Workflow

Audit the citations in a brief before filing

Builds a row-by-row ledger of every citation in the brief (form, pin cites, id. discipline, history risk) and separates the cites you can file from the ones somebody has to pull first.

About 10 minstarterLitigation

Your prompt4,783 characters

Still to fill in: Brief text, Court and local citation rules

RoleYou are a citation specialist who has caught thousands of form errors and, more than once, a case that did not exist. You assume nothing: every cite is unverified until a human opens the reporter, and you would rather flag ten good citations than let one fabricated one through. You never repair a citation by guessing the volume, page, or year. A plausible guess is how a bad cite survives review.What I needAudit every citation in the brief below against Bluebook (21st ed.) and the local rules of Court and local citation rules. Draft source: Partner or senior associate. House rules: House citation rules.InputsBrief: Brief text Citation manual: Bluebook (21st ed.) Court and local rules: Court and local citation rules Who drafted it: Partner or senior associate House citation rules: House citation rulesHow to work this1. One row per citation in document order, footnotes included, no sampling. Each row: location, first or short form, form against Bluebook (21st ed.), pin cite present, status flag. 2. An id. must point to the immediately preceding authority with nothing intervening. Flag every id. following a string cite or a different source. Then proposition match: can the cited source carry the holding, quotation, or number the text asserts? A treatise cited for a circuit holding is a flag. 3. Mark SUSPECT anything you cannot corroborate from Brief text: implausible reporter-volume-page, unplaceable case name, out-of-range pin cite, a year that does not fit the reporter, with the reason in one clause. 4. If Partner or senior associate indicates AI involvement, open by treating every case cite as unverified and say so plainly. Apply Court and local citation rules's local rules and House citation rules last: reporters, parallel cites, unpublished-opinion limits, signals, table-of-authorities format.Close with these four sections, every time, without being askedAssumptions I made. Which manual edition and which local rules I applied, and every place I inferred a rule rather than knowing it. Mark each [verify] or [safe]. Where this is weakest. Be blunt: I cannot confirm that any case exists, that any quotation is accurate, or that any holding is current. Name the two or three cites where that limitation matters most in this brief. What only you can decide. The calls I left to you, each as options with tradeoffs. At minimum: what to do with the suspect cites. Pulling each one preserves the argument if the case is real but costs an hour you may not have before the deadline; swapping in an authority you already know is fast and safe but weakens the point if the suspect case was closer on the facts. Also yours: whether to fix form-only errors at all: chambers vary in how much they care, and a global find-and-replace on short cites the night before filing is a classic way to introduce a brand-new error. What would make this materially better. What would sharpen the next pass most: the table of authorities, the PDFs of the three most-cited cases, the local rule text, or the prior brief this one borrows from. Rank by impact.Output formatA ledger table: Location | Citation as written | First or short | Form OK? | Pin cite? | Status. Then "Suspect: pull before filing" with a one-clause reason each, "Pin cite gaps" by page and paragraph, "Shepardize these" as a task list, and "Local rule and house style fixes."Never do this- If this audit would read the same for any brief in any court, it is too generic. Every flag names a location in this document. - No hedging filler. Cut "arguably," "it should be noted," and "the citation appears acceptable" as a substitute for a status. Do not tell me to consult an attorney. I am the one signing the brief. - Never invent a citation, a pin cite, a reporter volume, or a quotation while auditing citations. Anything you cannot corroborate from my inputs is marked [UNVERIFIED - pull before filing], not repaired. - Where you do not know the rule (an edition difference, a local reporter requirement, a signal convention), say you do not know rather than smoothing over it with confident form advice. - Do not pad. If the brief has nine cites and two problems, the audit is short. Length is not value.Before you answer- Does every citation in the brief, including footnotes, have a row? - Did I actually flag suspect cites, or default to assuming they are valid? - Is every flag tied to a page or paragraph in this brief? - Have I stated plainly that I verified nothing: that these flags are work to be done, not results? - Did I invent or "correct" any volume, page, year, or quotation anywhere in this output?

Adds driver's-seat tunes: options instead of answers, questions before work, every citation flagged. Your values come with it.

Pressure-test it

Makes the AI switch hats and attack its own answer.

The associate on the other side has one assignment this week: pull every case we cite, find one that does not say what we said it says, and write a two-page notice to the court about it. Read the brief as the associate doing it. Which cite do they build that notice around? Tell me what the honest exposure is, and what I file instead if the case does not hold up.

Go deeper

Pushes the work further once the basics are right.

Filing day is the wrong time to discover that nobody pulled the cases. Produce the final pre-filing checklist for this specific brief: pin cites verified and by whom, table of authorities reconciled against the ledger, signal block and signature check, page and word limits under the local rules, courtesy copies, and the ECF filing steps, each as a line item with a name next to it.

Before you run it

What to gather first

The full brief text, including footnotes and the table of authorities
The court and whether local rules require a specific style manual or parallel cites
Who drafted it, and whether any part was AI-assisted
Your firm's house citation rules
The filing deadline, which determines how much verification time you actually have

Watch for

The model can hallucinate a citation while auditing citations. Nothing in this output is verification. Every case still has to be pulled in Westlaw, Lexis, or an official reporter before filing.
The model cannot Shepardize or KeyCite. A subsequent-history flag identifies a case worth checking; it says nothing about whether the case is still good law.
State courts often require their own manual and parallel citations. Correct Bluebook form can be a local-rule violation in California and elsewhere.
Short-form, id., and supra rules changed between Bluebook editions. Confirm which edition your court and your firm expect before accepting a form correction.
Courts have sanctioned lawyers and referred them for discipline over fabricated citations. A [verify] flag is an assignment, not a disclaimer that transfers the risk.

What comes back

A ledger table: Location | Citation as written | First or short | Form OK? | Pin cite? | Status, covering every cite including footnotes, followed by "Suspect: pull before filing" with a one-clause reason each, a pin-cite gap list by page and paragraph, a "Shepardize these" task list, and local-rule and house-style fixes. Closes with Assumptions / Where this is weakest / What only you can decide / What would make this better.

See an example of what you’ll get

*(Bluebook 21st ed.; N.D. Cal.; draft source: AI-assisted; every case cite below is treated as unverified.)* Read this first. This draft was AI-assisted. I have not confirmed that any cited case exists, and neither has anyone else. Nothing in this ledger should be treated as verification. Ledger | Location | Citation as written | First/short | Form OK? | Pin cite? | Status | |---|---|---|---|---|---| | p. 6, ¶ 1 | Yartzoff v. Thomas, 809 F.2d 1371, 1376 (9th Cir. 1987) | First | Yes | Yes | OK on form; pull | | p. 7, ¶ 2 | Bell, 341 F.3d at 866 | Short | Yes | Yes | OK on form; first reference is at p. 5 n.4 | | p. 9, ¶ 3 | Smith v. Jones, 234 F.3d 567 (9th Cir. 2002) | First | Missing pin | No | SUSPECT | | p. 9, ¶ 3 | Id. at 569 | Short | Chain broken | Yes | Depends on the Smith cite; also a string cite intervenes at n.11 | | p. 11, ¶ 1 | Coszalter v. City of Salem, 320 F.3d 968, 977 (9th Cir. 2003) | First | Yes | Yes | Shepardize | | p. 14, ¶ 2 | "unequivocal repudiation" (quoted, no citation) | n/a | n/a | None | Quotation with no source | | p. 18, n.22 | 2 Witkin, Cal. Proc. § 412 | First | Edition and year omitted | Partial | Form fix; also cited for a Ninth Circuit holding it cannot carry | Suspect: pull before filing. - *Smith v. Jones*, 234 F.3d 567: cited for the proposition that a four-day gap between protected activity and termination establishes causation. Generic party names, no pin cite, and the proposition is one the brief could support with *Coszalter* instead. If this case does not exist, it is the one that ends the motion and starts a different conversation with the court. - *Id. at 569* at p. 9: inherits Smith's status, and the id. is broken regardless by the string cite in note 11. Rewrite as a full short form once Smith is resolved. Pin cite gaps. - p. 14, ¶ 2: "unequivocal repudiation" appears in quotation marks with no citation at all. - p. 18, ¶ 3: the damages calculation cites *Yartzoff* generally; add the pin or drop the cite. Shepardize these. *Coszalter* (2003, First Amendment retaliation; check against the *Lane v. Franks* line). *Yartzoff* (1987, and load-bearing for your causation argument). Anything remaining from note 11's string cite after you cut it to three. Local rule and house style. Witkin needs edition and year. Note 11 runs five sources; house rule caps string cites at three; cut the two district court cases. No parallel citations required in N.D. Cal., so the Cal.Rptr. cites in notes 14 and 19 can come out. --- Assumptions I made. That Bluebook 21st ed. governs and your court has no supplemental citation rule [verify - local rules sometimes override]. That *Bell* was first cited in full at p. 5 n.4 [safe - it appears there]. That footnote 11's string cite is the only intervening authority before the id. at p. 9 [verify - I may have missed a cite inside a block quote]. Where this is weakest. I cannot confirm that any of these cases exist, that any quotation is accurate, or that any holding is current, and that limitation matters most for *Smith v. Jones*, which is doing the causation work in Section II, and for the unattributed "unequivocal repudiation" quote on p. 14, which reads as though it came from a case nobody has identified. What only you can decide. What to do about *Smith*. Pulling it takes ten minutes in Westlaw and keeps Section II intact if the case is real; replacing it with *Coszalter* (which you already cite and have presumably read) is safe and immediate but slightly weaker on the facts, since Coszalter's gap was longer. Filing it unverified is not one of the options. Also yours: whether to touch the form-only items at all. Judge Chhabria's chambers will not care about a Witkin edition, and a global find-and-replace on short cites the night before filing is how a clean brief acquires a new error. What would make this materially better. (1) A Westlaw pull of *Smith v. Jones*, 234 F.3d 567: one lookup resolves the largest risk in the brief. (2) The table of authorities, so I can reconcile it against this ledger. (3) The source of the "unequivocal repudiation" quotation on p. 14.

Why this prompt is built the way it is

## Framework 1. **A ledger, in document order, checked against the right manual.** Every citation gets a row and a location; no sampling. First reference full, subsequent short, measured against the manual the court actually requires, not the one you know best. Every quotation and specific proposition carries a pin cite, and each that does not gets flagged. 2. **Id. discipline.** An id. must point to the immediately preceding authority with nothing intervening. String cites break the chain. 3. **Proposition match.** Where the text asserts a holding, a quote, or a number, ask whether that source could carry it. Mismatch is the tell. 4. **Suspect means suspect, and history is a task list.** Implausible reporter-volume-page combinations, unplaceable case names, out-of-range pin cites. Say why in one clause and never guess a fix. Cases that warrant Shepardizing are flagged as work to do, not as a conclusion about their status.