Helium’s bias analysis provides an empirical, AI-driven measure of news bias. By evaluating articles across dozens of dimensions without human input, we quantify the tone, intent, and quality of journalism.
How It Works
Using zero-shot learning, our AI probabilistically scores each article across a wide spectrum of biases, fallacies, and quality metrics.
Instead of just labeling an article “biased,” we break it down into specific, measurable dimensions:
Spectrum Scores (-10 to 10)
Each score places an article between two poles. Zero is neutral.
Political & Economic
-
Liberal <—> Conservative 
-
Libertarian <—> Authoritarian 
-
Anti-establishment <—> Pro-establishment 
-
Bearish <—> Bullish 
-
Dovish <—> Hawkish 
-
Individualist <—> Collectivist
-
Anti-corporate <—> Pro-corporate
How They Think
-
Objective <—> Subjective 
-
Truth-seeking <—> Delusional 
-
Descriptive <—> Prescriptive -
Rational <—> Irrational
-
Scientific <—> Superstitious
Quality & Sentiment
-
Uncredible <—> Credible 
-
Low Integrity <—> High Integrity 
-
Low Intelligence <—> High Intelligence 
-
Negative <—> Positive 
-
Boring <—> Interesting
Bias Indicators (0 to 10)
These flag the intensity of specific toxic or lazy reporting patterns. Zero means absent.
Emotional Manipulation
-
Sensationalism -
Fear -
Manipulation -
Victimization -
Scapegoating -
Suicidal Empathy — spineless capitulation disguised as compassion -
Cruelty -
Immaturity -
Overconfidence
Logical Fallacies
-
Begging the Question -
Circular Reasoning -
Appeal to Authority -
Double Standards -
Oversimplification
Social & Ideological
-
Gossip -
Virtue Signaling -
Woke -
Ideological Rigidity -
Conspiracy -
Hypocrisy
Content Flags
-
Opinion -
Political -
Covering Responses — reporting reactions to events instead of the events themselves -
Spam -
Advertising
-
Speculation
-
Terrorism sympathy
AI Detection & Transparency
-
Written by AI (0–10) — detects the likelihood an article was generated by AI.
-
My Bias — the AI explicitly introspects on its own training data limitations and blind spots for each evaluation, so you know what it might get wrong.
What Gets Extracted
Beyond scores, the system pulls out structured context to save you time:
-
A nuanced bias summary explaining the specific biases present and why.
-
A concise article summary and factual context placing the news in its broader setting.
-
Key data points extracted directly from the article.
-
Automatic categorization (Finance, Politics, Science, Technology, etc.).
By separating what happened from how it’s reported, Helium lets you see the news clearly.
