Skip to main content
Every answer the AI produces is automatically checked against the actual documents and cases it cited, claim by claim, before you see it. This is not web fact-checking - it verifies that what the model said is actually supported by the sources it pulled.
Response Verification panel showing overall confidence and each claim marked Supported or Refuted against its sources

Every claim traced to its cited source text

Verification runs on every standard chat, Deep Research output, Legal Tool result, and Agent Mode final answer. There is nothing to enable - it is on by default.

​
What Gets Caught

The verification layer is designed to catch the failure modes that matter to lawyers:

​
How It Works

Each claim in the AI’s response is checked against the source it cites:
1

Exact match

For direct quotes, the words must appear in the source verbatim.
2

Citation validation

For case and statute citations, the citation must resolve to a real authority.
3

Semantic match

For paraphrased points, the meaning must be supported by the source passage.
4

Reasoning check

For inferences, the chain of reasoning must hold up against the cited material.
Claims that pass all four are marked Verified. Claims that pass some but not others are Partially Verified. Claims that fail are flagged or removed before the response reaches you.

​
What You See

Each response carries a verification badge:
  • Verified - Every material claim was fully supported by sources
  • Partially Verified - Most claims passed; some had weak support and are marked inline
  • Unverified - One or more material claims could not be backed up - the response shows what failed and why
Hover over any sentence to see which source supports it and at what confidence level.

​
Why This Matters

The traditional risk with AI research is that an answer looks polished but contains a hallucinated case or a misstated holding. Verification flips the default: instead of you having to cite-check every paragraph, the system does it first and tells you exactly where to double-check. You should still confirm critical authorities before filing, but the verification badge tells you where to focus that review.

​
When Verification Runs

  • Standard chat responses
  • Deep Research outputs
  • Legal Tool outputs (contract review, NDA triage, compliance check, etc.)
  • Agent Mode final answers

​
Limits

Verification works against the sources the system retrieved. It cannot:
  • Tell you whether a case is still good law (use Citation Graph for precedent freshness)
  • Tell you whether a statute has been amended (use Statutes & Regulations for amendment history)
  • Tell you whether the cited source is itself reliable - that judgment is yours
A Verified badge confirms a claim is supported by the cited source. It does not certify the source itself is current, controlling, or applicable to your facts. Cite-check critical authorities before filing.
Treat verification as a strong filter against fabrication and misquotation, not as a substitute for legal judgment.

​
Tips

Click through the badge. The expanded view shows every claim mapped to its source - hover any sentence to see exactly which passage supports it and at what confidence level.
Investigate Partially Verified responses first. These are the spots where the AI hedged or stretched; that is where misstatements hide.
Pair with the Citation Graph. Verification confirms a case was cited correctly; the Citation Graph tells you whether that case is still good law.

​
Verify an answer yourself

Cite-Check - re-verify an existing answer claim by claim against the actual cited sources.

Deep Research Mode

Multi-hop research with built-in verification at each round.

Citation Graph

Check whether cited cases are still good law.

Document Search

How retrieval feeds verification.