How to Verify AI Legal Citations Before You File (ABA 512 Checklist)

To verify AI legal citations before you file, run every cite through three checks. First, confirm the case exists by opening it in a primary source, not by asking the AI that produced it. Second, confirm the quoted language actually appears in that opinion. Third, confirm the case supports the proposition it is cited for.

ABA Formal Opinion 512 says relying on or submitting a tool's output without "an appropriate degree of independent verification or review" could violate the duty of competence, and that output used in court must be "carefully reviewed to ensure that the assertions made to the court are not false." The checklist below turns that into a routine.

Reviewed and corrected on 12 October 2026. This is general information, not legal advice; check the cited opinion or rule before relying on it.

The case that should change how you check cites

On 13 March 2026 the Sixth Circuit decided Whiting v. City of Athens (Nos. 24-5918/5919 and 25-5424). It sanctioned two Tennessee lawyers, Van R. Irion and Russ Egli, $15,000 each as punitive sanctions. It also ordered them to reimburse the appellees' fees in full and to pay double costs (opinion).

The court said it found "over two dozen fake citations and misrepresentations of fact" in the briefs. Many of the cited cases "either did not exist, did not include the quoted language he claimed they did, or did not discuss or support the proposition for which Whiting cited them." That is three distinct failure modes, and only one of them, the fabricated case, is the kind your gut already knows to look for.

The court asked counsel whether generative AI was used to write the briefs. They did not answer, so the opinion does not decide it. The rule it relied on is broader. Quoting a California appellate court, it said that no filing should contain citations a lawyer has not personally "read and verified," whether they came from generative AI or any other source. Whether a chatbot, a junior associate or tired memory produced the cite, the signing lawyer owns it.

The dangerous failures are the real citation with the wrong quote, and the real citation that does not stand for what you claimed. Those survive the one check most people run, which is "does this case exist."

This post is the narrow, practical piece: the task you run before the brief leaves your hands. The backstory from Mata v. Avianca onward is in our AI hallucinations and sanctions guide.

TL;DR

• A "does the case exist" check is not enough. In Whiting the court listed three failure modes: a fake case, a missing quotation and a citation that does not support the point.
• In Farris (6th Cir., 3 April 2026) none of the cases was fake, but the briefs carried invented quotations and misdescribed the holdings of real cases. The court denied counsel his fee for the appeal and referred the matter for possible discipline.
• ABA Formal Opinion 512 (29 July 2024) does not give a fixed verification standard. It says the level of review "will necessarily depend on the GAI tool and the specific task that it performs."
• A tool built for lawyers does not transfer the duty. In Farris, counsel had used Westlaw's CoCounsel, and the court said the risk persists "even when new tools are sponsored by trusted legal technology providers."
• Charlotin's AI Hallucination Cases database listed 2,184 decisions worldwide, 1,492 in the US, when last updated on 10 October 2026. See our read of it.

Part of our legal AI verification and hallucination guide series.

For related coverage, see Damien Charlotin's Hallucination Tracker, Read Like a Risk Manager and the AI hallucination sanctions tracker.

Why the old checklist fails

For two years the standard advice was simple: pull every cite in Westlaw or Lexis and confirm it is real. That advice was built for the Mata v. Avianca era. In Mata (S.D.N.Y., 22 June 2023) ChatGPT produced non-existent opinions, including the "Varghese" decision, and the court ordered letters to six judges falsely named as their authors. A database search catches those.

The harder cases are half right. A real case name, a real reporter citation and a quoted holding the opinion never contained. Or a real case that says something adjacent. The Stanford researchers who tested legal research tools count both kinds as hallucinations: a response that is false, and one that cites a real source that does not support the claim. In their 2024 testing, tools from LexisNexis and Thomson Reuters "each hallucinate between 17% and 33% of the time" (Magesh et al., Journal of Empirical Legal Studies, 2025). Those were 2024 versions, so current tools may differ.

So the existence check is necessary and not enough. If your protocol stops at "the case is real," a brief like the one in Whiting can pass straight through.

