TL;DR: Every citation has to pass two separate tests: is it real, and is it still good law? The first guards against hallucinated or misquoted authority; the second against citing a case that has been reversed, overruled, or superseded. AI helps with both - by grounding answers in retrieved sources with exact-page citations, and by surfacing later treatment - but it does not replace the verification. This guide explains the two questions, what negative treatment means, and a workflow you can rely on before you file.
Citation checking sounds like a clerical task, yet it is where some of the most damaging mistakes in legal practice happen. A brief built on a case that does not exist, or one that was overruled last year, undermines everything around it - and courts have shown they will sanction lawyers who get it wrong. The confusion is that two different failures both look like a citation problem, and they need different checks. Separating them is the first step to a verification habit that actually protects you.
What are the two questions behind every citation?
Behind every authority you cite sit two independent questions. The first is whether the citation is real: does the case exist, and does it actually say what you are citing it for? The second is whether it is still good law: even if it is real and on point, has it survived - or has it been reversed, overruled, or superseded since it was decided? These are not the same check, and clearing one tells you nothing about the other.
A real citation can be bad law, and a hallucinated citation is not even real. Treating citation checking as a single step is how lawyers come unstuck: they confirm a case exists and stop, never asking whether it still stands, or they trust a familiar authority without checking it was quoted correctly. The discipline is to run both checks, deliberately, on every authority that carries weight in your argument.
Is the citation real? The hallucination problem
The is-it-real question has become urgent because generative AI can fabricate citations that look entirely plausible. A general-purpose chatbot, asked for authority, may produce a confident case name, a realistic reporter string, and a persuasive quotation for a decision that was never handed down - or attribute a real case to a holding it never reached. The Mata v. Avianca matter, in which lawyers were sanctioned for filing fabricated cases from a chatbot, is the cautionary tale of the era.
The fix is structural. Tools that generate from memory are the high-risk case; tools that retrieve from real databases and quote the exact passage let you confirm reality in seconds. But no tool removes the final human step: open the source and read it. The unbreakable rule is that you never cite a case you have not read in the original, however convincing the summary. Our guide on how to avoid fake AI citations covers the failure modes in detail.
Is it still good law? Negative treatment explained
The is-it-good-law question asks whether a real, on-point case still carries the authority you need. Cases are not static: they are tested, narrowed, and sometimes destroyed by later developments. The umbrella term for these developments is negative treatment, and it comes in several forms with very different consequences.
| Type of treatment | What happened | Effect on authority |
|---|---|---|
| Reversed | A higher court overturned this decision on appeal in the same case | The decision no longer stands |
| Overruled | A higher court rejected this decision's rule in a later case | The rule is no longer good law |
| Superseded | Legislation has replaced the rule the case applied | The statutory position now governs |
| Distinguished | Later courts confined the decision to its particular facts | Its reach narrows; the core may survive |
| Doubted or criticised | Later courts questioned the reasoning without overruling it | Weakened authority; use with caution |
Not all negative treatment is fatal. A case distinguished on its facts may still be perfectly good for its central principle, and a decision merely doubted remains binding until a court with power actually overrules it. That is why you read the treatment rather than just reacting to a warning flag - the signal tells you to look, not what to conclude.
Why does checking only one of them fail?
Skipping either check produces a different kind of failure, and both are serious. If you confirm a case is good law but never check it is real, you risk relying on a hallucinated or misquoted authority - the case may not exist, or may not say what you think. If you confirm a case is real but never check its status, you risk citing a decision that was overruled or reversed, which an opponent or the bench will expose.
The two checks also tend to fail in opposite situations. Hallucination is the classic risk of generative tools used carelessly; bad-law citation is the classic risk of relying on memory or an old precedent without tracing it forward. A verification routine has to cover both, every time, because a brief is only as strong as its weakest authority. For the broader habit, see how to verify AI legal research.
What is a citator, and do you still need one?
