Why we made WhatsApp Watch

From its humble origins as an economical way to send text messages, WhatsApp has grown in size and features to become a behemoth with over two billion users in 180 countries. As its popularity has grown, the platform has also become a common way of sharing news and media for people across the world, especially in global majority countries. 

Researchers and journalists have documented the significant role the platform has played in enabling the rapid transmission of false information and hate speech during periods of national elections. In 2018 this led to devestating consequences in India, when viral misinformation about child kidnappings disseminated on the platform, leading to mob violence and lynching. In response, WhatsApp put limits on the amount of times users can forward a given message. However, the decentralized and private nature of the platform make it hard to study the efficacy of these measures. 

With more than half of the world’s population living in countries experiencing national elections, 2024 has been called the year of democracy. This includes countries like India, where WhatsApp is commonly used for political campaigning, and has more than half a billion monthly active users. In this context, studying how information spreads on WhatsApp is crucial.

This is why we built WhatsApp Watch, a platform that persistently monitors public WhatsApp groups. Computational social scientists have done pioneering work developing techniques to study the platform in a responsible manner. WhatsApp Watch builds on these techniques and makes them available to a wider community of journalists and researchers interested in studying the platform. 

‍

How we collect data

We take a multi-pronged approach to data collection.

• We also collect data from relevant, publicly accessible groups discovered via social media and search engines.
• Our opt-in data donation tool allows researchers, journalists, and individuals to donate data from public groups they are a part of using their mobile device.

How we define public groups 

WhatsApp makes no distinction between public and private groups. While the platform promotes itself as a way to stay in touch with friends and family but in practice it has a much broader set of common use cases. 

Some of the types of groups that fall into the public category include: 



• Groups operated by political parties, government agencies, or public institutions in a given region.
• Groups created to share public information about a city, village, or other locality.

• Groups created to share content around a shared topic of interest such as religion, sports, or inspirational messages.


• Groups that are advertised widely and can be joined simply by clicking on an invitation URL.

We prioritize privacy and safety.

Applying our experience working with investigative and open source intelligence (OSINT) journalists around the globe, we built our platform implementing state-of-the-art security protection.

• We automatically redact personally identifiable information (PII) such as phone numbers and display names of group members.
• We only collect data from public groups where there is no reasonable expectation of privacy.

Our opt-in data donation tool

Individual WhatsApp users can use our custom tool to donate data from groups they belong to that meet our public group definition.

We built consent into the onboarding process at multiple steps, so we can ensure the data donor is aware of their choices and it is easy for them to update their preferences or opt-out of the platform at any time.  Please contact us if you would like to learn more about data donation.

How We Study The Data

We begin analysis of WhatsApp groups starting from the moment data from a group is donated. Our system automatically archives all message media (video, pdfs, images) and metadata, such as forwarding counts and post timestamps. In order to study what's being shared on the platform, we have built a suite of tools to enable various types of analysis.

 Some of these tools include:

• Real time image hashing to detect identical and similar images shared across groups.
• Automatic language translation.
• Web-based tools to run complex queries on the data, including semantic search, similar image detection, and network graph analysis.
• An interface that allows for viewing messages in the context of the group thread, allowing researchers and data annotators to get a better understanding of a message’s meaning based on context.