The pre-filing checklist, mapped to ABA 512

ABA 512 ties generative AI use to existing duties. Under Rule 1.1, relying on a tool's output "without an appropriate degree of independent verification or review" could violate the duty of competence. Under Rules 3.1, 3.3 and 8.4(c), output used in court must be "carefully reviewed," and the opinion notes that "even an unintentional misstatement to a court can involve a misrepresentation under Rule 8.4(c)." Run the checklist on every citation, every quotation, every proposition.

1. Existence: catches the fabricated case

Open the actual opinion in a real source. Not a summary and not the AI's restatement. Confirm that the case name, reporter citation, court and year all resolve to one real document. If you cannot open it, treat it as fake.

Do not verify a citation with the AI that produced it. In Mata, ChatGPT assured the lawyer the fake case was real. Take the cite to an independent source. Free options include the Legal Information Institute, govinfo.gov for federal materials, the Supreme Court Case Finder for Supreme Court opinions, Google Scholar and the court's own site. A cite that returns nothing is a red flag.

2. Quote fidelity: catches the real case with a fabricated quote

For every quoted passage, find that exact language in the opinion, in that order, on the page you pin-cited. In Noland (Cal. Ct. App., 12 September 2025), 21 of 23 case quotations in the opening brief were fabricated, and counsel was ordered to pay $10,000. A quote that is paraphrased, stitched from two paragraphs, or invented is wrong even if the case is genuine. Search the opinion for a distinctive phrase from your quote. If it does not appear verbatim, fix it.

3. Proposition support: catches the real case cited for the wrong point

Read the surrounding holding, not just the quoted line. A case can contain your sentence in dicta, in a dissent, or in a recitation of the losing argument and still not support your proposition. In Farris the real cases were misdescribed. This is the hardest failure to catch because it requires reading. The Ninth Circuit in Lnu v. Blanche wrote that a competent and diligent attorney "must also read and reason," not only detect and eliminate false AI results.

4. Good law: catches overruled or limited authority

Confirm the case is still good law. A well-known example is Loper Bright Enterprises v. Raimondo, 603 U.S. 369 (2024), which overruled Chevron U.S.A. Inc. v. NRDC, 467 U.S. 837 (1984). A model trained on older administrative law may cite Chevron deference as live doctrine. Run a citator such as KeyCite or Shepard's, or another service that shows negative treatment.

5. Pin cite accuracy: catches misplaced page references

Confirm the pin cite points to the page where the proposition appears. A court that flips to your cited page and finds nothing relevant will read that as carelessness at best.

6. Statutory and regulatory text: catches stale or misquoted statutes

If you cite a statute or regulation, pull the current official text. Codes are amended, and a model's training cutoff is not the current code. Confirm the section number, the operative language and the version in effect on the date that matters. For federal materials, the US Code and the eCFR are the starting points.

7. Confidentiality of the inputs: catches a different kind of exposure

This is about how you ran the check. Under ABA 512 and Rule 1.6, many self-learning tools are designed so that output could lead directly or indirectly to disclosure, and informed consent is required before inputting information relating to the representation into such a tool. Merely adding "general, boiler-plate provisions to engagement letters" is not enough. Check a tool's terms before pasting client facts. Our notes on whether ChatGPT is confidential for legal work and where your legal AI data actually goes cover that.

Run every cite top to bottom. A no at any gate stops the cite from reaching the brief.

What a half-right citation looks like

Suppose an AI hands you a case name, a plausible reporter citation and a quotation about what a contractual limitation of liability can and cannot cover. You search the case name and find a real opinion. Step 1 passes.

Now search the opinion for a distinctive string from the quotation. If it returns zero hits, the quotation is invented, no matter how real the citation looks. If the string does appear, read the paragraph around it. The sentence might be the court describing the losing party's argument. That is step 2 and step 3, and they take a few minutes per cite once you build the habit.

