RTFP

Assessing reports can be very complex because it requires both an understanding of context and specific content moderation expertise. As such, we’ve structured our teams and processes to help ensure that we are making accurate decisions when it comes to reviewing reported content. Content moderation at Pinterest is overseen by a dedicated Trust and Safety team, which reports to senior management. The team includes specialists trained in Pinterest’s Guidelines and specialists trained to review content against laws of the EU and its Member States. In the event of complex reports, team members can escalate any questions to leads of our Trust & Safety Enforcement Operations team, who may in turn consult with specialists within and outside the company trained in laws of the EU and its Member States.

For the purposes of this report, “own initiative” content moderation refers to decisions to impose restrictions on an account or certain content that have not been preceded by an order from an EU Member State authority, or a notice submitted through a notice and action mechanism under Article 16 DSA. This includes actions taken due to policy violations, whether following user reports or through proactive detection. Pinterest engages in content moderation at its own initiative to detect and address content that violates our policies. However, we do not engage in content moderation at our own initiative specifically for the purpose of identifying illegal content. Any actions we take related to illegal content are conducted either in accordance with our policies or in response to an order from an EU Member State authority or an Article 16 notice, and we do not proactively search for illegal content outside of these contexts. Reported content Anyone can report content they think violates our policies and, if we confirm a violation of our policies, we take appropriate action. Proactively detected content In addition, content that may violate our guidelines is also found via proactive methods. The mechanisms used to detect different potential violations vary based on the state of available technology, the volume of violative content and other factors such as the complexity of evaluation. We continue to iterate and evolve our tools and expect ongoing advancements as our teams continue to strive to maintain a positive and safe environment. How we action content When we become aware of a violation of any of our policies, we enforce our policies through automated tools, manual review and hybrid approaches that combine elements of both. These systems may use logic-based rules as well as machine learning. Where appropriate, we may take into account information provided by trusted third parties and industry tools. The Pins reported in tab 6_own_initiative_TC received 9,866,494,958 impressions globally during the reporting period. These impressions include impressions both before and after any content moderation action took place.

The total number of human resources dedicated to content moderation is calculated based on staffing levels as of the end of the reporting period (31 December, 2025). This count includes specialists from our internal operations teams, team members who consult on complex cases, and external moderation agents.

We hire individuals with diverse professional experience and educational backgrounds, with all members of the team having at least a high school diploma. Our review team is further assisted by members of our Legal department, who are qualified lawyers with in-depth experience regarding content moderation under our Community Guidelines and applicable EU and Member State laws. Pinterest is primarily a visual platform, nevertheless every member of our review team speaks at least one official language of a Member State of the EU, and we have additional external resources available to assist with any EU Member State official languages not covered by our internal specialists. Our internal and external specialists have a level of linguistic expertise at least equivalent to level B2 under the Common European Framework of Reference for Languages (CEFR).

We are committed to evaluating and improving the effectiveness of our automated content moderation measures, and we use a variety of metrics to measure their performance. These metrics include indicators such as accuracy, precision, and recall. Precision reflects how often our automated enforcements are correct, while recall measures the proportion of policy violations successfully identified by automation. Accuracy provides an overview of how many overall decisions - both enforcements and non-enforcements - are correct. To calculate accuracy, precision, and recall for our automated content moderation, we adhere to the classical definitions of these metrics as closely as possible. We calculate each indicator as follows: - Precision = True Positives / (True Positives + False Positives) - Accuracy = (True Positives + True Negatives ) / (True positives + True Negatives + False Positives + False Negatives) - Recall = True Positives / (True Positives + False Negatives) True positives are defined as automated enforcement actions that were never reversed. False positives refer to automated enforcement actions that were reversed, for example, actions reversed on appeal. False negatives are enforcement actions that are not reversed but were not made by automated systems, indicating they were missed by automation. True negatives are cases where policy enforcement review was performed by a human but no violation was found, representing cases where automated systems correctly did not apply any enforcement action.

We use automated tools and hybrid approaches (that combine elements of manual and automated actions) to enforce our policies. These systems may use logic-based rules as well as machine learning. Where appropriate, we also consider information provided by trusted third parties and industry tools. Automated actions: Our automated tools use a combination of signals to identify and take action against content identified as potentially violating our policies. For example, our machine learning models assign scores to content added to our platform. Our automated tools can then use those scores to perform appropriate enforcement actions. Hybrid actions: Sometimes, actions combine elements of manual and automated enforcement. Hybrid actions include those where a team member determines that a Pin violates our policy, and automated systems help expand that decision to enforce against machine-identified matching Pins. Depending on the volume of matching Pins, a hybrid action may result in a number of additional Pins actioned or none at all. For the purposes of this report, hybrid actions have been classified as automated actions and have been reported on accordingly. Parameters for automated means Our automated content moderation systems utilize a range of parameters designed to maximize accuracy while minimizing the potential for false positives. These include: Model scores: Machine learning models assign scores to content based on their likelihood of violating our policies. Thresholds are carefully calibrated to ensure that only content with a high probability of violation is flagged for further review or action. Image-matching strategies: For images, we employ per-policy strategies to determine how broadly content is matched across our platform. This includes settings for "exact match" (identifying identical content) and "near duplicate" (identifying visually similar content) fanouts. These strategies are tailored for each policy area to reflect varying risk profiles. Optical character recognition (OCR): OCR technology is used to extract text from images, which further informs our image-matching approach. Detected text can influence whether an image is compared using exact match or near duplicate methods, ensuring that textual violations embedded within images are identified while limiting unnecessary fanout. All of these parameters are continually refined and optimized to avoid excessive false positives and to ensure our moderation processes remain fair, effective, and proportional to the risks involved.

