Catch the fake citation before the court does.

We extract the legal citations from the text you paste or the file you upload — up to 50 per check — then run each one through two checks. First, existence: we look the citation up live against the official source — CaseRead's public law database for statutes and regulations (over 4 million sections across all 53 US jurisdictions: 50 states, DC, Puerto Rico, and federal) and the Free Law Project for US court opinions (over 9 million decisions from 2,000+ courts) — plus, for court rules and state constitutions, the court-rules and constitution texts in CaseRead's public law database where it carries them and the court's own official rules website where it does not. Second, support: where the source is found, we read its own text — plus a live web search — to judge whether it actually backs the claim it's attached to. Citations we can't match are flagged for manual review; citations that exist but don't back the claim are flagged as possibly misused.

Up to 50 per check. That covers most briefs and memos outright. If your text contains more than 50 citations we check the first 50 and the results page says so — every citation past the 50th is never looked up, and a clean result does not cover it. On a long brief, split the document and run it in parts. This is a deliberate bound on what each free check costs to run, not an estimate.

The tool is free, so we ask for an email in exchange for the full verification breakdown. You'll see the summary and full detail for the first 3 citations (highest-risk first) immediately; the remaining citations unlock after email. No signup, no credit card, no password. CaseRead's founder, Collin, may also send you one personal email about your report, from collin@caseread.ai.

Not necessarily. It means the live lookup didn't return a match — the citation could be fabricated, mis-cited, very obscure, from a municipal code, a foreign jurisdiction, a treatise, or an unpublished opinion. Always review 'Not found' citations manually before relying on them.

CaseRead does not persist the pasted text anywhere in our systems — only aggregate counts (how many citations were found, how many verified) are recorded against your email. To run the check, your text and the citations we extract from it are sent to a few external services: Anthropic (Claude) extracts the citations and judges whether each source supports its claim; our public-law and case-law data providers (named on our privacy page) receive the individual citations to look them up; federal enacted-law citations (public laws, Federal Register documents, CFR sections) may also be checked directly against three U.S. government sources — GovInfo, the eCFR, and the Federal Register; Brave receives search queries derived from each citation for the web cross-check; and, for court rules, the citing jurisdiction's own official court-rules website receives a request for the rule's page (Utah Courts today; the registered list is in the product). Each provider's standard API policy applies. One exception: if you run a check while signed in to a CaseRead account, the verification results (not the raw pasted text) are saved to your private history so you can return to them. If you're handling privileged material and need stricter data-handling guarantees, please contact us before using the tool.

Free — every visitor gets 2 checks per day. Signed in with a free CaseRead account, your checks come back in full with no email step, are saved to your private history, and pull supporting evidence from your own firm's vault; members on a paid plan get unlimited checks in the in-app Shield. This is a public utility we maintain as part of CaseRead.ai, our AI-powered legal research platform — starting an account is free, no credit card required.

Accuracy depends on the coverage of the sources we look up — CaseRead's public law database, which covers all 53 US jurisdictions (50 states, DC, Puerto Rico, and federal, including the US Code and CFR), and the Free Law Project for US court opinions. Ambiguous citations (short forms, partial references) are flagged as 'inconclusive' so you can review them yourself.