and efficient review operations. Our reactive detection processes complement our proactive detection efforts. We provide in-app and web-based reporting tools that enable EU users to report illegal content pursuant to our Article 16 notice mechanism, or to separately report content and accounts they think violate our policies, including our Community Guidelines. We also have mechanisms enabling non-users in the EU to report content on Snapchat they believe is illegal or a violation of our policies. We apply a combination of human and automated tools to enforce policies against violative content. Enforcement actions vary depending on the severity of the violation and can range from removing recommendation eligibility (least severe) to removing the account (most severe). We work to ensure these rules are applied consistently. When applying them, we take into account the nature of the content, including whether it is newsworthy, factual, and relates to a matter of political, social, or educational value. During the reporting period, we saw a Violative View Rate (VVR) of 0.01% percent globally, which means that out of every 10,000 Snap and Story views on Snapchat, 1 contained content found to violate our Community Guidelines. Among enforcements for what we consider to be “Severe Harms,” we saw a VVR of 0.0003% percent.

In order to compute the number of human resources dedicated to content moderation, we identified teams of employees and contractors who regularly evaluate content on Snapchat. The number reported herein is the number of employees and contractors who were on those teams as of 31 December 2025.

The teams working on content moderation are composed of full time employees (FTEs) and contractors, who are employed with third party partners. The composition of the teams change as we see incoming volume trends or submissions by language/country. We review our metrics weekly to see if we need additional support for particular markets to meet our service level commitments. Our content moderation teams operate across the globe, enabling us to help keep Snapchatters safe 24/7. In situations where we need additional language support, we use translation services. Moderators are recruited using a standard job description that includes a language requirement (depending on the need). The language requirement states that the candidate should be able to demonstrate written and spoken fluency in the language. Candidates moderating in European languages need to be certified at a minimum C1 level for supported languages. For entry-level positions, candidates are required to have a high school diploma and at least one year of work experience (though, the majority of our moderators hold a BA+). Candidates must meet the educational and background requirements in order to be considered. Candidates also must demonstrate an understanding of current events for the country or region of content moderation they will support. We employed 1,430 unique moderators proficient in European languages. The combined sum across individual European language categories reached 1,939 due to the multilingual proficiency of our team members. In situations where we do not employ moderators sufficiently proficient in a particular language, as specified in the quantitative template, and need additional language support, we use translation services.

We monitor the accuracy and error rates of our automated enforcement tools by selecting random samples of tasks for which automation made a moderation decision and submitting them for re-review by our human moderation teams. We monitor the accuracy and error rates of our automated enforcement tools by selecting random samples of tasks for which automation made a moderation decision and submitting them for re-review by our human moderation teams. The accuracy rate is the percentage of sampled items for which the automated decision matches the decision reached on human re-review. The precision rate is the percentage of sampled items that were flagged by automation as violative and were confirmed as violative on human re-review (true positives). For the purposes of this report, in the context of automation, we use the term ‘recall’ to refer to the automation rate, defined as the percentage of all moderation outcomes in a given category, including enforcement and non-enforcement outcomes, that are determined by automated tools. Snap evaluates the auto-moderation tools in place to assess the precision and make adjustments as needed to maintain a high precision. We also use human moderators to review the results of the automoderation models, to check its efficacy (using precision and recall metrics as the ground truth). These results are incorporated into the model to retrain in areas where there might have been identified discrepancies. While our primary reporting is aggregate, we also monitor these accuracy metrics across different control groups, specifically distinguishing between different violation categories to ensure the automated tools do not disproportionately underperform on specific types of content.

As explained above in rows 2 and 3, we deploy automated tools to proactively detect and, in some cases, enforce violations of our Terms and policies at our own initiative. Our automated detection tools include hash-matching tools (including PhotoDNA and Google CSAI Match), Google Content Safety API, Abusive Language Detection models (which detect and reject content based on an identified and regularly updated list of abusive keywords and emojis), other custom technology designed to detect abusive text and emojis, and technologies leveraging artificial intelligence, including machine learning and large language models. We continuously update these tools, set standards, and test these automated systems for accuracy. When our automated tools detect a potential violation of our policies, the potentially violating material is either sent for human review or automatically actioned in accordance with our policies. (Notices submitted through our Article 16 and Trusted Flagger Notice and Action Mechanisms undergo human review).

