"Just Shepardize it" is common advice for catching bad citations, and it is genuinely useful advice — for a problem Shepardizing (or the KeyCite equivalent) was never designed to solve. Shepardizing confirms whether a real, existing case is still good law. It does not confirm the case exists in the first place, and it does not confirm a quotation attributed to that case is accurate.
What Shepardizing actually does
Shepard's Citations (on Lexis) and KeyCite (on Westlaw) track the subsequent history and treatment of a case: whether it has been overruled, reversed, criticized, or followed by later courts. This is essential for confirming a real case is still authoritative on the point you're citing it for.
Why this doesn't catch AI hallucinations
If an AI tool fabricates a case that never existed, there is nothing to Shepardize — the citation simply will not appear in the system, or worse, a distracted user might not notice the "no citing references found" result is actually a sign the case itself doesn't exist, rather than just an uncited one. Shepardizing also does nothing to confirm a quotation is accurate; it only speaks to the subsequent treatment of a case that is assumed to be real and correctly cited.
A documented example of the gap
Several cases in the Charlotin AI Hallucination Cases database involve fabricated citations that a Shepardizing-only review would not have caught, because the underlying assumption of Shepardizing — that the case exists and is what it claims to be — was false from the start. In Mata v. Avianca, No. 22-cv-1461 (S.D.N.Y. 2023), the fabricated cases were not caught through Shepardizing; they were caught because opposing counsel could not locate them in any database at all.
What to run instead, or in addition
Existence check first (does the case exist at all), then quote-match (is the quoted language accurate), then Shepardize or KeyCite for good-law status, then proposition-support (does it actually stand for what you're citing it for). Shepardizing is a valuable and necessary step — it is simply the fourth step, not a substitute for the first three.
Expert perspective
Legal research trainers have long taught Shepardizing as the final step in a research chain that starts with confirming a source is real and accurately represented. AI drafting tools have made it more important than ever to keep that ordering explicit, since a fabricated citation can otherwise slip through a workflow built around Shepardizing alone.
A common question
If a citation doesn't show up in Shepard's or KeyCite at all, does that prove it's fake?
Not necessarily on its own — a very recent or obscure real case might have limited citing history. But combined with a failed existence check against a primary source like CourtListener, it is a strong signal worth investigating immediately rather than assuming it's simply an under-cited case.
Related reading
- Existence, Quote, and Proposition Checks: What Each One Actually Catches
- How CourtListener Works for Lawyers
- Bluebook Citation Checking With AI Tools
- How to Check for AI Hallucinations in Legal Briefs
- Mata v. Avianca: The Lessons Two Years Later
Check a brief before you file it →
What Shepard's and KeyCite were actually built to solve
Both systems predate AI drafting tools by decades and were designed to solve a narrower, older problem: tracking whether a case has been affirmed, reversed, overruled, criticized, or distinguished by later courts, so a researcher relying on it knows whether it remains good law. That is a genuinely valuable and still-necessary function. It was simply never designed to answer the question "does this case exist" or "is this quotation accurate," because those questions were not the failure modes researchers using traditional legal research methods typically encountered.
A concrete scenario showing the gap in action
Imagine an AI tool generates a citation to a real, existing case, but attributes to it a holding from a completely different, unrelated case on a similar topic — a plausible mix-up given how language models generate statistically associated text. Shepardizing the cited case will show accurate treatment history for that case as it actually exists, giving a false sense of confidence, while saying nothing about whether the specific proposition attributed to it in your brief is actually supported by that case's real holding. This is a proposition-support failure, not a good-law-status failure, and Shepardizing simply operates on a different axis entirely.
Practical sequencing for a verification-conscious researcher
Run existence and quote-match checks first using a primary source, since there is no point Shepardizing a citation that turns out not to exist. Once existence and quote accuracy are confirmed, Shepardize or KeyCite the case for good-law status. Finally, read the actual passage to confirm proposition-support. This ordering front-loads the checks most likely to reveal a fabrication early, before investing time in the more traditional research steps.
Final takeaway
Keep using Shepard's or KeyCite exactly as you always have — nothing about AI drafting tools changes their value for confirming good-law status. Just don't mistake a clean Shepard's report for proof that the underlying citation was ever real or accurately used in the first place.
A common question
If Shepard's shows no negative treatment at all, does that mean the case is safe to rely on?
It means the case has not been overruled or criticized, which is valuable but incomplete information. It says nothing about whether the case actually exists in the first place (if you have not independently confirmed that) or whether the specific quotation or proposition attributed to it in your draft is accurate. Treat a clean Shepard's report as one confirmed data point among the four checks that together constitute full verification, not as a substitute for the other three.
Related reading
Build the quote-fragment search habit into your standard proofreading pass, right alongside checking for typos, so it happens by default rather than requiring a separate deliberate decision every time.
A one-page reminder taped near your desk, or a recurring line item on your filing checklist, is often enough to make this habit stick across a busy practice.