How VeriTrace verifies claims
No black box. This page explains what the AI is allowed to do, where evidence comes from, and the exact rules that turn evidence into a verdict.
The AI never decides what is true
The AI has three narrow jobs: extract checkable claims from your input, label how each retrieved passage relates to a claim (supports, contradicts, context, irrelevant) and write a short explanation using only those labels. Tiers, independence and verdicts are plain code.
Where evidence comes from
- Live web search and news search across several providers.
- Fact-check databases, including published fact-checks.
- Wikipedia and other reference sources.
- Academic indexes: Crossref, OpenAlex and PubMed.
- GDELT news data and the Wayback Machine for archived copies.
Verdict rules
- Supported: at least two independent sources of tier T2 or better support it, and no credible source contradicts it.
- Contradicted: at least two independent credible sources contradict it, or one primary (T1) source does.
- Mixed: credible sources disagree.
- Insufficient evidence: everything else.
Limitations
VeriTrace checks written claims and links. It cannot judge images, audio, video or satire, and fast-moving stories may lack coverage. Always open the cited sources.
Frequently asked questions
Why does VeriTrace sometimes say “Insufficient evidence”?
Because the rules require independent credible sources. When they are missing, saying so is more honest than guessing.
Are results cached?
Identical claims may be cached briefly to keep checks fast. Verdicts are always based on the retrieved sources shown in the report.
Who can see my checks?
Only you. Your history is private to your account and you can delete any check or your whole account at any time.
Is my data used to train AI?
VeriTrace does not train models on your data. Claim text is sent to search and AI providers only to retrieve and label evidence.