A citator is a tool that traces a case forward through everything that has cited it, flagging negative treatment. The best-known are Shepard's on LexisNexis - the origin of the verb to Shepardize - and KeyCite on Westlaw, which attach signals to indicate a case's status. Citators are the traditional answer to the is-it-good-law question, and where you have access to one, it remains the most thorough way to check subsequent history.
You still need that function, even if not always that product. The concept - systematically reading how later courts treated a decision - is what matters, and it can be done with free sources too, by following the cases that cite yours on services like CourtListener and reading how they treated it. AI can accelerate the gathering, but the judgment about whether treatment is fatal stays with you. The courts' own sites, such as uscourts.gov, help you confirm procedural posture and appeals.
What does a reliable verification workflow look like?
A dependable routine runs the two checks in order and leaves a trail. The steps below are quick once they are habit.
| Step | What you do | Which question it answers |
|---|---|---|
| 1. Open the source | Retrieve and read the actual decision, not a summary | Is it real? |
| 2. Confirm the proposition | Check the quoted passage supports the point you are making | Is it real? |
| 3. Note court and date | Record the deciding court, level, and date | Both |
| 4. Trace it forward | Read how later cases and any statute treated it | Is it good law? |
| 5. Preserve the record | Archive the source and capture the pinpoint citation | Both |
The order matters: there is no point checking the status of a case you have not confirmed is real, and no point relying on a real case whose status you have not checked. Done as a routine, the whole sequence takes minutes per authority and turns citation checking from an anxious afterthought into a reliable, repeatable step.
How does AI make citation checking faster?
AI helps most at the gathering stages. For the is-it-real check, a retrieval-based tool returns the actual passage with a page-level citation, so confirming a quotation is a quick read rather than a separate search. For the is-it-good-law check, AI can surface the later decisions that cite your case and quote how they treated it, compressing the forward trace that would otherwise mean reading through a citation list by hand.
What AI does not do is reach the conclusion for you. Whether a line of doubting cases has crossed into overruling, or whether a distinction confines a case fatally or trivially, is a judgment. The right division of labour is that the tool finds and presents the material at speed, and you decide what it means for the authority you intend to rely on.
How does Judicio support citation checking?
Judicio is designed to make the is-it-real problem hard to fall into. Its Legal Research grounds every answer in a retrieved source and cites the exact page and quoted passage, with deterministic citation labels, so you can open and confirm each authority directly. Every web source is archived as a permanent PDF at the moment it is retrieved, which means a citation cannot rot between research and filing, and you can export an evidence pack of the whole set for your file or a court.
On the is-it-good-law side, Judicio helps you gather later treatment quickly, but it is honest about its role: it is not a citator, and it does not declare a case good law on its own authority. You perform that confirmation, using a dedicated citator where you have one. Outputs are not legal advice. Because one upload into the File Library feeds every tool, the verified authorities flow straight into Drafting. See the whole feature set in one place.
What are the limits you must respect?
Two limits are worth stating plainly. First, no tool - AI or traditional citator - removes your responsibility for the authorities you cite; the duty of candour to the court is personal. Second, automation can lull you into skipping the read. A green signal or a confident summary is an invitation to verify, not a substitute for verification, and the cases where lawyers were sanctioned almost always involved trusting output without opening the source.
The practical takeaway is to keep the human read non-negotiable, especially for any authority that carries real weight in your argument. AI and citators make the checks faster and more thorough; they do not make them optional. Treat both the is-it-real and is-it-good-law questions as your responsibility, and the tools as the means to discharge it efficiently.
How do you get started?
Adopt the two-question routine on your next brief: for every authority, confirm it is real by reading the cited passage, then confirm it is good law by tracing its later treatment, and preserve the source as you go. Use AI to gather the material quickly and a citator where you have one to be thorough. Build it into your process so it happens every time, not just when something feels off.
You can try Judicio's citation-first research on your own matters with a 7-day free trial - 500 credits, no credit card - and move to Professional access at $200 per month for 5,000 credits. For a walkthrough, contact us. To pair this with the wider verification habit, read how to verify AI legal research and our US guide to researching United States law with AI.