An AI legal research system must confirm that every case it cites remains good law, a judgment no language model can make on its own. Knowing whether a single case has been overruled, reversed, or limited requires a record of how every later decision treats every earlier one, across millions of opinions. A language model holds no such record, whatever its training data contains. That bookkeeping lives in the citator, curated treatment data the profession has maintained for two centuries. Built into the research process and visible in the work product, treatment analysis is what makes an AI-cited case safe to rely on.
Good Law and Treatment Analysis
What does it mean for a case to be good law?
A case is good law if no later decision or statute has overruled or undermined its authority. The concept is a term of art with real professional stakes. Citing overruled law to a court damages the argument and the lawyer's standing with the judge. Advice built on bad law can constitute malpractice.
A direct overruling is only the clearest way a case loses authority. A decision may be reversed on appeal. A later court may overrule it by name or reject its reasoning. A legislature may abrogate its holding by statute. An opinion that articulates several rules of law can also lose authority on one point while remaining good law on the others. Citators therefore track treatment point by point rather than case by case.
The principle that an authority is only as good as its subsequent treatment has governed legal research since Simon Greenleaf published the first American citator in 1821, after learning the lesson in a Maine courtroom at his client's expense. The full story is in Chapter 2, and its narrative telling in Servient's companion blog series.
What is treatment analysis in legal research?
In legal research, treatment analysis is the citator's record of how each later decision has dealt with an earlier case, whether it followed, distinguished, criticized, questioned, limited, or overruled it. Each notation records the treating court's action. References tied to the individual point of law then show which of the earlier case's rules the treatment touched.
Treatment is richer than a binary valid-or-not flag. A case followed in forty decisions carries different weight than one narrowly confined to its facts. One distinguished in a recent decision may be perfectly good law that simply does not reach your situation. Criticism and questioning weaken a case's persuasive force without removing its authority. An overruling removes the authority itself, often on only a single point of law.
Reading treatment well is a lawyerly skill. The researcher weighs which court issued the treating decision, how recent it is, and which point of law it addressed, then judges what remains of the earlier case's force. An agentic legal research system must both perform that judgment and expose it for the lawyer's review.
What is the difference between direct history and subsequent treatment in a citator?
In a citator, direct history is what happens to a case within its own litigation, the appeals, affirmances, reversals, and remands of the dispute itself. Subsequent treatment is how courts handle the opinion in other cases, following, distinguishing, criticizing, or overruling it. Either record can end a case's authority.
Direct history follows the single dispute through the court system, from appeal through affirmance, reversal, or remand to review in a higher court. A reversal there is decisive, because the opinion loses its authority in the very case that produced it. Checking direct history answers whether the opinion in hand survived its own appeal.
Subsequent treatment accumulates outside the case, in later decisions that cite the opinion. A court can overrule its own precedent in an unrelated case when it changes its view of the law. A decision never disturbed on appeal can therefore lose its authority years later, in litigation between strangers to the original dispute. Courts that cannot overrule the case, in other jurisdictions or lower in the hierarchy, can still distinguish, criticize, or question it, notations that narrow the opinion's reach or invite later courts to reject it.
A statute can end a case's authority as well. When a legislature changes the rule of law an opinion announced and relied on, the opinion is effectively overruled even though no court has treated it. Citators mark the case as abrogated or superseded by statute. An AI legal research system must check all three, the case's own history, the later decisions that cite it, and the statutes that may have displaced it, before treating the opinion as good law.
The Citator in AI Legal Research
Why does AI legal research need a citator?
AI legal research needs a citator because a language model cannot track how every later decision treats each earlier case. Whether an authority remains good law is a fact spread across the later case law, held in curated citation data that the system consults as a tool. Without that check, an AI system can present an overruled case as confidently as a controlling one.
The check runs against the citator's records. The agent confirms the case's direct history, that the opinion survived its own appeals, and its subsequent treatment, that no later decision has overruled, limited, or otherwise undermined it. A statute that displaces the rule of law counts as well. None of those answers appears in the text of the opinion the agent is reading.
An AI legal research system must therefore confirm every authority against current citation data before placing it in the work product. In an agentic workflow the agent checks treatment as it reads each authority, not in a final pass over a finished draft. This is the same principle Greenleaf established, applied to a new researcher. The citator's record of subsequent treatment is what makes the transition to AI legal research trustworthy.
What happens when AI legal research cites an overruled case?
When AI legal research cites an overruled case, the consequences run from embarrassment to sanctions to malpractice exposure. The argument collapses the moment opposing counsel or the court finds the overruling decision. The error then costs the lawyer credibility on every point that follows.
In federal court, Rule 11 of the Federal Rules of Civil Procedure makes every legal contention in a filing the responsibility of the attorney who signs it. A contention that rests on an overruled case as controlling law is not warranted by existing law and exposes the signer to sanctions. A lawyer who discovers the error after filing must correct the record.
The professional duty involved predates the technology. Lawyers have always been required to confirm their authorities. What changed is the pace. An AI system produces cited work product faster than any lawyer could vet each authority by hand, which is why the good-law check belongs inside the research system rather than in a reminder to the reader.
How does an AI agent use citator treatment data to find related cases?
An AI agent uses citator treatment data to find related cases, following each recorded relationship as a retrieval path through the law. Every treatment notation joins two decisions that address the same point of law, so the record built to answer the good-law question doubles as a guide to the case law around it.
Treatment leads the agent to law no query would return. When the agent finds that an authority has been distinguished, the treating opinion becomes the next case it reads. That opinion supplies both the limit on the earlier case and the rule the later court applied in its place. An overruling works the same way, redirecting the research from the superseded rule to the law that currently controls. The agent also reads the treating opinion itself instead of relying on the notation alone. A distinction on an unrelated point of law leaves the case's authority intact, while one on the point under research changes the analysis.
An agent that follows those connections moves along the joints of doctrine itself. Chapter 5 describes this graph traversal in detail. Citator data built for validity checking turns out to be one of the richest maps of the law ever assembled.
Verification in Practice
How do you verify that an AI-cited case is still good law?
You verify that an AI-cited case is still good law by checking its direct history and its subsequent treatment against current citation data. A system built for verification has run that check before the case reaches you and shows you what it found.
Three capabilities mark a system built to that standard. First, every cited authority should arrive already checked, with its treatment visible, including the negative treatment. Second, the system should show the treating decisions themselves, so confirming a case's status means reading its treatment rather than re-running the research. Third, the treatment should be tied to the point of law the work product relies on, since a case overruled on one point may remain good law on the point being cited.
Verification a lawyer can complete in seconds is the practical standard, because a check that consumes the time the system saved is not a solution.
Treatment analysis answers whether a real, relied-upon case still stands. Whether an AI system has invented a case, or misstated what a real one says, is a separate problem with a separate engineering answer, the subject of Chapters 7 and 8.
