At a glance
Google Scholar's case law tab gives lawyers, journalists and self-represented litigants free access to millions of US court opinions. It is not a citator, and courts are starting to price that difference in indemnity costs.
Google Scholar's case law tab has been open since 2009, and in the years since it has become the most widely used free citator on earth — a title it never asked for and does not earn.
What it does earn is narrower and genuinely useful: full-text search across published American court opinions, at no cost, with the citations inside each judgment live and clickable. For a solo practitioner in a county courthouse, a reporter checking a quote, or a litigant who cannot afford a subscription, that is not a small thing. Westlaw and LexisNexis built two very large businesses on the assumption that it would never exist.
The trouble starts where the searching stops.
What the tab holds, and where the floor gives way
The coverage boundaries are unusually well documented for a Google product: US Supreme Court opinions from 1791, federal district, appellate, tax and bankruptcy courts from 1923, and state appellate and supreme court decisions from 1950. Google publishes those dates itself and then disclaims the consequences — opinions are offered for informational purposes only, and the company does not warrant that they are complete or accurate.
1950 is the number that catches people. A 1938 state supreme court ruling on a contract dispute is not there. Neither are state trial courts, statutes, regulations, or the annotated codes that tell you how a statute has actually been applied. Scholar indexes judicial opinions, and only some of them.
Inside those limits the corpus is vast, and ranking is the real constraint. Results are ordered by an algorithm that weighs citation counts and full-text relevance, and the list stops at a thousand. If the controlling case sits at position 1,001, you will not see it, and nothing will tell you it exists.
How Cited is a lead generator, not a citator
Every opinion page carries two tabs: read the case, and "How cited." The second is the feature that made Scholar indispensable to lawyers without a subscription. It splits into three parts — a panel of excerpts showing how later courts have characterised the decision, a "cited by" list of everything in Scholar that references it, and a set of related documents.
Since March 2012 the citing list has been ordered by how deeply each later opinion engages with the case rather than by prominence, with a three-bar marker giving a rough sense of the depth of discussion. It is a good piece of design. It is also entirely mechanical.
Nobody at Google has read these cases. No editor attaches a treatment flag; no warning appears when a decision has been overruled, abrogated, or distinguished into oblivion. A dead case and a live one look identical in a search result.
The Library of Congress guide to free case law is careful about this, describing How cited as a place to begin rather than an authority. Commercial citators — Shepard's, KeyCite — employ lawyers and editors to read and classify how courts have treated a case. Scholar employs an index.
A white paper published by LexisNexis, which competes with the free services it criticises and should be read with that in mind, lists Supreme Court and appellate opinions that courts corrected after publication and which continued to circulate uncorrected in free databases: a 2014 Supreme Court case where the text read "judgment debtor" instead of "judgment creditor," a district court opinion in which the parties had been transposed. The same paper notes that a withdrawn Arkansas decision, Bradford v. Bradford, stayed visible on free services including Scholar.
Treat the vendor's motive as suspect and its examples as checkable. The point survives either way. Scholar reproduces what it crawled, and courts do not always publish cleanly.
Photo by Rubaitul Azad on Unsplash
A junior solicitor, a native title claim, and an indemnity costs order
In July 2025 the Federal Court of Australia published its reasons in Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria. The facts are worth setting out slowly, because they describe a workflow rather than a scandal.
A junior solicitor at a Melbourne firm, Massar Briggs Law, prepared the footnotes for a summary document in a native title proceeding. Working from home, without access to the firm's copies of the reports she was citing, she used Google Scholar to assemble the references. When the respondent's legal service went looking for those sources, it found that most did not exist and that others existed but had been wrongly cited. Justice Bernard Murphy described the citations as fabricated.
Asked to explain, the solicitor re-ran her searches and got different results, and could not say why. She had used Scholar throughout her studies without trouble. The firm was ordered to pay the respondents' costs on an indemnity basis. Neither the junior solicitor nor her principal was referred to the Victorian Legal Services Board.