Making the check hard to skip

A seven-step check on every cite sounds heavy. The steps that get skipped under deadline pressure are the quote and the proposition, which are also the expensive ones to skip.

• Divide the work. One person pulls and opens each opinion. A second person checks each quotation and reads each holding.
• Keep a record. Note which source you opened for each cite. ABA 512 tells lawyers to consult the applicable court's local rules on AI use, and in Lnu the Ninth Circuit ordered the sanctioned firm to include in future filings a statement, under penalty of perjury, on whether generative AI was used. A record of what you checked is your answer if a judge asks.
• Supervise. ABA 512 says managerial lawyers "must establish clear policies regarding the law firm's permissible use of GAI." In Lnu the court sanctioned lawyers whose firm had unlicensed brief writers drafting the legal arguments with no attorney vetting their citations.
• Check local rules. Requirements differ by court and judge. The Fifth Circuit considered and declined to adopt a circuit-wide rule and relies on existing rules. Our state-by-state guide tracks bar opinions and court rules.

A tool that shows the source text beside each claim makes checking faster. It does not replace reading the opinion. Anyone who tells you the duty goes away with a better tool is telling you something the courts have not said.

The shorter version you can tape to your monitor

1. Does this case exist, in a real source I opened myself?
2. Is this quote in that opinion, word for word, on this page?
3. Does this case hold what I said it holds?

Then add three: is it still good law, is the statute text current, and did I run this check somewhere that protects my client's confidences. Six questions, every cite.

FAQ

How do I verify an AI legal citation before filing?

Open the case in a primary source you trust, not the AI that produced it. Confirm the quotation appears word for word on the pin-cited page and that the holding supports your proposition. Then run a citator for negative treatment and keep a record of what you checked.

How can I check if an AI-generated case citation is real for free?

Free sources include the Legal Information Institute at Cornell, govinfo.gov, the Supreme Court Case Finder, Google Scholar and the courts' own websites. Search the case name and reporter cite. If a cite returns nothing in any of them, treat it as fabricated until you find the opinion.

Can ChatGPT verify its own citations?

No. In Mata v. Avianca ChatGPT told the lawyer that the fake case was real and could be found on Westlaw and LexisNexis. Verify against an independent primary source.

What is ABA Formal Opinion 512?

It is the ABA's July 2024 formal opinion applying the Model Rules to generative AI. It covers competence, confidentiality, communication, candor, supervision and fees. It says the amount of verification "will necessarily depend on the GAI tool and the specific task that it performs," and does not set a checklist.

Does a tool built for lawyers remove the need to verify citations?

No. In Farris the Sixth Circuit said the risk persists "even when new tools are sponsored by trusted legal technology providers." Stanford's 2024 testing found tools from LexisNexis and Thomson Reuters each hallucinated between 17% and 33% of the time. Versions change, but the duty to check stays with the lawyer.

How do I check whether a case is still good law?

Run a citator such as KeyCite or Shepard's, which flag negative treatment including reversals and overrulings. Loper Bright Enterprises v. Raimondo, 603 U.S. 369 (2024), overruled Chevron, so a model trained on older law can cite it as live.

What happens if I file a brief with a fake AI citation?

Courts have imposed fines, fee awards, suspensions and referrals. In Whiting, two lawyers were fined $15,000 each plus the other side's fees and double costs. In Couvrette v. Wisnovsky (D. Or.) the orders of 12 December 2025 and 23 March 2026 total $110,204.38 across two lawyers. See the sanctions guide and the Oregon breakdown.

Are ABA Formal Opinions binding on lawyers?

They are persuasive, not binding, but state bars and courts often draw on their reasoning. We break down how much weight Opinion 512 carries in are ABA Formal Opinions binding.

For the US statutes, regulations and court rules these laws sit in, see the US primary law API.

Arshita Anand

Co-Founder & CEO · Attorney

Arshita leads product and strategy at Vaquill AI, the US primary law data API. As an attorney, she decides which law we carry, how each section is sourced and cited, and what legal AI and compliance teams need from the data.