Verification as a Craft: Existence Is Not Enough
Why it Matters
A citation in a brief answers five questions, whether or not anyone asks them. Does the case exist, did it decide what the brief says, does the cited page say it, are the quoted words the court’s own, and is it still good law? A database lookup settles the first question in seconds; the other four fail quietly, inside real cases, and someone has to open the opinion to catch them. Courts sanction lawyers for those quiet failures too: a real case cited for a result it never reached, or a quotation borrowed from another opinion. Your job here is to grade three machines’ cite-checks of a brief that drew sanctions, then check the same five citations yourself, reading every cited page and timing every check.
Current Context
On September 16, 2026, in Lisandrillo v. Palozzi, Florida’s Fourth District Court of Appeal said the trouble with a lawyer’s filings (opens in a new tab) “has nothing to do with hallucinated law or false legal citations.” The court suspected an AI tool had a hand in them. It faulted other failures: allegations the hearing transcript contradicted, a citation to an unpublished order with no value as precedent, and scattershot arguments. In May 2026, Princeton researchers reported that an AI agent able to search legal databases caught more than four in five planted citation errors (opens in a new tab), every fake case among them, but stumbled on wrong page numbers. Neither the problems the court found nor the ones the machine missed would show up in a check that stopped at whether each case exists.
Key Concepts
Existence
Whether the case is real: decided by the court the citation names and printed at the volume and page it gives. An invented citation usually has a plausible name and a correct format, and its volume and page may lead to a real case with a different name, so the citation’s look proves nothing. A citation you cannot find is unverified, which is not the same as fake: databases have gaps, so a search by citation, by name, and by any quoted words comes before that conclusion.
Holding Accuracy
Whether the case decided what the brief says it decided. A real case can be cited for a result it rejected or for a question it never reached, and nothing in the citation itself shows it. The check is the opinion’s disposition (affirmed, reversed, dismissed) read with the reasoning behind it.
Pinpoint Support
Whether the page the citation names, called the pinpoint or pincite, holds the words or reasoning that support the sentence. A proposition supported somewhere else in the opinion still carries a wrong pinpoint, and a reader sent to the wrong page cannot check the claim. Checking it takes a copy that marks the reporter’s page numbers, as Google Scholar and Westlaw do for most opinions.
Quotation Fidelity
Whether every word inside quotation marks appears in the cited opinion, in that order. A quotation can be invented outright, or real but taken from a different case, and the two look identical on the page until someone searches the opinion for the exact words.
Subsequent History
The later life of a case: what happened to it on appeal, and whether a higher court, a statute, or a rule has since displaced it. A case can be real and quoted perfectly and still not be good law. A citator such as KeyCite in Westlaw flags that later history; in Google Scholar, the list of opinions citing the case is a slower check that can miss a recent change.
Automated Citation Check
A tool’s check of a citation, from a database lookup of the reporter citation to an AI agent that searches for the opinion and reads it. A lookup can show that a citation leads nowhere or to a different case, but the other questions need the opinion’s text, and free databases often lack the reporter’s page numbers. A tool’s “confirmed” or “fake” is a lead to follow, since tools also flag real cases they fail to find.
Verification Cost
The minutes a proper check takes, citation by citation. Knowing the number lets a paralegal tell a supervising lawyer honestly how long a cite-check will take, and why forty citations cannot be checked in the ten minutes before a deadline. A skipped check moves its cost to others: the court that must check every citation itself, the opposing party, and the client whose filings are now read with suspicion.
Resources
- Citing Cases from Westlaw, Citation Literacy series, UNC School of Law (opens in a new tab) (The video at the top of this page. Watch the first seven minutes, where a law professor shows where the court’s own words begin below the publisher’s headnotes and how the star paging tells you the page a sentence is on. The screens are Westlaw’s from 2022, and Google Scholar’s *587 markers work the same way.)
- Joint Opposition to Second Motion to Dismiss, Woodward Harbor, L.L.C. v. City of Mandeville, No. 2:23-cv-05824 (E.D. La. June 3, 2025), ECF No. 97 (opens in a new tab) (The brief you check, filed by licensed lawyers in a Louisiana federal court and later the subject of a sanctions order. The five citations in Part 1 are on pages 6 through 12.)
- Order and Reasons, Woodward Harbor, L.L.C. v. City of Mandeville, No. 2:23-cv-05824 (E.D. La. Feb. 5, 2026), ECF No. 101 (opens in a new tab) (The ruling on the motion the brief opposed. Pages 20 through 22 give the court’s own account of the brief’s citations; read them only after your log is finished.)
- Google Scholar (opens in a new tab) (Choose Case law, then search by citation or by case name. Most opinions mark the reporter’s page numbers in the text, such as *587, which is how you check a pinpoint.)
- Westlaw through the Daytona State College Library (opens in a new tab) (Sign in with your college account. The KeyCite flag beside a case shows whether a later court has reversed, overruled, or questioned it. See whether the library’s Westlaw offers a citation checker you can use for the third report in Part 2.)
