VeritasAI extracts discrete factual claims from any article you're reading, verifies each one against primary sources and cross-outlet coverage, and surfaces confidence levels — without ever issuing a verdict.
✓Claim-by-claim verification
Extracts 5–8 checkable claims and verifies each independently with statistics, quotes, attributions, stated events. Not a summary. Not a sentiment score.
↔Political lean labeling
Every source labeled Left to Right so you see where support comes from. Not just how many outlets agree, but whether they span the political spectrum.
⊕Coverage divergence
CNN says X, Fox says Y, primary source says Z, all surfaced automatically. Identifies where outlets with different lean labels specifically disagree, and why it matters.
◎Live credibility score
A confidence bar updates in real time as each claim resolves — you see how an article holds up overall before all claims finish. Color-coded green to orange based on the average across all checked claims.
⊙Breaking story detection
Articles classified as Breaking, Developing, or Stable before fact-checking starts. Sources older than 12 hours on a breaking story trigger automatic confidence downgrades — stale evidence is never treated the same as fresh reporting.
✎Highlight to fact-check
Select any sentence in the article and fact-check it on demand. Results appear at the top of the sidebar, verified with the same rigor, lean labeling, and primary-source ranking as the full pipeline.
Article is loaded
Open any news article and click the VeritasAI icon. The extension extracts the article body and sends it to the pipeline.
Claims are extracted
An AI agent identifies 5–8 discrete, checkable factual claims — statistics, quotes, attributions, stated events. No summaries, no sentiment.
Parallel verification
Each claim is independently searched against live web results. Sources are lean-labeled, primary sources separated, and a synthesis agent weighs the evidence.
Results stream in
Confidence levels, source cards, and divergence summaries arrive claim by claim — no waiting for the full analysis to finish before you see the first result.
Never issues a true or false verdict.
Confidence levels and cited sources — you decide.