Question 1

How does the AI bias checker work?

Accepted Answer

Paste an article or news story. The tool tokenizes it into sentences and scans every phrase against built-in lexicons of loaded verbs, subjective adjectives, weasel words, hedges, passive-evasion patterns, left-leaning and right-leaning terms, and clickbait patterns. Each flagged phrase is highlighted inline, and sentences are tagged fact/opinion/mixed/neutral. The tool computes scores across five axes (political lean, emotional tone, factual density, one-sidedness, sensationalism) and offers a neutral-rewrite of the most loaded sentences.

Question 2

Does it judge whether the article's claims are true?

Accepted Answer

No. Bias detection is about language patterns — not truth. The tool flags how something is said (emotive adjectives, missing attribution, framing) and is explicit in the UI that it does not fact-check the claims themselves. Verifying accuracy requires independent fact-checking; this tool only surfaces linguistic signals of bias and rates the text, not the outlet.

Question 3

How is political lean estimated?

Accepted Answer

Lean is estimated from the relative frequency of left-leaning vs right-leaning loaded terms (e.g. 'progressive' vs 'woke', 'gun control' vs 'second amendment rights'). Confidence is reported alongside the lean label (left / lean-left / center / lean-right / right) and is lower when both sides appear or when total political vocabulary is sparse. The reasoning is shown so you can audit the call.

Question 4

Can I get a neutralized rewrite of the article?

Accepted Answer

Yes. Each flagged phrase carries a suggested neutral replacement. Click 'Neutralize' on any sentence to apply suggestions, or use the 'Neutral rewrite' button to produce a full-text version with loaded terms replaced, subjective adjectives stripped, and weasel words rewritten with attribution placeholders. The rewrite is a starting point for editing, not a final polish.

Question 5

What extra features does this tool have compared to others?

Accepted Answer

(1) 7 lexicon categories (left-lean, right-lean, loaded verbs, subjective adjectives, weasel words, hedges, absolutisms). (2) Passive-evasion detector ('mistakes were made'). (3) Clickbait pattern detector with 7 regex patterns + caps-density check. (4) Multi-axis scoring: political lean (−100 to +100 with confidence), emotional (0-100), factual (0-100), one-sidedness (0-100), sensationalism (0-100). (5) Inline phrase highlighting with category + suggestion. (6) Per-sentence fact/opinion/mixed/neutral tagging. (7) Missing-perspective checklist with detection cues. (8) Flesch readability + grade level. (9) One-click neutralize (single sentence) and full-text neutral rewrite. (10) Lean reasoning (human-readable explanation). (11) Sensationalism heatmap (per-sentence). (12) Export to Markdown, JSON, and highlighted HTML. (13) History (localStorage, last 20). (14) Shareable URL. (15) Optional BYO-key LLM enhancement (OpenAI/Anthropic) for nuance the offline lexicons miss.

Question 6

Is my data sent anywhere?

Accepted Answer

No. All lexicon matching, scoring, highlighting, and neutral rewriting runs locally in your browser. The article text never leaves this device. The only network call is if you paste your own LLM API key and click 'Enhance with LLM' — that request goes directly from your browser to the LLM provider you choose (OpenAI or Anthropic).

AI Bias Checker for Articles

How to use