Misinformation and its Moderation on WhatsApp

Why WhatsApp matters

Scale and global reach: Over 2 billion active users globally; 600 million in India, nearly 100% of Brazil's internet population, ~25% of the US (comparable to Twitter penetration but vastly overlooked by researchers and policymakers).

Platform design reduces misinformation spread:

• End-to-end encrypted: WhatsApp servers cannot see message content; only know sender and recipient metadata.
• Decentralized one-to-one communication by default; no broadcast button to reach millions instantly.
• No algorithmic feed, news feed personalization, or engagement metrics (likes, shares, follower counts).
• No artificial virality incentives; users cannot see how many people saw their message.
• Higher trust: information flows through close friends and family, not strangers or influencers.

Result: Spreading misinformation on WhatsApp is structurally harder than on Twitter or Facebook, but billions of users and strategic actors still make it happen.

Political coordination and cross-platform amplification

Infrastructure in India: Political parties (especially BJP) operate thousands of WhatsApp groups with coordinated messaging. Study finds ~60% of collected groups were BJP-related in one dataset, >50% in another.

Surprising finding — limited explicit political content: Despite being party-created groups, only ~40% of messages are politically related. The rest: local selfies, religious events, community gatherings. Hypothesis: groups build local community identity and social bonds, not just message dissemination.

Cross-platform coordination: Researchers documented coordinated inauthentic behavior linking WhatsApp to Twitter. Political operatives send WhatsApp group members Google Docs containing:

• Specific hashtags and timestamps.
• Template tweets (dozens to hundreds of examples).
• Instructions to log into Twitter at a set time and post the template tweets verbatim or with minor edits.

Result: Hundreds of users copy-paste the same tweet simultaneously, creating artificial trending hashtags and perceived grassroots momentum on a public platform.

Data collection methodology

Challenge: End-to-end encryption makes external monitoring impossible; WhatsApp does not provide researcher access.

Solution — Opt-in data donation: Researchers go door-to-door or work with local surveyors:

1. User scans a QR code on researcher's system (similar to logging into web WhatsApp).
2. User consents to share specific groups (>5 members, sufficient activity).
3. Researcher receives: 2 months retroactive + 2 months prospective data; anonymized group messages, contact network, user survey.
4. All identifiers (names, emails, phone numbers, faces) removed before storage.
5. Users receive modest monetary compensation.

Pilot results: ~400 users in north and central India (Uttar Pradesh, Jharkhand) recruited via door-to-door canvassing. Challenges: rural populations have low WhatsApp adoption; random sampling yields few large group participants.

Moderation and solutions

Problem: Scale mismatch. WhatsApp reports 55 billion messages sent daily. During 2019 Indian elections, WhatsApp's heavily-advertised public tip line received 100,000 reports over 3 months—1 per 16.5 billion messages. Opt-in fact-checking does not scale.

Community-driven fact-checking: 40% of misinformation in datasets is already debunkable via reverse image search or easy verification. Rather than top-down expert fact-checking:

• Crowdsource fact-checking within trusted networks (friends, family, local community).
• Use technology (e.g., image hashing) to identify previously-debunked content in encrypted contexts.
• Deploy Community Notes-style annotations (bottom-up correction, not removal).

Challenges:

• Requires operating within encrypted channels without content visibility.
• Risk of gaming or misinformation in crowdsourced corrections.
• Beliefs rooted in decades-long identity and caste-based hierarchies resist correction.

Scalable direction: Privacy-preserving data collection (aggregate statistics computed on-device, not transmitted) paired with opt-in donation for representative sampling.

Notes

The talk is valuable for its focus on an understudied platform (WhatsApp relative to Facebook/Twitter) and context (Global South political misinformation). The opt-in data donation method is innovative and privacy-respecting, addressing a real barrier to studying encrypted systems. The cross-platform coordination findings reveal an important mechanism: encrypted platforms as organizational infrastructure for coordinating attacks on public platforms.

The transparency about fact-checking's failure at WhatsApp's scale is refreshing and grounded in real data (100,000 tips / 55 billion messages/day). The proposal for community-driven alternatives is pragmatic but requires more development on incentive design, bot-resistance, and belief-persistence research.

Limitations: The talk focuses on political groups, which may not represent private WhatsApp dynamics where most real misinformation may flow. Year of the talk is indeterminate from transcript alone; context clues (2019 election references, 2023 mention, India 2026 election planning) suggest delivery in 2024 or 2025, but confidence is low.