"The error was centrally one of failing to check and verify the output of the search tool," Murphy wrote, attributing it to inexperience and to a supervisor who conceded he was not aware that anyone had checked the work.
Reporters described an AI-assisted tool inside Scholar. The judgment does not turn on which button was pressed. It turns on a document going to a court with sources that nobody confirmed, and on a search engine being treated as though it were a library.
The scale of that mistake is now measurable. Damien Charlotin, a researcher at HEC Paris, maintains a database of decisions in which courts found or clearly implied reliance on fabricated material: 2,202 decisions worldwide as of 19 October 2026, of which 1,501 were in the United States. Lawyers account for 872 of them; self-represented litigants for 1,267. New cases fell in each of the first three quarters of 2026 — 455, then 423, then 359 — which Charlotin credits partly to better models and partly to fear.
The free-law plumbing Scholar now sits on
For sixteen years Scholar was the only serious free option. That has changed, and the change matters to anyone who has been using it as a workaround.
Harvard Law School's Library Innovation Lab digitised the whole of its printed case law collection — roughly six million cases, some forty million pages — and in March 2024 the commercial restrictions on that data expired. The lab then wound down case.law's search and API and handed the interactive layer to the Free Law Project's CourtListener, keeping case.law as a bulk download and browsing site. The old interface retired that September.
CourtListener absorbed nearly nine million historical decisions in 2024. In April 2026 the Free Law Project began scanning case reporters for opinions published since 2018 — the versions with official pagination, the ones courts actually cite — using a machine-learning tool called Blackletter to strip out publisher editorial material that may be copyrighted. It has also been building an AI citator with more than eighty volunteers drawn from law librarians, court clerks and practitioners, to flag unreliable case law.
Read those developments together. Google built a better index and stopped there. The nonprofit sector is now building the thing Scholar deliberately never was: a citator, free and open, with humans in the loop.
Scholar Labs answers questions. Courts ask different ones.
In November 2025 Google introduced Scholar Labs, a generative-AI layer that takes a research question, breaks it into facets, and returns a short list of papers with an explanation of why each was chosen. Access began with a waitlist for logged-in accounts.
Early reviewers found it fast and precise within limits, and the limits are instructive. It lacks the custom year-range filter that ordinary Scholar offers. The Singapore Management University library's assessment found it evaluates a few hundred results at most and stops after fifty relevant ones, which rules it out for systematic review. The same assessment cites a study in which Scholar held 97.2 percent of the papers found in a set of systematic reviews but returned only 46.4 percent of them through an actual subject search.
That gap between what a database contains and what a search can surface is the entire problem in miniature, and it is not new. It is why the index everyone uses is not the same thing as the record everyone needs.
Photo by Shutter Speed on Unsplash
How to use it without joining a sanctions docket
None of this argues for abandoning Scholar. It argues for knowing which of its two jobs you are asking it to do.
- Treat How cited as a list of leads. Open the citing opinions and read the paragraph where your case appears, not the excerpt Google chose for you.
- Check the posture and the date. A 1950 appellate decision in a state that has since rewritten the statute is not authority, whatever the search result implies.
- Verify the text against a reporter page. Scanned originals at case.law and CourtListener exist precisely because web versions of opinions drift.
- Run a real citator on anything you file. Shepard's or KeyCite if you have them, CourtListener's tools if you don't, and a human read of the most recent citing case either way.
- Never outsource a citation to a model. The failure in Melbourne was not that a tool lied. It was that nobody looked.
There is a version of this story in which the machines get good enough that none of it matters, and that version may even arrive. Until then the discipline is the one a nineteenth-century lawyer applied to a shelf of reports: find the case, read the case, then find out what happened to it after it was decided.
A search engine will always answer the question you typed. The law has a habit of turning on the one you didn't.
- legal-research
- legal-technology
- google-scholar-case-law
- case-law-search
- courts

