Overview

Citation checking is the systematic verification that every legal authority cited in a document actually exists, is accurately cited, is accurately quoted, and remains good law. A single bad citation can undermine the credibility of an entire brief. In the age of AI-assisted legal research, citation checking has become even more critical due to the well-documented problem of AI-generated “hallucinated” citations.

The Citation Checking Workflow

Phase 1: Existence Verification

For every citation in the document:

1. Search by case name and citation in a legal database (Westlaw, Lexis+, Bloomberg Law, or Google Scholar)
2. If found: Proceed to Phase 2
3. If not found by citation: Search by case name only — the citation may have a typo
4. If not found by name or citation: The case likely does not exist. Flag as RED — potential fabrication
5. For statutes: Verify the section exists in the current U.S. Code (uscode.house.gov or law.cornell.edu)
6. For regulations: Verify the section exists in the current CFR (ecfr.gov)

Phase 2: Citation Format Verification

Check every citation against the Bluebook (21st edition) or the applicable court’s local rules:

Phase 3: Pinpoint Verification

For every pinpoint citation (specific page or paragraph reference):

1. Navigate to the exact page cited
2. Confirm the cited proposition appears on that page
3. If the proposition appears on a different page: Correct the pinpoint
4. If the proposition does not appear in the case at all: Flag as RED — the case does not support the attributed proposition

Phase 4: Quotation Verification

For every direct quotation:

1. Compare word-for-word against the original source
2. Check for:

• Omissions: Are ellipses (…) used where text has been removed?
• Alterations: Are brackets [ ] used where text has been changed?
• Emphasis: Is “[emphasis added]” noted if emphasis was added? Is “[emphasis in original]” noted if emphasis existed in the original?
• Capitalization: If the first word of the quotation was not capitalized in the original, is a bracket used? [L]ike this

1. Flag any discrepancy — even a single changed word

Phase 5: Authority Verification (Citator Check)

For every case citation:

1. Run the case through a citator (KeyCite, Shepard’s, or BCITE)
2. Check for negative treatment:

• Overruled / Reversed → RED — cannot be cited as authority for that point
• Questioned / Criticized → YELLOW — investigate whether the negative treatment affects the specific proposition cited
• Distinguished → Usually acceptable, but verify the distinction does not undermine the citation

1. Check for subsequent history: Was the case affirmed, reversed, or vacated on appeal?
2. Note the treatment in your citation checking report

Phase 6: Statutory and Regulatory Currency

For every statute and regulation cited:

1. Check for amendments — Has the statute been amended since the document was drafted?
2. Check for repeal — Has the statute been repealed?
3. Check for renumbering — Has the section been moved to a different part of the code?
4. For regulations: Has the regulation been amended, revoked, or replaced by a new rule?
5. Note effective dates — Some changes have future effective dates

AI-Generated Citation Risks

The Problem

Large language models (including ChatGPT, Claude, and others) are known to generate plausible but nonexistent legal citations. These “hallucinated” citations include:

• Fabricated case names that sound real
• Fake reporter citations with realistic volume and page numbers
• Fabricated holdings that perfectly support the argument
• Real case names paired with wrong citations, wrong courts, or wrong holdings

Why It Happens

LLMs generate text by predicting the most likely next token based on patterns in training data. Legal citations follow predictable patterns (Party v. Party, Volume Reporter Page), so the model generates realistic-looking but fictional citations.

Documented Consequences

• Mata v. Avianca, Inc. (S.D.N.Y. 2023): Attorney sanctioned $5,000 for filing brief with six AI-generated fabricated citations. The attorney’s co-counsel was also sanctioned. The judge’s opinion specifically addressed the dangers of AI-generated legal research.
• Multiple subsequent incidents across federal and state courts
• Several courts have adopted local rules requiring disclosure of AI assistance and verification of AI-generated citations

Prevention Protocol

1. Verify every citation independently. Never trust an AI-generated citation
2. Search the exact case name and citation in a legal database
3. If the case exists, verify the holding matches what the AI attributed to it
4. If the case does not exist, it was fabricated. Remove it
5. Do not attempt to “find” a substitute case that supports the proposition — research the issue properly from scratch
6. Disclose AI assistance if required by court rules
7. Run all AI-assisted work through the full citation checking protocol before submission

Citation Checking Tools

Citation Checking Report Format

For each citation checked, record:

Standards by Document Type

Common Bluebook Errors