How RageCheck detects manipulative patterns in content.
If you're seeing the "Techniques Detected," "Viral Triggers," or "What Follows" sections in a report, this page explains the model behind them.
RageCheck uses a two-stage analysis pipeline: rule-based pattern detection followed by optional AI-powered contextual analysis. This hybrid approach balances speed, transparency, and accuracy.
The system analyzes text for linguistic patterns commonly associated with manipulative framing—language optimized to provoke high-arousal reactions over understanding. It does not assess factual accuracy or political bias.
RageCheck organizes analysis the same way influence typically works in the real world:
Note: This is probabilistic pattern detection. It does not infer intent and it is not truth scoring. It estimates patterns that tend to correlate with higher-arousal, lower-nuance media.
Content is analyzed across five distinct signal categories, each targeting specific manipulation patterns:
Language designed to activate strong emotional responses—fear, anger, disgust, or outrage—rather than inform.
Examples: "Shocking," "horrifying," "unbelievable," "disgusting," inflammatory adjectives, ALL CAPS emphasis, excessive punctuation, urgent calls to action.
Framing that constructs an adversary—dehumanizing groups, attributing malicious intent, or creating artificial us-vs-them divisions.
Examples: "They want to destroy," "the elite," "those people," "the enemy within," collective blame, conspiracy framing, dehumanizing labels.
Appeals to moral outrage, purity rhetoric, and righteous indignation that frame issues as battles between good and evil.
Examples: "Evil," "immoral," "corruption," "betrayal," "disgusting behavior," virtue signaling, moral absolutism, purity tests.
Black-and-white framing that eliminates nuance, presents false dichotomies, or reduces complex issues to simple narratives.
Examples: "Always," "never," "everyone knows," "the only solution," "it's simple," false equivalences, strawman arguments, ignoring counterevidence.
Direct provocations, engagement bait, and language designed to mobilize action or spread content virally.
Examples: "Share before they delete this," "wake up," "fight back," "they don't want you to know," rhetorical questions designed to provoke, call-outs.
The first stage uses pattern matching against curated dictionaries of manipulative phrases. Each category has weighted terms—stronger manipulative signals receive higher weights.
Scores are normalized per 1,000 words to account for content length, ensuring short tweets and long articles are compared fairly.
When available, Claude AI reviews the rule-based findings to add context. This stage can adjust scores based on factors rules can't capture:
Minimal manipulation signals
Some concerning patterns
Significant manipulation density
RageCheck can analyze screenshots of social media posts, memes with text, and news headlines using computer vision.
This allows analysis of content that can't be linked directly—screenshots shared in group chats, posts from private accounts, or memes circulating on social media.
RageCheck's signal categories draw from established research in media psychology, propaganda studies, and affective computing: