The citation errors in your brief that a read-through will not catch.
You read the brief twice. The citations look right. But a quoted passage may differ by a word from what the opinion actually says. A sentence attributed to the majority may come from the dissent. A case may cite-check clean by name but point to an opinion that says something different from what the brief claims. These are the errors that survive proofreading and land on a judge's desk. A mechanical pre-filing check finds them before you file.
General information for Nevada legal professionals. This is not legal advice. Read the full disclaimer.
The candor duty that applies before the brief leaves your office
Nevada RPC 3.3(a) prohibits a lawyer from making a false statement of law or fact to a tribunal and from citing authority the lawyer knows to be invalid. The rule is a floor, not a ceiling: an attorney who files a brief with a fabricated citation, a misquoted passage, or dissent language dressed as a holding has made a false statement to the court, whether or not the attorney knew it at the time of filing.
That last phrase is where AI drafting tools create a new category of risk. A competent attorney catches an argument that is wrong on the law. What is harder to catch, on a read-through of a brief that otherwise sounds right, is a quotation that is off by a clause, a parenthetical that mischaracterizes a concurrence as the court's holding, or a citation to a real case that stands for a different proposition than the one the brief asserts. The brief reads well. The error is mechanical, not substantive. And a read-through does not find it.
Why AI-drafted briefs surface this failure mode
The problem is not unique to AI-drafted content, but AI tools make it more common and more convincing. A Stanford RegLab study published in 2024 found that leading cloud-based legal research tools hallucinated in roughly one out of six or more queries on a structured benchmark. (Magesh et al., Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools, Stanford RegLab 2024; see Stanford HAI summary.) One in six is a rate that guarantees problems across a working docket.
The hallucinations the study documented are not obvious. They are not nonsense citations. They are citations to real cases, plausible reporters, and coherent propositions, where the case exists but the proposition is wrong, slightly shifted, or simply invented. On a fast read, they look fine. What flags them is not legal judgment but mechanical verification: pull the opinion, compare the words, check the source of the quote.
Attorneys face this whether they use AI drafting tools or not. A junior associate who paraphrased from a headnote rather than the opinion, a quoted passage copied from a prior brief without re-checking the source, a case the attorney remembered correctly but cited to the wrong reporter: these errors predate AI. AI makes them more frequent, more plausible, and harder to spot because the surrounding prose is smooth.
The four failure modes a pre-filing check targets
A mechanical check before filing addresses four specific, verifiable errors:
1. The misquote
The brief says the court held that a standard "requires clear and convincing evidence of willful misconduct." The opinion says "requires clear and convincing evidence of intentional misconduct." One word. The quote has quotation marks around it. It looks right. A human read-through at filing speed will not catch the substitution. A word-for-word comparison against the opinion text will.
2. Dissent or concurrence cited as holding
This error is subtler and more damaging. Language from a dissenting opinion is frequently clearer and more quotable than the majority's holding, especially in close cases. An AI drafting tool will pull the language that best supports the argument, which is sometimes language that only one or two justices adopted. When that language appears in the brief without attribution to a dissent, the attorney has misrepresented the state of Nevada law to the court. RPC 3.3(a)(1) directly applies.
3. An unverifiable or miscaptioned cite
A citation may be unverifiable for two reasons: it points to a case that does not exist, or it points to a case that exists but cannot be confirmed to say what the brief claims. Both are problems. The first is fabrication. The second may be fabrication or may be a citation to a real case that was misread, paraphrased beyond recognition, or confused with another opinion. Either way, the supervising attorney cannot sign a brief containing a citation that cannot be verified against the record or a publicly available source.
4. A citation present in the body but absent from the Table of Authorities
NRAP 28(a)(12) requires a Table of Authorities that lists all cases, statutes, and other authorities cited in the brief with references to every page on which they appear. A citation that appears in the argument section but is missing from the TOA is a rule violation. It is also a tell: if a citation was dropped into a paragraph but never made it into the table, the source may not have been checked at all.
Why a read-through does not find these errors
Attorneys are trained to read for argument. When you read your own brief, you read what you meant to write, not what is actually on the page. A misquote that is one word off will read correctly to you because you know the case and your brain completes the sentence. A passage from the dissent that you expected to find in the majority will read correctly because the argument makes sense. A citation that looks real will pass because you are not comparing it against the opinion text; you are checking whether it fits the proposition.
None of this is negligence. It is how human reading works. Catching this class of error requires a different kind of check: one that compares the quoted text in the brief against the actual opinion, character by character; one that knows which portions of an opinion come from a dissent or concurrence; one that can confirm whether each citation can be verified against a source the attorney can point to. That check is mechanical, and it does not require legal judgment to run. It requires that someone actually run it.
For most Nevada appellate practices, that check has historically been skipped because running it manually, against every quotation in a forty-page opening brief, takes hours that do not exist in the days before filing.
How DilloLex runs the pre-filing check
DilloLex's pre-filing QC gate reads the draft brief on the firm's own appliance and flags each instance of the four failure modes above. Here is what that looks like in practice:
- The attorney submits the draft for review. The brief stays on the firm's hardware. No client content goes to an outside AI. The check runs locally.
- The gate reads every quotation and compares it against the source opinion. Where the text of the brief differs from the opinion text, the gate flags the discrepancy, shows both versions side by side, and identifies the location in the opinion.
