Previous work by Jakesh, Garimella et al. examined how hybrid cross-platform organization of political volunteers produced hundreds of nationwide Twitter trends during the 2019 Indian national election. Our work also focuses on Trend Alerts and how they coincide with trends on X (formerly Twitter). In addition, we examine Facebook posts that coincide with the time window of specific Trend Alerts to assess their reach on that platform. We demonstrate that Trend Alerts are not limited to the ruling Bharatiya Janata Party (BJP), whose high level of online organization is well documented, but are used by other national and even regional Indian political parties.
WhatsApp Trend Alert Tracker provides a model for tracking cross-platform influence campaigns, which relies on platforms like X and Facebook making their APIs accessible to researchers. At the time we started this project, access to the X/Twitter API was already cost prohibitive. After this project’s kickoff, Meta announced they would be ending access to their Crowdtangle API. Without access to these tools, cross-platform analysis of manipulation campaigns is not possible.
By providing a visual tool for exploring a range of WhatsApp-based political activity, we aim to raise awareness of how the platform is used by political actors and aid better scrutiny of inorganic and deceptive political campaigns in India and other regions of the world.
Anatomy of a Trend Alert WhatsApp Message Trend Alerts for this study refer to WhatsApp messages that contain the keywords “trend alert” and provide instructions for what researchers have called “participatory social media manipulation campaigns.” Trend Alert messages are shared through WhatsApp groups and provide instructions and links to pre-written posts, images, and videos. These messages ask volunteers to share these assets, along with hashtags, on social media channels within a specific time window, with the goal of making a message go viral. Unlike self-organizing advocacy groups, the messaging in a Trend Alert campaign does not emerge organically, but is created by political parties and their affiliates for campaign purposes. Unlike astroturfing where automation and paid participants post messages to create artificial campaigns, previous research has shown that participants who participate in Trend Alert campaigns do so willingly and are not paid for their posts.
Annotated example of a Trend Alert message shared in political WhatsApp group A Trend Alert message typically contains a hashtag, description of the associated campaign, and a specific time window for posting. Most of the Trend Alert messages we found also contain a link to a Google Drive folder with visual assets and Tweet banks for members of the group to share or personalize. While campaign operatives create a narrative, volunteers spread the message.
Screen capture of social media assets shared in Google Drive link for Trend Alert campaign #HBDBabu Background While our focus is on Trend Alert messages on WhatsApp, Alt-News first exposed similar tactics on Twitter in 2017. To track party-driven campaigns on the platform, Alt-News co-founder Prateek Sinha set up an alert in Tweetdeck for the terms “trend alert” and “docs.google.com,” as he had observed instances of such Tweets linking to instructions in both English and Hindi for how to make a hashtag trend at a particular time. X has, and had specific guidelines prohibiting using their services in “a manner intended to artificially amplify or suppress information.” However, Trend Alert campaigns persist. Previous research has shown no evidence X monitors or removes posts involved in Trend Alert campaigns or takes any actions to discourage or preempt them.
Researchers first documented WhatsApp Trend Alert campaigns led by the BJP during the 2019 Indian national elections. The researchers discovered these messages in WhatsApp groups whose invitation URLs were indexed by search engines or advertised on social media by political parties and their supporters. Those researchers filtered for messages that referred to a Twitter hashtag and linked to a Google document—similar elements used in 2017’s Twitter Trend Alert campaigns. Of the 75 Trend Alert campaigns organized by the BJP that they discovered, 69 successfully reached Twitter’s trending topics section in India.
When campaign coordination takes place on WhatsApp, a messaging service, Trend Alerts develop out of the public eye, yet are intended to sway public opinion during critical democratic processes. Our WhatsApp Trend Alert Tracker sheds light on the origins and processes behind these coordinated political campaigns, providing needed transparency to potentially critical political tactics. The tool tracks 29 campaigns that successfully trended on X between January and May, 2024. Some campaigns coincided with landslide election wins that defeated the incumbent party.