- Russell v. Mells, 426 So. 3d 913 (Fla. 2d DCA 2025) (opens in a new tab) (Florida’s Second District referred a lawyer to The Florida Bar over an answer brief with three case citations: one case did not exist, and the other two were misquoted. Read Part IV, pages 7 through 12.)
- Noland v. Land of the Free, L.P., 336 Cal. Rptr. 3d 897 (Ct. App. 2025) (opens in a new tab) (A California appellate court found that 21 of the 23 quotations in one opening brief were fabricated, most of them attributed to real cases, and sanctioned the lawyer $10,000. Read pages 2 and 3, where it states its rule, and pages 9 and 10, where it walks through the false quotations.)
- Whiting v. City of Athens, Nos. 24-5918, 24-5919, 25-5424 (6th Cir. Mar. 13, 2026) (opens in a new tab) (The Sixth Circuit’s sanctions opinion: more than two dozen fake citations and misrepresentations across three appeals, $15,000 from each of two lawyers, and the matter sent to the chief judge to consider discipline. Read pages 1 through 3 and page 18, the first page of the appendix, which reads like a cite-checker’s log.)
- Federal Rule of Civil Procedure 11, Legal Information Institute (opens in a new tab) (The rule under which the Woodward Harbor court sanctioned the lawyer who prepared the brief. Read subdivision (b)(2): by signing, a lawyer certifies that the filing’s legal contentions are warranted by existing law or by a nonfrivolous argument for changing it.)
- Florida Rules of General Practice and Judicial Administration, The Florida Bar (opens in a new tab) (Open Chapter 2 and read Rule 2.515(d)(2), the Florida counterpart: every signer of a Florida court filing represents that the legal authorities it cites exist and are accurately cited.)
What to Do
The same five citations get checked four times here: by two chatbots, by a tool that can look the cases up, and by you, by hand. By the end you can answer all five questions a citation must answer, and you know which of them a machine can answer for you. Your worksheet is what you post: the verification log, your scores and ranking of the three machine reports with a short comparison, and one repaired paragraph.
Part 1: Read the Standard and the Brief
Start with the background reading and the first seven minutes of the video at the top of this page. The video shows how to tell the court’s words from the publisher’s, and how to find the page a sentence is on. Then read Rule 11(b)(2), the federal rule the brief’s signers were held to, and Rule 2.515(d)(2), its Florida counterpart. Read the pages of Russell v. Mells, Noland, and Whiting that Resources names. Last, open the Woodward Harbor brief and read pages 6 through 12, where the five citations below appear. The court that decided the motion found the brief full of citations to cases that did not exist or were quoted or described wrongly. Some of the five are among them, and some are sound.
The five citations. Each is quoted from the brief with the sentence it supports. Copy them into your log exactly as written.
Citation 1 (page 6), for the sentence “To survive a Rule 12(b)(6) motion, Plaintiffs need only allege facts that, if true, state a plausible claim for relief”: Bell Atl. Corp. v. Twombly, 550 U.S. 544, 570 (2007).
Citation 2 (page 7): “Likewise, in Wood v. City of Utica, 890 F.3d 273, 285 (2d Cir. 2018), the Second Circuit reversed dismissal of an equal protection claim where the plaintiff alleged the city singled out his property for enforcement action that it did not apply to others. The court emphasized that ‘a plaintiff is not obligated to identify actual instances where others were treated differently at the pleading stage.’”
Citation 3 (pages 7 and 8), for the sentence “To state a ‘class-of-one’ equal protection claim, a plaintiff must show that they were intentionally treated differently from others similarly situated and that there is no rational basis for the difference”: Village of Willowbrook v. Olech, 528 U.S. 562, 564 (2000).
Citation 4 (page 11), for the sentence “However, courts routinely allow class-of-one equal protection claims in the context of land use”: Also see Integrity Collision Ctr v. City of Fulshear, 837 F. 3d 581, 587-88 (5th Cir. 2016) (zoning reclassification of business supported class-of-one claim).
Citation 5 (pages 11 and 12), for the sentence “The Fifth Circuit mandates liberal allowance of amendments”: See Lowrey v. Texas A&M Univ. Sys., 117 F.3d 242, 245 (5th Cir. 1997) (“Rule 15(a) evinces a bias in favor of granting leave to amend.”).
Part 2: Run Three Cite-Checks
Run the Citation Screener three times on the same text: the five citations, each with its sentence, pasted where the prompt expects a brief. Use two different general-purpose chatbots for the first two runs, each in a fresh chat, with web search turned off if the chatbot lets you. For the third run, use a tool that can look cases up: the citation checker in the library’s Westlaw, if the library offers one. Otherwise, use a chatbot’s web search or research mode, and before you send the Screener, delete its sentence that begins “You do not have access.” Save all three reports. Time each run from the moment you send the prompt until you finish reading the answer.
