Check every citation
before you file
Paste a draft. Every case citation is cite-checked against 10 million opinions in seconds: invented cases, wrong volumes, misquotes, and whether the case supports the point. For AI-drafted and hand-drafted briefs alike.
US case law and Swiss decisions (BGE/ATF/DTF, BGer, BVGer, BStGer, all 26 cantons; drafts in German, French or Italian). Processed in memory, then discarded. Only counts are logged.
Built on open registers
Two checks.
One open register.
Default check
Exists, is the case you named, says what you quoted.
Deep check
Says what you cited it for, with the passage.
What it checks, and how often it is wrong
Existence
89% of invented cases caught.
Quotation
14 of 15 invented quotes caught.
Support
55.0% of good citations confirmed, 4.5% of wrong pairs marked as support, at the confidence it shows. A review queue, not a verdict.
Unverifiable ids
A quarter of citations. Listed, never flagged.
invented cases caught
invented quotes caught
good citations confirmed by the deep check at the confidence it shows; 4.5% of wrong pairs marked as support
Register lookup, name matching and quotation matching.
On LePhantomCite, Princeton's public benchmark, the default check (engine 0.5.0) caught 31 of 31 citations in reporter series that do not exist, 70.6% of case-name mismatches and 54.3% of altered quotations; it does not check the other two classes. Results and caveats.
Word, PDF, your own tools, or your agent
Attach the PDF report to the file. Or give the check to Claude or an OpenAI agent: an open-source MCP server for Claude Desktop, Claude Code, Cursor and the OpenAI Agents SDK, npx -y proofread-mcp. MCP server API
Questions lawyers ask
How do I check whether a case citation is real?
Paste the draft above. Each citation is looked up in a register of 10 million opinions: the case at that volume and page, its name, court and date, and the opinion text. A citation that is not there is flagged with the volume's coverage, so an absence is shown as what it is. Then open the flagged ones at the source. The steps by hand, with free tools.
Does ChatGPT invent case citations?
Yes, and so do the other models. In our measurement, 300 paragraphs drafted by three language models carried 7 to 21% invented cases and 4 to 14% misquotes; 7,067 citations from 141 briefs filed by lawyers carried none. Courts have sanctioned lawyers for filing the invented ones. The numbers.
What is a cite check?
Reading every citation in a brief to confirm the case exists, the volume and page are right, the quotation matches the opinion, and the case says what the brief cites it for. A paralegal does it by hand in hours; this does the first three in seconds and the fourth with an opt-in deep check that shows the passage.
Does it check Westlaw and Lexis citations?
Westlaw and Lexis identifiers (2024 WL 123456) cannot be resolved by any open register. They are listed as unverifiable, never flagged; about a quarter of the citations in filed briefs. Reporter citations (F.3d, U.S., S. Ct., state reporters) are checked. Whether a case is still good law is not checked; this is not a citator.
Does it check Swiss citations?
Yes: BGE/ATF/DTF, Federal Supreme Court dockets (4A_281/2015), the federal courts and the courts of the 26 cantons, 1.1 million decisions, in German, French or Italian. The report comes back in the language of the draft. Coverage.
Can I check the other side's brief?
Yes. A filed brief is a public record, so there is no privilege question: upload the PDF and every case citation comes back with what the register holds at that volume and page. How to check their citations.
Is my brief stored?
Not unless you save it to your account. The text is processed in memory and discarded when the report is returned; only counts are logged. Privacy.
"Zero invented cases in 7,067 citations from filed briefs. 21% in the LLM drafts."
