Law or Grad School Essay Flagged as AI? What Your Certification Says

A checker's score and the certification you sign answer different questions. The certification is about what you did: Columbia Law's Fall 2027 application has you certify that no "artificial intelligence tool or service has written or modified any component" of it, and Duke Law requires that you "wrote all drafts and the final product without the assistance of generative artificial intelligence." A detector only guesses from the finished text, and LSAC's own blog has described a tool repeatedly labelling its author's writing as ChatGPT's. If no AI tool touched your essay, the flag does not make your certification untrue, so keep your drafts and do not rewrite to move the number. If an AI tool did edit it, even just for grammar, that is what you need to sort out before you sign, whatever the checker said.

We read four application pages for the current cycle on 2 October 2026, two from law schools and two from graduate schools. They are far apart.

Northwestern's career office described the same split in Fall 2024: "most law schools prohibit the use of generative Artificial Intelligence (AI), especially pertaining to any application essays", while "most graduate schools do not have specific policies that explicitly address the use of generative Artificial Intelligence (AI) in preparing applications." That summary is two years old. The four pages above fit it, but four pages are not a survey, and Northwestern also notes that a handful of law schools explicitly allow generative AI. What one school says tells you nothing reliable about the next one on your list.

Columbia's certification opens with a paragraph that is easy to skim past. The application and its supporting materials "will be retained by Columbia Law School and made available to the committee of character and fitness of the state(s) in which you intend to practice", and "Honesty in the law school application and admission process is essential in satisfying the requirements of good character required by state bars." Northwestern spells out what that means for AI: if an applicant is found to have used generative AI to prepare the application, "that can result in a Character and Fitness issue for state bar admission purposes for the applicant." So this is not a box that stops mattering once you are admitted.

Now look at the verbs. Columbia's word is "modified". A chatbot that only fixed your commas has modified the essay. A version of this worry turns up in applicant forums: someone wrote every word themselves, asked ChatGPT to clean up the grammar, and then a checker flagged it. On Columbia's wording, the grammar pass is the problem, and the flag is beside the point.

Duke draws the line in a different place. Its rule covers "all drafts and the final product", so an AI pass on an early draft counts as well. It also says the rule "does not preclude asking family members, friends, pre-law advisors, and others for proofreading assistance or general feedback." Columbia has no line like that. What it prohibits is "Assistance in completing any components of your Columbia application", and the agents, third-party vendors, educational consultants and AI tools it names come after "including but not limited to": they are examples, not the whole list. Friends and advisers are not mentioned, and the certification calls the application "exclusively my own work". If someone read or commented on your essay before it goes to Columbia, ask the admissions office in writing whether that counts. Do not guess.

UNC-Chapel Hill's Graduate School allows some use and asks you to cite it: "The starting point for writing statements of purpose and other application requirements should be applicant-generated. While the use of generative AI tools may be useful for structuring and editing writing, it should primarily be the voice of the applicant. Any use of AI must be cited clearly." BU's Graduate School of Arts & Sciences, on its MA and MS application page, allows tools "to support your original ideas rather than to compose your work in its entirety" and expects applicants to meet the same academic honesty standards as its current students. It asks for no disclosure, and its PhD and MFA application page has no AI statement at all. Cornell's Graduate School also requires disclosure of any generative AI tool, even one used for a permitted purpose, and specifies the form it takes. We quoted its wording in Do colleges check application essays for AI?, which covers the undergraduate side.

The graduate school's page is not always the last word. UNC says the statement's form and content requirements "may vary by program". Read the program's own application page too.

One question applicants ask: if you used no AI at all, should you add a line saying so? Neither UNC nor BU asks for one. UNC asks you to cite the AI you did use. If a program's form has a disclosure question, answer exactly what it asks. If it has none and you are unsure, email the program and ask, rather than writing a statement nobody requested.

It is not evidence that you used AI. A post on the blog of LSAC, which runs the LSAT and the law school application service, put it plainly. In August 2023 the author described testing AI detectors and wrote: "I found myself scandalized when one tool repeatedly and erroneously identified my own work as likely written by ChatGPT." He returned to it that October: "False positives seem to me to be a problem far greater than some students using ChatGPT as they might lead to real accusations with real consequences, up to and including not receiving an admissions offer they otherwise would have received." Both posts are from 2023 and describe one person's testing, not an LSAC rule.