Part 3: Score the Reports Before You Check
Score each report against the five questions, one line each, before you check anything yourself. Does the case exist, did it decide what the brief says, does the cited page support the sentence, are any quoted words the court’s own, and is it still good law? On each line, give the report one point for every citation it answered that question for, with a reason you could check, for a total out of 25. Then rank the three reports, and write two or three sentences on which one you would hand to a supervising lawyer and why. Keep these first scores as they are; they record how the reports look before anyone checks them.
Part 4: Check the Five by Hand
Now answer the five questions yourself in Google Scholar or the library’s Westlaw, one citation at a time, and time each citation from start to finish. Build your log as a table with one row per citation, one column for each question, and a last column for minutes. Every cell gives your answer and the source it came from, such as the link and the page. For existence, record the search that found the case, or every search that failed. For the holding, record the disposition and one sentence on what the court actually decided. For the pinpoint, say what the cited page says. For a quotation, say whether a search of the opinion found the exact words. For history, record what KeyCite or the list of citing opinions showed. Write “no quotation” or “no case to check” where a question does not apply.
Search for a case you cannot find by its citation, by its name, and by any words it supposedly quotes. If a search turns up a case with a similar name, compare the court, reporter, year, and subject before you treat it as the same case. When the log is done, score it against the same five lines before you open the Error Taxonomist: a cell counts only if it names the source it came from. Then open a fresh chat, run the Error Taxonomist, and paste your log. Where its category differs from yours, go back to the opinion and decide which is right.
Part 5: Score Again and Repair a Paragraph
Score the three reports a second time, now against your log. On each line, a report earns a point only when its answer matches what you found. Where your log says there was no case to check, the report earns the point only if it did not vouch for the case. Mark every answer a report gave with confidence that your log shows was wrong. Then choose one paragraph of the brief that holds a citation your log shows is flawed. Rewrite it so that every citation in it is one you verified, cited only for what the source supports, at the right page. Check each citation in your paragraph against your log first, then run the Rewrite Checker with the paragraph and the text of each case it cites, and fix anything it finds. Last, read pages 20 through 22 of the court’s order and compare its account with your log.
Part 6: Post Your Worksheet
Create a single new discussion thread with your worksheet. It holds your log as a table, a table of each report’s first and second scores on each line, your repaired paragraph, and a comparison of about 250 words. The comparison names the report you ranked first before checking and the one you rank first now. It then says what each method caught and missed, question by question, where a report was confidently wrong, and how your minutes compare with the machines’. End it with a sentence on whether your log agrees with the court’s account.
Part 7: Share Chat Link
Include one AI chat link with a 1–2 sentence explanation of what the conversation shows and why you chose to share it.
Suggested AI Prompts
Use these prompts as a starting point, then adjust them to fit your goal. Strong prompting develops through trial, revision, and testing. It’s a foundational skill that grows into more advanced AI work such as context engineering and agent-based workflows.
Citation Screener
Act as a cite-checking assistant. I will upload or paste a legal brief. List every citation to a case, statute, or rule in it, and for each one tell me whether you can confirm it exists, whether the proposition it is cited for is one the authority plausibly supports, and whether any quotation attributed to it raises a concern. Label each citation as confirmed, suspicious, or unable to verify, and explain your reasoning in one sentence. You do not have access to a case law database, so do not claim to have confirmed a case unless you are certain; unable to verify is an acceptable answer and a better one than a guess.
Naming what the model cannot do, and accepting “unable to verify” as an answer, gives it permission to abstain, which lowers the odds that it confirms a fake case as confidently as a real one. Notice that the prompt never asks about the cited page or later history.
Error Taxonomist
Act as a legal research supervisor. I will paste my verification log for a sanctioned brief, listing each citation and what I found when I checked it. Classify every problem as a fabrication (the authority does not exist), a holding misstatement (a real case described wrongly), a pinpoint failure (the cited page does not support the proposition), a false quotation, a subsequent history problem (overruled, reversed, or criticized), or a citation form error. For each, explain in one sentence why it belongs in that category, and flag any entry where my own finding is ambiguous enough that you cannot classify it. Wait for me to paste the log.
The log is yours and the model only sorts it: because you supply what each check found, it classifies evidence instead of generating it, and its categories give you the vocabulary for your comparison.
Rewrite Checker
Act as a strict appellate editor. I will paste a rewritten paragraph from a brief along with the full text of each case I cited in it. For every citation, quote the exact language from the pasted opinion that supports the proposition and give the page it appears on, or state that the opinion does not support the proposition. Confirm that every quotation in my paragraph appears verbatim in the pasted text. Use only the opinions I provide; do not add authority or rely on your memory of the law.
Pasting the opinions confines the check to supplied text and turns it into a reading task, which models do more reliably than recalling law. Barring outside authority keeps the model from patching your paragraph with a case you have not read.