Data We collected messages from a selection of public WhatsApp groups throughout India, based on access: via publicly-advertised links to join the WhatsApp group, data donations by members of the group who determined that the conversations were public in nature, and through campaign volunteers who chose to anonymously share groups they administer. In total, we monitor around 2,000 groups.
As of June 6, 2024, searching for the phrase “trend alert,” we found 137 Trend Alert messages promoting campaigns in 2024. Those messages were posted in 45 WhatsApp groups, which had between 17 and 1,920 members. (The median group had 315 members.) The WhatsApp groups in our sample are predominantly from the national Bharatiya Janata Party (BJP), the Telugu Desam Party (TDP), Yuvajana Sramika Rythu Congress Party (YSRCP), Janasena Party (JSP), regional parties based in the state of Andhra Pradesh, and the All India Trinamool Congress (TMC), a state party based in West Bengal.
Of the 45 groups, 43 either literally mention a political party or politician in their group name or express support for a political party. The two other groups do not explicitly mention a political party but their political affiliations are easy to determine based on the messages being shared.
WhatsApp makes no official distinction between public and private groups. We designed our own criteria to maximize privacy considerations. We consider groups to be “private” when they are small in size, where group members have personal relationships, and there is an implicit expectation of privacy. WhatsApp groups of friends, relatives, family members, or other personal communities, such as school alumni or former colleagues fall into this category. We do not collect data from private groups.
We consider WhatsApp groups to be public if they contain at least six members and broadcast messages to a community or discuss public events. For extra privacy, we redact members’ phone numbers and names attached to all messages shared with us.
How we define a hashtag campaign A Trend Alert message sent through a WhatsApp group is usually one of many repeats in multiple groups in support of a particular campaign. In defining a campaign, we relied on specific elements of the WhatsApp Trend Alert message. We manually organized these elements in addition to two category labels into a table that was used for the WhatsApp Trend Alert Tracker:
Labeling campaigns based on theme We label each campaign’s theme as belonging to one of three categories, depending on language in the WhatsApp messages and when applicable, in the content in any linked Google Drive folder. These categories are:
Attack campaign: the trend alert campaign targets an opponent. ex: “Let's join forces in this trend to uncover YSRCP's connection to the drug syndicate! #JaganForDrugs” General campaigning: a candidate is specifically promoted or celebrated.ex: “ #BabunuMalliRappiddham Join the Twitter hashtag trend movement to bring back the visionary leader Chandrababu Naidu, who is committed to the welfare and development of the state.” Temporal events: a trend alert that promotes a rally or political campaign event. ex: “Everyone Should Ready For The Positive Buzz, Our PM Sri.Narendra Modi ji's Visit to Tamilnadu on 27th & 28th Feb the tag to trend on X is *#VendumMeendumModi*” Multiple messages for one campaign In total we identified 40 distinct campaigns that were shared in 137 Trend Alert WhatsApp messages across 45 WhatsApp groups. Many of the campaigns were shared across multiple WhatsApp groups. In one instance, a campaign was shared via 23 distinct messages in 17 different WhatsApp groups. There are slight variations in the text of some of the messages, but the general instructions—such as the start and end times for primary campaign hashtags (either a singular hashtag or the first hashtag in a list of hashtags) —were consistent, so we treated these messages as a single campaign.
Searching for Trend Alerts on Social Media Once we manually identified the primary hashtag for a campaign, secondary hashtag, and campaign start/end times, we searched for the hashtags in a public archive of trending topics on X/Twitter and for Facebook posts using Meta’s CrowdTangle tool.
We used data from these third-party tools from the day before the campaign’s intended start time through the day after its intended end time. For example, if the campaign’s intended timeframe was March 20, 2024 at 10am to 12pm IST (4:30am–6:30am UTC), we collected X/Twitter trending topics and Facebook posts from March 19, 2024 UTC through March 21, 2024 UTC.
