Contraceptive Misinformation on Social Media

This research is supported by the Society for Family Planning.

SPF19-MDI2 (01.2026 - 12.2027)

Ritwik Banerjee, Principal Investigator

Contraceptive misinformation is not a monolithic phenomenon amenable to simple keyword detection or binary fact-checking. It propagates through polarized epistemic networks — faith-based abstinence communities, wellness conspiracy ecosystems, reproductive justice advocates — each with its own rhetorical vocabulary, internal amplification dynamics, and relationship to scientific evidence. Algorithmic recommendation systems on platforms like TikTok, YouTube, and Reddit do not merely host this discourse: they actively accelerate it, by prioritizing engagement over accuracy and surfacing belief-consistent content to users already predisposed to accept it. The result is a self-reinforcing feedback loop in which community-specific distortions of contraceptive evidence gain virality precisely because they confirm what a given audience already believes.

Despite the urgency — heightened by the post-Dobbs landscape in which contraceptive access and reproductive rights have become actively contested — the computational research base is thin. Existing studies rely on broad keyword tracking, which sacrifices specificity, or on qualitative thematic analyses, which cannot scale to the volume and velocity of social media discourse. There is no unified computational framework for identifying how rhetorical strategies vary by community and platform, how algorithmic amplification biases operate at the level of individual psycholinguistic profiles, or how misinformation propagates and gains reinforcement as it traverses interconnected platforms. This project addresses these gaps directly.

Community-Specific Rhetorical Models

The first research question asks how distinct communities differentially distort contraceptive evidence, and which rhetorical strategies most effectively amplify misinformation within each community’s belief network. The computational approach combines frame-semantic parsing to detect distortion tactics (science-denial framing, false causality claims, evidence contradiction), discourse analysis grounded in rhetorical structure theory adapted for multi-platform contexts, and Hawkes process modeling to quantify the self-exciting dynamics of misinformation diffusion across over one million posts from Twitter/X, YouTube, TikTok, Reddit, and health forums. Across this corpus, the analysis will produce the first cross-platform taxonomy of contraceptive misinformation strategies — a resource enabling community-sensitive NLP classifiers for real-time detection of community-tailored distortions.

Algorithmic Amplification and Belief Reinforcement

The second research question addresses a more subtle mechanism: the extent to which algorithmic recommendation systems reinforce belief-consistent distortions, and how this amplification varies by user psycholinguistic profile. Rather than relying on observational correlational analyses, this work employs a controlled agent-based auditing framework. AI agents simulating users with diverse psycholinguistic profiles — varying in cognitive rigidity and identity-driven motivation, operationalized through linguistic analysis — interact organically with platform recommendation systems, generating counterfactual interaction traces that isolate how algorithmic parameters respond to identical queries from users with divergent belief profiles. Cross-platform propagation is tracked through temporal knowledge graphs of shared URLs, capturing how contraceptive misinformation gains reinforcement as it traverses platforms. Causal inference methods — specifically Neyman-Rubin counterfactual modeling — then establish whether, and how strongly, user psycholinguistic profiles causally drive increased exposure to belief-distorted content.

Equity at the Center

Computational rigor without equity is insufficient for research of this kind. This project explicitly centers communities that are disproportionately targeted by contraceptive misinformation — including youth in abortion-restricted states, low-health-literacy groups, and LGBTQ+ communities — by operationalizing vulnerability through intersectional indices derived from geolocation data, linguistic markers of structural marginalization, and network isolation metrics. Detection models are trained on data enriched with content from Black, Latinx, and LGBTQ+ digital communities to ensure recognition of culturally specific misinformation narratives. All outputs will be released as open-source toolkits designed to be accessible to under-resourced public health departments, with no requirement for specialized technical infrastructure.

This is an active project. Publications will be listed here as they become available.

This project page is hosted and maintained by the principal investigator, Dr. Ritwik Banerjee.