Do Law Schools Use AI Detectors? Every Published Number (2026)

1 Do Law Schools Actually Run Detectors on Applications?

The most useful evidence here is the disagreement inside a single source. GradPilot’s table marks Columbia Law with the label Uses screening tools. Its Columbia page quotes a certification that no artificial intelligence tool has written or modified any component, then says the school has not indicated how they check in admissions essays. The label covers forensic checks, timed samples and draft requests as well as software, so it cannot carry the weight some readers will put on it. (GradPilot, 2026)

Commercial interests run through this whole record, and readers should weigh them. Pangram sells detection. GradPilot, the fullest tracker of these rules, says partner with Pangram Labs, which means the best dataset in this field comes from a firm with a detector partner. Spivey sells admissions consulting. None of that makes their statements false. It does mean the question has no neutral survey behind it, unlike undergraduate admissions, where more numbers exist. (Pangram, 2026)

The body closest to the applications is sceptical. LSAC runs the application system nearly every American law school uses, and its technology officer tested detectors personally, then advised everyone to be especially cautious about relying on them. That advice dates from 2023, so it predates current models, but we found no later LSAC statement replacing it. (LSAC, 2023)

2 How Many Law Schools Prohibit AI in the Admissions Essay?

These figures are two years old and were self-reported by admissions officers, so treat them as a floor on how fast policy was moving, not a current census. They still show the pattern that matters. Nearly half of offices had a written position on AI writing the essay, and about seven in ten had none on lighter uses, which is where most applicant use sits. Kaplan’s own executive warned that no policy may be read by applicants as a de facto policy of allowing it.

An earlier American Bar Association survey points the same way, with 62 percent of responding deans still undecided on applications. Its sample was only 29 deans, and its authors cautioned against leaning on it, so it is direction rather than measurement. (GovTech on the ABA survey, 2024)

3 What Do Law Schools Actually Publish?

Set this beside Kaplan and an apparent contradiction appears. Nearly half of surveyed officers reported an official ban on AI writing, yet only about one tracked school in seven publishes a J.D. rule. Both can be true, because a certification inside the application is invisible to anyone reading the public requirements page. GradPilot makes this point itself: silence on a public page is not permission. That is the single most practical sentence for an applicant on this topic.

The wording varies more than a strict-to-lenient scale suggests. UCLA says any AI tool used to assist in completing the application is prohibited. Georgetown allows feedback but reserves the writing to the applicant. Richmond permits defined editing and requires disclosure. Harvard, Yale, Berkeley and Texas publish no first-year J.D. essay rule, as GradPilot checked them in July. Read the sentence your school actually wrote, then read the certification you are about to sign. (GradPilot, 2026)

4 What Are the Rules Once You Are Enrolled?

The blind spot in most writing on this topic sits here. Law schools have a form of evidence other departments lack: a citation either exists or it does not. Berkeley Law states that citations to sources that do not exist raise a presumption of prohibited AI use. No probability score is involved, and the student carries the burden from that point. For a law student, an invented case is a more dangerous trace than any percentage, and it is the same error that has followed AI use into professional settings. (Berkeley Law, 2026)

Detectors do still appear, with caveats attached. San Diego Law’s research guide says a suspicious faculty member may choose to use an AI detector, and the same guide says detectors are not recommended as a sole indicator. That is the most candid statement we found from a law school, and it is rare because Backer’s study describes policy text that is often absent, gated or oral. He argues this sits in tension with the fair notice that integrity enforcement requires. Applicants to a program should expect to ask, because the university-wide tool list may not settle what the law faculty does. (University of San Diego Legal Research Center, 2026) (Backer, SSRN, 2026)

5 How Accurate Are the Detectors Law Schools Are Told About?

Vendor claims and independent results are not measuring the same thing, and we cover that at length in our false positive data by tool and our Turnitin statistics. What matters for law school is arithmetic rather than a tool ranking. Vanderbilt’s sum was 75,000 papers and roughly 750 wrong labels at the vendor’s own stated rate, which is why it disabled the tool. Law school cohorts are small, so a school sees few errors and has little means of noticing them. (Vanderbilt University, 2023)

No published study we found tests detectors on legal writing specifically. Legal prose is formal, formulaic and full of quoted authority, and that is a reasonable place to worry about false positives, but worry is not data. LSAC’s own experiment is the nearest thing: AI comparing personal statements with writing samples got authorship right better than two-thirds of the time, and LSAC decided that was not good enough for applicants, partly because a 35-minute timed essay and a months-long statement differ so much in tone. (LSAC, 2023)

6 What Has Actually Happened to Applicants and Students?

The Juris sample comes from an anonymous Google Form run by an admissions consultancy, so its percentages describe its respondents and not all applicants. Its breakdown is still instructive, because most reported use falls in exactly the territory that Kaplan found most schools had not addressed. (Juris Education, 2025)

The Purdue Global case is worth reading for what the accusation rested on. According to The Indiana Lawyer, it rested in part on the speed and appearance of the student’s writing, along with citation concerns. The school’s policy says faculty cannot rely solely on detection tools, and the article names no detector and no score. The court had ruled on a sealing motion by April 27 and the reporting shows no ruling on the merits, so nothing here establishes who is right. It does show a law school dismissal can proceed with no detector score reported at all, which is why other AI lawsuits are only a partial guide, and why keeping a record of your drafts matters more than checking a score. (The Indiana Lawyer, 2026)

We found no published count of AI honor cases at American law schools, and no reliable figure for how such a finding affects bar admission. Anyone quoting one should be asked for the source. What is certain is narrower. A certification is a signed statement, Michigan’s warns that false statements may cost an offer or a degree according to GradPilot’s quotation, and an integrity finding can follow you in ways a detector percentage alone does not.