We use the latest modelling techniques and continuously iterate on these models. By adding new data and exploring new technical breakthroughs, we maintain or improve their performance over time to effectively address violative content. To control the quality of a machine learning model, we evaluate the performance of that model through offline analysis and online experiments before launching. We look at a number of metrics to review the effectiveness of our models, both prior to and subsequent to launch of the model. We regularly review data and other relevant information to continually identify areas for improvement that our teams investigate and address as appropriate.

We use machine learning models to help enforce our policies across Pinterest. These models assist in identifying content that may require action under our policies. In addition to these detection capabilities, we also use automation to efficiently remove identical or near-identical content that has been previously flagged by human review. This approach allows us to enforce our guidelines across all Pins and re-Pins, helping to ensure that violating content is removed from our platform.

Pinterest’s Terms of Service require users to abide by various guidelines and policies regarding content on Pinterest. These include our Community Guidelines and related policies, which explain what is and isn’t allowed on Pinterest, as well as our Advertising Guidelines and Merchant Guidelines, which explain additional policies that apply to Advertisers and Merchants on our platform. Our Enforcement Guidelines detail how we put our policies into practice, including any restrictions that we may apply to content. Together, we refer to these as “Guidelines” or our “policies,” and we explain each in more detail below. Community Guidelines Pinterest's mission is to bring everyone the inspiration to create a life they love. That being said, not all content is inspiring - so we have Community Guidelines to outline what we do and don’t allow on Pinterest and all users must abide by them. In most cases, our Community Guidelines already cover material that is illegal in the EU as well. Our goal is to be a positive and inspiring place on the internet, so our Community Guidelines may also go further and prohibit certain categories of content that, while legal, are not consistent with the positive nature or purpose of our platform. Merchant and Advertising Guidelines We have additional rules for merchants and advertisers to set clear expectations about what is and is not acceptable for product Pins and advertisements, called (respectively) our Merchant Guidelines and Advertising Guidelines. These Guidelines include high standards that are for the safety of all audiences who use Pinterest. Our Terms of Service also reference our Copyright and Trademark Policies. Pinterest respects the intellectual property rights of others and we expect people on our platform to do the same. It’s our policy to disable content in response to complete and valid claims of infringement and, in appropriate circumstances, to deactivate the accounts of people who repeatedly infringe or are repeatedly charged with infringing copyrights or other intellectual property rights. We work hard to enforce these policies to create a positive and safe environment for users. When we become aware of a violation of any of our Guidelines, we take appropriate action to keep our users and the community safe. That action may consist of: - Deactivation: When we deactivate a Pin, that Pin is no longer available to anyone on the platform. Deactivation may also be referred to as “removal.” - Limiting distribution: When we limit the distribution of a Pin, that Pin will continue to be accessible to the user who posted it, but it won’t be distributed through recommendation or discovery surfaces such as Pinterest search results or home feed. When we become aware of a violation of our Advertising Guidelines, we’ll take action: - Rejection: When an advertisement is rejected, it will not be shown to users. We may reject an ad before ever showing it to anyone, or after the campaign has already begun to run - for example, in response to a report from a user who saw the ad. - Limited approval: Limited approval means that an ad may run, but with restrictions - for example, it may only be shown to people in certain regions or to people over a certain age. - Advertiser removal: When we remove an advertiser, they no longer have access to Pinterest’s advertising tools. This doesn’t affect their ability to use other, non-advertising Pinterest products. When we become aware of a violation of our Merchant Guidelines, we’ll take action: - Merchant rejection: If a merchant is rejected at any time, they are denied access to Pinterest’s merchant products. This doesn’t affect their ability to use other, standard features on Pinterest. We may also block links that lead to content or are associated with content that may violate our policies. If someone on Pinterest has verified that they own a domain that we block, we provide that person with notice and a chance to appeal.

Our systems and processes are set up to provide ongoing support to our specialists. In order to help ensure that we consistently make accurate decisions on complex issues, specialists can always ask questions to their team leads or escalate more complex requests to other relevant experts, such as the Policy or Legal team, as appropriate. We send out timely updates regarding current events or changes in our policies so that team members have up-to-date information on continuously evolving historical and cultural context. In addition to specialised training, we also invest heavily in wellness. Pinterest provides mental health and well-being resources to employees moderating content. Additional ongoing education, virtual wellness classes, and activities that promote mental and physical well-being are offered on a regular basis. We also require our external content moderation partners to provide wellness programs. The programs include wellbeing workshops, resilience training and access to counselling.

Pinterest provides structured, continuous training to all personnel involved in content moderation. Our training program is designed and maintained by a dedicated Learning team that develops onboarding curricula, policy-update trainings, and knowledge management systems for Trust & Safety workflows. We are continuously training our review team members in an effort to build the expertise and knowledge critical to content moderation. As part of our onboarding process, each team member undergoes workflow-specific training on Pinterest’s Terms of Service and our Guidelines. The goal of this training is to learn how to review content against these requirements using existing processes and tooling. In addition to the onboarding training, we also provide continuous training to content reviewers for policy and process updates. Depending on the scope of the change, this may involve notifications and bulletins, updated presentations and other learning resources. Major updates and new policy launches require mandatory completion of defined training tasks. We continue to provide team members with regular education, training, and/or certifications as appropriate.