Our method

AI Truth runs on AI too, so it can be wrong. Its edge is the method, and we show it openly.

How one question is answered

1. Understand the question and load an expert pack for its field (health, money, science, products, news).
2. Gather free sources first: research databases, health agencies, government records, recall lists, encyclopedias and news indexes.
3. Look for what's missing: one-sided or old sources, missing evidence types, hidden assumptions in the question.
4. Several AIs from different companies answer, each twice: once on its own, and once from the gathered evidence. The difference between the two is "What AI won't tell you."
5. A separate judge (never one of the panel) checks every claim against the sources and writes the answer page.

What the labels mean

Proven two or more independent quality sources agree (official data, reviews of many studies, studies, primary records).

Likely good evidence points this way, but it is not settled, or only the AIs agree.

Disputed credible sources disagree. We show both.

Unknown nobody has good evidence yet. We say so instead of guessing.

AIs agreeing is weak proof: they often share the same mistakes. So agreement alone never makes a claim proven. This rule is enforced in code after the judge, not left to the AI.

How we stay honest

• The other side is always shown, in its strongest honest form.
• Sources are scored by evidence type, not by politics.
• Anyone can flag an error. Fixes will be logged in public.

Built by AI and checked against sources. Check the sources before big decisions. Not medical, legal or financial advice.