We are mindful of the potential impact of automated moderation tools on fundamental rights, and we deploy safeguards to help minimise that impact. Our automated content moderation tools are tested prior to being deployed. Models are tested offline for performance to ensure their proper functioning prior to being fully phased into production. We perform pre-launch Quality Assurance (QA) reviews, launch reviews, and ongoing precision QA checks during rollouts. Following the launch of our automated tools, we evaluate their performance and accuracy on an ongoing basis, and make adjustments as needed. We actively monitor the precision of our detection tools and promptly remediate rules that drop below our quality thresholds. Our automated moderation tools similarly undergo ongoing monitoring for accuracy. This process involves the re-review of samples of automated tasks by our human moderators to identify models that require adjustments to improve accuracy. We also monitor the prevalence of specific harms on Snapchat via random daily sampling of certain Public Content, and leverage this information to identify areas for further improvement. Our policies and systems promote consistent and fair enforcement, including by our automated tools, and provide Snapchatters an opportunity to meaningfully dispute enforcement outcomes through notice and appeals processes that aim to safeguard the interests of our community while protecting individual Snapchatter rights. We strive to continually evolve our automated content moderation tools to improve their accuracy and support the consistent and fair enforcement of our policies. We use human review for, among other things, notices submitted through our Article 16 notice and action mechanism (including those from Trusted Flaggers) and all appeals of account locks on the basis of violations of Community Guidelines (including illegal content).

Snap applies automated means throughout various stages of the content moderation process, including where appropriate to proactively detect violative content and/or to make enforcements (e.g. restrict visibility, delete content, etc.). The type of automated means deployed differs based on the public or private nature of the content surface, and the type of harm at issue. Across both private and public surfaces, Snap uses automated hash-matching technologies and classifiers to detect severe harms and known illegal content. The purpose of these tools is to identify and remove severe harms such as Child Sexual Exploitation and Abuse (CSEA). Snap deploys a range of automated content moderation mechanisms to scan media uploads for CSEA specifically, including: (i) PhotoDNA, CSAI Match and other hash-based detection (including NCMEC’s Take It Down hashes) to detect known CSAM; (ii) Google’s Content Safety API to detect novel CSAM and (iii) proprietary signal based tools to detect sextortion and other sexual harms against minors. Snap deploys additional automated content moderation tools to scan public surfaces (e.g. public Stories, Discover, and Spotlight submissions). Tools include abusive language detection, other keyword-based detection, and machine-learning-based proactive detection models. Snapchat uses these automated means to detect violative content including the sale of prohibited and regulated goods and services, self harm and suicide, sexual content, CSEA, terrorist content, violent and dangerous behavior, harassment and bullying, and fraud and spam. For public-facing content surfaces (such as Spotlight, Discover, and Public Stories), a principal purpose of automation is to address the risk of viral amplification of harmful content. Content that is approved for recommendation to broader audiences must comply with both our Community Guidelines and our Content Guidelines for Recommendation Eligibility. All Spotlight and non-professional user generated Discover content goes through automoderation against these guidelines before it is eligible for recommendation to a wide audience. As part of this process, we monitor content that is achieving large-scale reach so that it can be human reviewed and interrupt potential virality, and a means of checking that our pre-moderation systems are working effectively.