Screen capture of Google doc shared in a Trend Alert message with example Tweets for #JaganLandGrabbingAct X/Twitter To determine if the primary or secondary hashtags were trending on X/Twitter within our custom search windows, we relied on archive.twitter-trending.com , a website that provides access to historical trend data in each country at 30-minute intervals. For most trending topics, the website—which collects its information from X/Twitter’s API—also lists the topic’s Tweet volume for the previous 24 hours. For each of a campaign’s hashtags, we identified whether they were listed in the top ten trending topics in India during our search window. To test the accuracy of the archive site, we manually compared the topics it listed as trending against current trending topics on the X platform at different times throughout our research; in all instances, the information matched exactly.
We relied on the third-party site rather than X/Twitter’s API subscription since the API does not provide historical data on trending topics.
CrowdTangle We used Meta’s CrowdTangle platform to search for Facebook posts that used any of the campaigns’ (primary or secondary) hashtags. CrowdTangle provides posts from public Facebook pages, public Facebook groups, and verified Facebook profiles. We performed each search query manually using the web platform and downloaded the results. Additionally, we used CrowdTangle’s API to identify which of the posts came from “verified” accounts, political accounts*, and media accounts*. We focused on the following pieces of information, gathered via those two methods, for each post:
Final Campaigns Selection The 137 Trend Alert WhatsApp messages we identified were in support of 40 distinct campaigns. Based on data from archive.twitter-trending.com , we found that the (primary or secondary) hashtags for 29 of these 40 campaigns were trending on X/Twitter during our search window. (It is possible that other campaigns trended but went undetected by the archive; see limitations section below for details.) We limited our subsequent analysis and visualizations to those 29 campaigns.
Examples of correlations between WhatsApp Trend Alert timing instructions and actual hashtag trending periods on Twitter/X
We cannot know for certain whether a hashtag was trending on Twitter as a result of a specific Trend Alert message. See the limitations section Quantifying Campaign Effectiveness for more details.
Analysis From the Facebook and X/Twitter data noted above, we calculated the following statistics for each Trend Alert campaign:
Unique Facebook posters
The total number of unique public Facebook accounts that shared the campaign hashtags during our search window, based on data available from CrowdTangle.
Facebook reach
We added the total number of followers of every Facebook account that shared a post with the campaign hashtags. We counted each account's followers only once, regardless of how many posts they shared. We did not deduplicate followers across different accounts. Put another way, if a Facebook user follows multiple accounts that shared the campaign they are counted multiple times in the Facebook reach metric, once for each account they follow.
Approximate periods and time trending on X/Twitter
The approximate total amount of time a campaign hashtag was among the top ten trends on X/Twitter within our search window. Since our source provides data every 30 minutes, we first compiled a list of every time a particular campaign hashtag was present in the top-ten trending topics within our search window. We multiplied the number of times that the campaign hashtag appeared in the top ten list by 30 minutes to estimate the amount of time it was trending. See limitations for more details.
Data portraits providing overview of campaign metrics for Tweet volume, trend duration, and campaign type Data Portraits
For Digital Witness Lab, Accurat design studio built a series of “data portraits'' to visually summarize several aspects of each Trend Alert campaign. Each portrait conveys the campaign’s hashtag, start time, end time, type, trend duration on X, and number of tweets posted during the trending period. The data portraits enable visual observation and comparison of individual campaigns while maintaining an overall understanding of the entire group.
Limitations Non-representative sample
The WhatsApp groups where we discovered these Trend Alert messages are not representative of all Indian political parties or campaigning activities on WhatsApp. Groups in our dataset are from a select sampling of political campaigns and political parties in India that were accessible through our research.
Short, focused timeframes
Our WhatsApp Trend Alert Tracker focuses only on the day before, of, and after a campaign was set to take place. We do not account for times the hashtag may have trended on previous days or if the trend persisted more than 24 hours after the initial Trend Alert message collected in our dataset.
Deleted posts
We collected Facebook posts from CrowdTangle after the campaigns had passed. We can not account for posts that may have been deleted before we downloaded the data.
