Public FVR Scorecard · updated every Monday
The only published error rate in legal AI citation checking
We publish our False-Verify Rate (FVR) — the rate at which we stamp VERIFIED on something that is actually fake or wrong — every week, per layer, with the eval-set size and last-measured date. No other legal citation tool we could find publishes this number at all. That is not a coincidence; it is the entire premise of the Argus Standard (see the whitepaper).
Citation Safe — live, by layer
No persisted run yet for this layer. Full live detail on /quality.
Industry benchmark — third-party research
The most-cited independent benchmark of legal AI tools is the Stanford RegLab / Stanford HAI study (Magesh, Surani, Dahl et al., 2024, “Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools”), which tested leading legal-research AI products against real queries and measured how often each one produced a hallucinated or incorrect legal citation/proposition. AI ethicist Reid Blackman has separately written about these findings as a cautionary tale for professional AI adoption. We did not run this study — we are citing it, with the source named, exactly as the no-unsubstantiated-claims standard requires.
| Tool | Published error rate | Source |
|---|---|---|
| Citation Safe (us) | See per-layer FVR above — updated weekly, live | Our own eval set, methodology at /argus-standard |
| Westlaw AI-Assisted Research | ~33% hallucination rate | Stanford RegLab/HAI, 2024 |
| Lexis+ AI | ~17% hallucination rate | Stanford RegLab/HAI, 2024 |
| Clearbrief | Not published | clearbrief.com/pricing (no error-rate disclosure found) |
| Casetext / CoCounsel (Thomson Reuters) | Not published | legal.thomsonreuters.com (vendor-commissioned Forrester stats only, not an error rate) |
| CiteCheck AI | Not published | citecheck.ai (existence-only checker, no published accuracy) |
| JurisCheck | Not published | jurischeck.com |
“Not published” means: as of this page’s last edit, we could not find a self-published, methodology-disclosed error rate on the vendor’s own site or in a peer-reviewed/third-party study. If that changes, tell us and we will update this table — that is the entire point of a page like this existing.
Methodology
Full methodology, eval-set construction, and refund contract are documented in the Argus Standard whitepaper. Live per-layer detail, dispute stats, and coverage map are on /quality. This page is the weekly-snapshot, competitor-benchmarked summary of the same underlying data.