- It identifies the authorial source of each quoted passage. A passage that appears in the brief as a statement of law is checked against the majority, concurrence, and dissent. If the source is a dissent or concurrence, the gate flags it and identifies the author and section.
- It attempts to verify each citation. Citations that cannot be matched to a verifiable source are flagged as unconfirmed. The gate does not delete them or make a legal judgment about them; it surfaces them so the attorney can check.
- It cross-checks the Table of Authorities. Any authority cited in the body but missing from the TOA is flagged for the attorney's attention.
The gate generates a flagged report. Every flag requires a human response: the attorney fixes the problem or records an explicit override with a reason. The gate does not file the brief, does not override the attorney's judgment, and does not suppress flags without the attorney's sign-off. It is a checklist that runs itself, with the attorney still holding the pen.
The gate has been run on Nevada appellate drafts in-house, and each flag was reviewed before filing. It is not infallible; the attorney reviews every flag and makes the call.
A concrete example
Assume a Las Vegas personal injury firm uses an AI drafting tool to prepare an opening brief in an NRAP appeal. The draft contains fourteen cited cases, four block quotations, and a Table of Authorities. On pre-filing review, the gate flags three items:
- One block quotation differs from the opinion by eight words: three substitutions and one omitted clause. The citation is accurate; the quote is not.
- One sentence cited as the court's holding comes from a two-justice concurrence. The majority did not adopt that formulation.
- One case appears in the argument section but is missing from the Table of Authorities.
None of these would have been caught by a read-through. All three are fixed before filing. The brief goes out clean, and the supervising attorney can account for every citation in it.
That is the purpose of the check: not to replace the attorney's review, but to make it complete.
How DilloLex helps
DilloLex is software that assists a firm's staff. It does not practice law and it does not replace the attorney's judgment. What it does is run a mechanical check that most Nevada appellate practices skip because they do not have time to run it manually: comparing quoted text against source opinions, identifying the authorial source of each passage, flagging unverifiable citations, and cross-checking the Table of Authorities. Each flag is delivered to the supervising attorney, who decides what to do with it. Nothing goes to the court without the attorney's approval. The system flags; the attorney fixes or overrides; the brief goes out under the attorney's name and professional responsibility.
Because the check runs on the firm's own appliance, no client content goes to an outside AI. The brief, the case files, and the flagged report stay on hardware the firm owns.
If you want to see how the gate works on a real Nevada appellate draft, book a demo. You can also read more about how DilloLex handles brief drafting and review, the firm's security posture, or the Bates numbering and production tools that work alongside it.
Common questions
Does RPC 3.3 apply to citation errors the attorney did not know about?
RPC 3.3(a)(1) prohibits knowingly making a false statement of law or fact. The word "knowingly" matters: an attorney who files a misquote without knowing it is a misquote has not necessarily violated the rule in its strictest reading. But the Nevada Supreme Court has also interpreted the competence rule, RPC 1.1, to require that an attorney's work product be accurate. Filing a brief with unverified AI-generated citations, without conducting any check, is difficult to defend as competent work under RPC 1.1 even if no specific rule violation is charged. The wiser position is that a supervising attorney should be able to vouch for every citation in a filed brief. This note is general information, not legal advice; consult the current Nevada RPC and relevant bar guidance before relying on any specific interpretation.
Is a citation to a dissent always wrong?
No. Citing a dissenting opinion is appropriate when it is presented as such: as the view of the dissenting justices, not as the holding of the court. The problem arises when dissent language appears in a brief as if it were the majority's rule. If the brief says "the court held" and quotes a dissent, that is a misrepresentation. If the brief says "the dissenting justice noted" and quotes the same language, it is accurate. The fix is attribution, not deletion.
What does NRAP require for the Table of Authorities?
NRAP 28(a)(12) requires that an opening brief include a table of authorities listing, with references to the pages of the brief, each case, statute, rule, or other authority cited. Every authority that appears in the brief must appear in the table, and every page on which it appears must be listed. Confirm the current rule text before relying on any specific requirement; NRAP is amended from time to time.
Can the pre-filing check catch every citation problem?
No. The gate checks the four mechanical failure modes it is designed to detect: misquotes, dissent-cited-as-holding, unverifiable citations, and TOA omissions. It does not make substantive legal judgments about whether a cited case is on point, whether an argument is legally sound, or whether the brief is persuasive. Substantive legal review remains the attorney's responsibility. The check is a floor, not a ceiling.
What happens if the attorney overrides a flag?
The attorney reviews the flag, makes a judgment, and records the override with a reason. The override is logged. The brief proceeds with the attorney having explicitly considered and resolved each flagged issue. The gate is not a gatekeeper; it is a checklist. The attorney retains full professional responsibility for the filed document.
A note on accuracy
This guide is general information for Nevada legal professionals. It is not legal advice, it does not create an attorney-client relationship, and it is not a substitute for the rules themselves. Nevada RPC 3.3, NRAP 28, and Nevada State Bar ethics guidance are amended from time to time; confirm the current text before relying on any specific requirement described here. The Stanford RegLab citation above refers to a 2024 study of cloud-based legal research tools; the study's findings are described as published and do not necessarily reflect the current accuracy of any specific product.