Reach vs. exposure
For our Facebook reach metric, we summed each unique posters’ follower count as a way to compare campaigns’ potential reach. We do not presume that all of the followers of each account viewed the post, and do not attempt to measure such exposure. Our metric also likely overcounts each message’s total reach, since any given user may follow more than one account that posted the campaign hashtags.
Cause and effect
Although our analyses, in many cases, identify substantial increases in Facebook and Twitter posts using Trend Alerts’ hashtags at the time the alerts suggested, we can only suggest that the Trend Alerts led to this activity. Campaign-specific research and participant interviews would be required to find more direct evidence of WhatsApp Trend Alert messages being catalysts for these trends.
Twitter data access
API access necessary to collect a comprehensive set of all X/Twitter posts mentioning a hashtag was prohibitively expensive. To overcome this obstacle, we attempted to collect Tweets programmatically via X/Twitter’s advanced search feature. However, the search results were inconsistent and unreliable; the same query returned different Tweets on different days, and for many campaigns, the search returned fewer than 50 results. This contradicts the post volume we found on the trend archives website, which is based on counts provided by X/Twitter’s official API. For this reason, we rely solely on the trend archive at the expense of losing granularity of individual Tweets and accounts.
WhatsApp group ownership ambiguity
We do not know if all of the WhatsApp groups in our sample are managed by political operatives or party volunteers. We assume that groups have a political affiliation based on party names being in the name of the group. All of our data is anonymized, so we do not see the names or numbers of those who shared Trend Alert messages.
Twitter trending timelines
The source we use to identify hashtags that trended on X/Twitter and the periods during which they trended, archive.twitter-trending.com/india , lists X/Twitter’s trending hashtags in 30-minute intervals. For this reason, we do not know the precise moment a trend began or ceased to be “trending.” Similarly, hashtags that trended only briefly may not be represented in this archive. Additionally, the archive contains a substantial gap in observations for parts of April 8 and 9, 2024, as well as much smaller gaps on a handful of other days. For this reason, we have under-counted the trending campaigns when we performed the filtering step described in the “Final Campaigns Selection” above.
No visibility into non-public Facebook content
CrowdTangle, the service we used to gather data on Facebook posts, only provides access to posts made by public accounts. Some campaigns may be popular in large-but-private groups. Due to lack of access, we cannot account for this activity.
Quantifying campaign effectiveness
We limited our analysis to campaigns that shared hashtags that also appeared as trending on the X/Twitter platform during our search window.
Based on manually checking campaigns against the X/Twitter interface, we did find several instances of Tweets containing text and images that matched the text in Tweet Bank documents shared in Trend Alert messages’ linked Google drive folders. The example below, duplicating even the numbering present in the Tweet Bank document, demonstrates the copy/paste nature of these campaigns.
Screen capture from X/Twitter search for #HBDBabu showing repeated text patterns and evidence of copy-pasted content, including numbered tweet examples from a source document
We faced similar challenges quantifying the effectiveness of campaigns on Facebook. While the CrowdTangle interface gave us access to posts for our given time frame, the average volume of Facebook posts was much lower than Tweet volume, demonstrating that X/Twitter is still the more popular platform for these viral campaigns. Since 2018, Facebook’s trending feature has been removed so it was not possible to measure whether a campaign trended on the platform. The CrowdTangle platform will no longer be available after August 14, 2024, so our purview into trending activity will be further limited.
Conclusion Digital Witness Lab’s WhatsApp Trend Alert Tracker illustrates how messages shared on political WhatsApp groups coincided and potentially influenced political campaigning on X/Twitter and Facebook during the 2024 Indian elections. When a hashtag appears in a flurry of posts across public platforms, coordinated partisan campaigning beyond these hashtags may seem like causes or attacks that emerged organically. Our WhatsApp Trend Alert Tracker gives a retroactive analysis of the cross-platform results that may be attributed to this coordination. This visual tool, built in collaboration with Accurat, provides an interactive comparison into the types of Trend Alert campaigns and the reach and influence they may have had in the lead up to India’s 2024 national and state elections.