It does not tell you what a school will see. None of the four application pages says whether or how essays are screened. In the text of all four (Columbia's 23-page preview plus the Duke, UNC and BU pages, about 157,700 characters together), "detect", "Turnitin" and "software" each appear zero times. "Applica", as in application and applicant, appears 540 times, so the search itself works. That silence is not a promise that nobody checks. Columbia states that "all materials are subject to verification by the Admissions Committee".

If you took the LSAT, law schools also have a sample of your writing done under supervision. LSAT Argumentative Writing is "a remote-proctored, online writing test", and LSAC says "Your writing sample will be shared with the law schools to which you have applied once it’s approved and your score is released." That does not mean schools run your statement against it by machine. In the 2023 post, LSAC's author tried using AI to compare personal statements with writing samples and concluded "the technology is not yet accurate enough to be integrated into LSAC’s current offerings". It does mean your file already holds writing you produced with no help available.

So what do you do with the score? Mostly, leave it alone. Reworking a statement you wrote yourself to move a free checker's percentage spends days on a test the school may not be running, and it changes nothing about what you can certify. At a school with Columbia's wording, running the essay through an AI paraphraser or rewriting tool is exactly an AI tool modifying your essay, the one thing you certify did not happen. If you want to revise because the essay reads as generic, that is a writing problem, and at a school with Columbia's or Duke's wording it is one you fix by hand.

1. Read the certification in the real application, one school at a time. Previews and admissions pages help you prepare, but Columbia's own preview tells you to "refer to the application on LSAC.org for the current application instructions and questions."
2. Write down honestly what touched each essay. A chatbot, an AI feature inside a writing app, a translation tool, a friend, a paid consultant. Then hold that list against each school's wording. The same essay can be clearly fine at Duke and something to ask Columbia about first, for example if a friend proofread it.
3. If an AI tool edited an essay going to a school that bans it, undo that before you sign. Go back to your last draft from before the AI edit and make the corrections yourself, or have a person help if that school allows it. If you are not sure whether what you did counts, ask the school before you submit, not after.
4. Keep every draft. Northwestern advises applicants to "save copies of all their drafts with their name and draft number in the case the application material is ever called into question for the use of generative AI." Google Docs and Word keep a version history that does the same job, and How to keep version history in Word, Google Docs and Overleaf shows where it is and how to export it. Do not retype a finished essay into a new file to create a history. That article explains why a record made after the fact works against you.
5. Ask in writing, and keep the reply. Northwestern's advice when you have doubts is to consult the school's admissions office or a prelaw adviser. An emailed answer from the school is something you can point to later.
6. For graduate programs, answer the disclosure question the form actually asks, in the form it asks for.

Do law schools run AI detectors on personal statements? The application pages we read do not say either way. Columbia says materials are subject to verification, and LSAC's October 2023 post advised "everyone to be especially cautious about relying on tools marketed as ChatGPT detectors." If it matters to you for a particular school, ask that school's admissions office.

I only used ChatGPT to fix grammar. Can I sign Columbia's certification? Its wording is that no AI tool or service "has written or modified any component" of the application. Grammar corrections modify the text. To sign it as written, redo the corrections yourself starting from your draft before the AI pass, or ask the admissions office first.

Can a friend or prelaw adviser read my essay? At Duke, yes: it expressly allows proofreading and general feedback from family members, friends and pre-law advisors. Columbia prohibits "Assistance in completing any components of your Columbia application"; agents, vendors, consultants and AI are listed as examples, not as the whole list. It does not mention a friend proofreading on its own, so ask Columbia's admissions office in writing before you rely on it.

• Columbia Law School, J.D. Application Fall 2027, preview copy (PDF), linked from its Regular Admission page, read 2 October 2026.
• Duke University School of Law, Juris Doctor Application Information, read 2 October 2026.
• The University of North Carolina at Chapel Hill Graduate School, Instructions for Graduate Applicants, read 2 October 2026.
• Boston University Graduate School of Arts & Sciences, MA & MS Application Requirements & Deadline Information, read 2 October 2026.
• Northwestern Career Advancement, Using AI for Grad or Law School (written "As of Fall 2024"), read 2 October 2026.
• LSAC, ChatGPT, Law School Application Personal Statements, and the LSAT Writing Sample (31 August 2023) and ChatGPT Detectors Revisited (5 October 2023), read 2 October 2026.
• LSAC, LSAT Argumentative Writing, read 2 October 2026.