The chart number of videos removed by YouTube for the period October 2017-March 2022, by first source of detection (automated flagging or human detection). Flags from human detection can come from a user or a member of YouTube’s Trusted Flagger program,which include individuals, NGOs, and government agencies. The chart shows that the number of automated flagging is significantly higher compared to human detection. When it comes to human detection, the biggest number of removed videos were first noticed by users, followed by individual trusted flaggers, NGOs and government agencies.

This chart shows the distribution of videos removed by YouTube, by the reason removal, over the period September 2018-March 2022. These removal reasons correspond to YouTube’s Community Guidelines. Reviewers evaluate flagged videos against all of YouTube's Community Guidelines and policies, regardless of the reason the video was originally flagged. As the chart shows, the most frequent reasons of removal of videos are child abusive content, violent or graphic content and nudity or sexual content. In the first quarter of 2022, the child safety content decline by 53.5% compared to the same period of 2021, while harmful or dangerous content increased in the same period by 463% and harassement and cyberbullying by 579%.

The chart shows the distribution of the videos removed by Youtube based on human detection, by flagging reason. The data represents average shares of videos removed for the period October 2017-March 2022 and are calculated based on the trimestrial values included in the transparency report. The results show that the users' main flagging reason of videos is the spam, mislinding and scam content, followed by sexual content and hateful or abusive content. When flagging a video, human flaggers can select a reason they are reporting the video and leave comments or video timestamps for YouTube's reviewers. This chart shows the flagging reasons that people selected when reporting YouTube content. A single video may be flagged multiple times and may be flagged for different reasons. Reviewers evaluate flagged videos against all of the Community Guidelines and policies, regardless of why they were originally flagged. Flagging a video does not necessarily result in it being removed. Human flagged videos are removed for violations of Community Guidelines once a trained reviewer confirms a policy violation.

High Frequency Indicator: This dataset presents year and month wise enforcement actions taken by Significant Social Media Intermediaries (SSMIs) from 2021 to the present, compiled from the mandatory monthly transparency reports published under Rule 4(1)(d) of the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021. It includes counts of content removed, accounts suspended or banned, and chatrooms, comments, edit profiles and livestreams restricted, along with the policy or violation category (e.g., child sexual exploitation, terrorism, hate speech, bullying, violence, regulated goods, misinformation, etc.).

To enable comparability across platforms with different reporting terms, the dataset uses a standardised enforcement classification:

1. enforcement_type:

The type of action taken: a. Content Actioned (any enforcement such as warning, downranking, age-gating), b. Content Removed (content deleted or made inaccessible), c. Account Banned (account suspension or disabling), d. Quality Metric (AI moderation accuracy indicators reported by some platforms).

1. proactive_flag:

Whether the platform identified and enforced before user reports: a. Proactive = Found via automated detection or internal review systems, b. Unknown = Platform did not specify proactive vs reactive.

Notes: 1. SSMI denotes to Significant Social Media Intermediaries, with over 50,00,000 registered users in India, which primarily or solely enables online interaction between two or more users and allows them to create, upload, share, disseminate, modify or access information using its services

1. Facebook, Instagram, and Threads (Meta) a. Content Actioned counts any enforcement, not only removals (e.g., removals, warning screens/covering, age gates, downranking). b. Proactive Rate = (items found & actioned proactively) ÷ (total content actioned).

2. X/Twitter a. Child Sexual Exploitation and terrorism suspensions are largely proactive, flagged using proprietary tools and industry hash-sharing systems. b. Data reflects global enforcement, not only India.

3. Google / YouTube a. Number of removal actions as a result of automated detection captures actions triggered by automated systems (ML + human-trained models).

4. ShareChat a. Content Removed / Taken Down / UGC discard / Comments/Chatrooms deleted are standardised as Content Removed. b. Also includes rights-holder reporting workflow for copyright/IP and automated proactive monitoring for harmful content.

5. WhatsApp a. Reports Proactively Banned Accounts, meaning accounts banned before any user reports.

6. Koo a. Distinguishes between Content Removed, Content Actioned (flagged/downranked), and Account Banned. b. Automation Correct/Wrong reflect AI moderation accuracy, not enforcement outcomes.

The chart shows the distribution of human flags on YouTube for the period October 2017 - March 2022, by type of flagger. Human flags can come from a user or a member of YouTube’s Trusted Flagger program,which include individuals, NGOs and government agencies. The chart shows that the majority of human flags come from users, followed by individual trusted flaggers. The share of flags from NGOs is insignificant compared to the other two type of flaggers.