How Scores Work
Last updated June 2026
This page explains how to read ChatVerify's consensus score and confidence levels so you can interpret a verification correctly. Scores are a guide, not a guarantee.
Key takeaways
- The consensus score (0–100) reflects how strongly AI systems agree across all dimensions.
- Confidence is separate — it reflects how much the consensus itself can be trusted.
- High-stakes topics always warrant independent verification, regardless of score.
The consensus score (0–100)
The consensus score measures how strongly credible AI systems agree across the core conclusion, reasoning, recommended actions, and caveats. We use a strict five-tier scale:
90–100: Near-unanimous agreement. 75–89: Strong agreement with only minor caveats. 50–74: Partial agreement — meaningful differences exist. 25–49: Significant disagreement. 0–24: Highly conflicting answers.
We calibrate conservatively. Because most real questions involve genuine nuance, the majority land in the partial band. A 'strong' score is reserved for settled, uncontested matters.
Don't just trust — verify
Run your question through ChatVerify and compare answers across leading AI systems.
Confidence level (high / medium / low)
Confidence is separate from agreement. It reflects how much you should trust the consensus itself, considering evidence quality, recency, source authority, and how fast the topic changes.
A topic can have strong agreement but only medium or low confidence — for example when data is sparse, fast-changing, or based on weak sources.
Why we still flag verification
For medical, legal, tax, investing, and insurance topics, we mark verification as required even when consensus and confidence are high. Consensus never eliminates the need for independent confirmation in high-stakes decisions.
Treat the score as a triage signal: high scores mean fewer things to check, low scores mean more — but neither replaces reading the sources or consulting a professional.
Frequently asked questions
Why is my score in the 'partial' band so often?
Because most real questions involve genuine nuance. We deliberately avoid overstating certainty and reserve high scores for truly settled matters.
Can the score be wrong?
Yes. The score reflects AI agreement, which can be confidently mistaken. Use it alongside the sources and verification gaps, not on its own.