In this report, Snap has refined its reporting methodology to align with the DSA transparency template. Previously, Snap aggregated all EU user reports regarding Community Guidelines violations into our Article 16 figures because our Community Guidelines substantially prohibit many categories of content and activities that are illegal in the EU. To enhance data precision, Article 16 data now exclusively encompasses illegal content notices received via Snap’s dedicated Article 16 notice and action mechanism forms (originating from in-app and Support Site reports), including from Trusted Flaggers. Enforcements resulting from user reports of Community Guidelines violations have been reclassified and are now reported under the “own initiative” field, as described below. Content moderation at Snap’s “own initiative” refers to enforcement actions taken on accounts in the EU that violate our Community Guidelines or Content Guidelines for Recommendation Eligibility, distinct from actions in response to an Article 9 order or Article 16 notice. This encompasses both proactive enforcement—where our safety teams and automated detection tools identify potentially violative content—and reactive enforcement, involving automated tools and human review, triggered by reports from Snapchatters or other individuals or entities outside of the Article 16 process. The reported enforcement figures on user accounts in the EU do not include actions taken against advertisers under our separate Advertising Policies. We use automated tools that detect and/or enforce against violations of our Community Guidelines, including enforcements against a user’s account, as well as automated tools that detect and/or take action on content on our public broadcast surfaces that does not align with our Content Guidelines for Recommendation Eligibility. Irrespective of how we may become aware of violating content, our safety teams, through a combination of automation and human moderation, promptly review identified content and make enforcement decisions. Enforcements may include removing the content, warning or disabling the violating account, removing a violative user name or display name, and blocking the violating account’s device. We also may report the account to law enforcement, if warranted, alongside one of these other enforcements. For some creators and some types of content, enforcements may also lead to visibility restrictions on, or demonetisation of, their content. If a user accrues too many Community Guidelines enforcements over a defined time period, their account may be automatically locked. In prior reports covering 2024 and the first half of 2025, Snap counted all strikes and deleted content as a part of its warning enforcement figures and content deletion figures. However, due to a discrepancy in data mapping, the ensuing automated locks were not reported as a separate, unique account lock measure. These automated locks amounted to 37,411 locks in the first half of 2024, 55,781 locks in the second half of 2024, and 53,856 locks in the first half of 2025. In the second half of 2025, Snap applied 80,173 automated locks based on strikes to EU accounts. Consistent with the new DSA template’s requirements to separately report each content restriction or account enforcement measure by type, Snap includes these strike-based automated locks as part of its own-initiative account-termination restriction measures. In this updated version 2.0 of the report for the period 1 July 2025 - 31 December 2025, Snap has addressed three formula-based errors and inconsistencies between parent-level categories and their underlying subcategory data. We have also made formatting and labelling updates and added an explanatory note to improve overall clarity. We corrected a formula in Sheet 6_own_initiative_TC that had double-counted 8,507,506 automated “own-initiative” enforcement measures taken under our Content Guidelines for Recommendation Eligibility. These duplicate measures have been removed from the overall own initiative totals, the totals in STATEMENT_CATEGORY_OTHER_VIOLATION_TC, and the totals of underlying categories “Editorial Quality,” “Spam-Adjacent,” “Sexually Suggestive,” “Depictions of Regulated Goods or Activities,” “Disturbing,” and “Other Violations of our Content Guidelines for Recommendation Eligibility.” We also refreshed the underlying dataset, which resulted in minor adjustments (below 0.6%) in the subcategories of this section. Second, we resolved a discrepancy in the subcategory breakdown of member state requests for information for Sweden and Slovenia in Sheet 3_member_states_orders, with no impact on overall country-level totals. Finally, we fixed the calculation of the median turnaround times of parent-level categories in Sheet 4_notices, without any changes to the accurately-reported subcategory median times.

We consider all team members who are on the frontlines of reviewing and moderating content on Snapchat to be digital first responders -- and believe they need to be supported in the same ways that first responders in other fields are. We prioritise their wellbeing first – and regularly reinforce that philosophy with our teams and leadership. We are constantly working to keep strengthening our approach to supporting our digital first responders, which includes: Properly disclosing the risk of what content a content moderator might be exposed to during the interview process and prior to a job offer; Providing comprehensive on the job wellness support and easy access to mental health services, including on-site counselors; Creating career paths for digital first responders to both grow within that path or transition into other roles. This includes allowing content moderators to opt out of a workflow if they no longer feel comfortable with the content; Wellness provisions are included in our contracts with third party partners, including the requirement of on-site counselors and space considerations; and Regular wellness reporting as part of our vendor contracts and audits by our FTE team.

Our content moderation teams apply our content moderation policies to help protect our Snapchat community. They are trained over a multi-week period, in which new team members are educated on Snap’s policies, tools and escalation procedures. Our moderation teams engage in refresher training relevant to their workflows, particularly when we encounter policy-borderline and context-dependent cases. In addition, each time a significant update is made to our policies, dedicated training sessions are delivered to ensure moderators understand and consistently apply the revised standards. We also run upskilling programmes and certification sessions, and administer quizzes, to ensure all moderators are current and in compliance with all updated policies. We conduct weekly calibration sessions with moderation vendors, maintain dedicated communication channels for real-time policy and workflow questions, and provide targeted supplemental training where needed. Finally, when urgent content trends surface based on current events, we quickly disseminate policy clarifications to ensure teams respond according to Snap's policies.