LAW · FOR STUDENTS
AI Essay Grader for Law Students: Why Your Self-Check Might Be Lying to You
Many law students paste their essay into an AI tool to find out what it would get. In our test, a law essay the lecturer graded as a fail came back from AI tools as a comfortable pass, and once as a perfect score. Here is why, and how to check your own work in a way you can trust.
By Eduface · September 2026 · 10 min read
It is 11pm, the essay is due at noon, and you have a finished draft. You paste it into ChatGPT with the question and ask what mark it would get. It says the essay is well structured, uses relevant authority and would likely earn a 2:1. You submit. Three weeks later, the mark comes back as a 2:2, and the feedback says your application was thin and your conclusion did not answer the question.
That gap is not bad luck. It is predictable, and it happens for a specific reason.
Can an AI essay grader tell a law student what their essay will get?
Not reliably, if it is a general tool. In our test on a real law essay graded 4.4 (a fail) by the lecturer, ChatGPT returned 7.1, Claude 7.2, Gemini 6.8 and CoGrader a perfect 10.0. General AI responds to how well you write, not how well you reason in law. A purpose-built grader calibrated on academic rubrics came much closer, at 5.5. Use AI to find weaknesses, not to predict your mark.
What happened when we ran a failing law essay through AI graders?
As part of our guide to AI grading tools, two students in the Netherlands tested eight tools on papers that already had a lecturer’s grade, written feedback and the official rubric. Neither student was paid by or connected to any of the tools. One of the papers was a law student’s constitutional law essay, written in Dutch. The lecturer graded it 4.4 on the Dutch 1-10 scale. That is a fail.
The lecturer’s main criticisms were clear:
The research question was restated six times instead of being answered.
Sources were cited but never applied to the legal criteria.
A bridging sentence between two parts of the argument was missing.
There were several typos.
The conclusion did not reach a legal determination.
Every tool got the same input: the full essay and the full rubric. Here is what they returned.
What five AI tools gave a law essay the lecturer failed
CoGrader
10
Claude
7.2
ChatGPT
7.1
Gemini
6.8
Eduface
5.5
Lecturer’s grade: 4.4. Pass mark on the Dutch 1-10 scale: 5.5. Every general tool put a failing essay comfortably above it.
Figure 1: Grades returned by five AI tools for a constitutional law essay the lecturer graded 4.4. Source: Eduface independent student test, July 2026.
CoGrader, a tool built for school pupils, gave the essay full marks on every criterion and described it as having “an excellent structure that flows seamlessly from your creative concept into a professional legal analysis”. If you were the student, you would have submitted a failing essay believing it was exceptional.
That is the real risk of a bad self-check. Not a slightly wrong number, but false confidence at the moment you could still have fixed the essay.
Why does AI over-mark law essays?
There are three reasons, and they compound.
It reads your writing, not your reasoning. General AI models learned from enormous amounts of general text. They are very good at recognising fluent, confident, well-organised prose. A model trained on general text responds to linguistic polish. A model trained on academic assessment responds to disciplinary argument quality. Law marking is almost entirely the second thing.
It sees problems but does not weight them. In our test, Claude spotted the missing bridging sentence, the inconsistent heading numbering and the conclusion that did not answer the research question. Then it graded the essay 7.2. A structural weakness that costs 0.3 points in a general model’s weighting can cost two full points with a university assessor. The observations were right. The judgement of what they cost was not.
It is trying to be helpful. A chat assistant is designed to be encouraging. That is fine for brainstorming. It is the wrong instinct in a marker.
There is a telling finding from the University of Minnesota. In 2023, four law professors had ChatGPT sit four of their real exams and graded its answers blind, alongside real students. It averaged a C+: a low pass in every course.¹ Fluent, confident, mediocre legal prose is what these models produce. It is also what they tend to reward when they mark yours.
What does a law marker actually look for?
Most law assignments reward the same things, whether they are problem questions or essays. If you can check these yourself, you do not need an AI to tell you your mark.
What the marker looks for
The question to ask of your own essay
The common way students lose marks
Issues
Have I identified every legal issue the question raises, including the less obvious ones?
Answering the issue you prepared for instead of the one asked
Rules
Is every rule stated accurately and supported by a case, statute or secondary source?
Stating a rule with no authority, or the wrong authority
Application
Have I applied each rule to these specific facts, step by step?
Restating the rule and jumping straight to a conclusion
Counter-argument
Have I dealt with the strongest argument against my position?
Arguing one side only
Conclusion
Does my conclusion answer the question, as advice a client could act on?
Ending with “it depends” or a summary
Authority and citation
Is every proposition supported and cited consistently, usually in OSCOLA?
Cited but not used: a list of cases with no analysis
Table 1: A self-check you can run on any law essay or problem question.
Most students recognise this as some version of IRAC: issue, rule, application, conclusion. IRAC is a scaffold, not a guarantee. You can have a perfect IRAC structure and very little analysis. The marks are in the A.
How can you use AI on a law essay without fooling yourself?
AI can help you find weaknesses. It cannot tell you what you will get. Used that way round, it is useful.
Ask for weaknesses, not a grade. “Which paragraph applies the law to the facts least convincingly?” is a better prompt than “What mark would this get?”. The first gives you something to fix. The second gives you a number you should not trust.
Give it the marking criteria. Paste in the rubric or marking criteria from your module handbook. Without them, the AI invents its own standard.
Ask it to argue against you. “What is the strongest counter-argument to my conclusion in section 3, and did I address it?” This is where AI is strong, because generating arguments is what it does well.
Check every case it mentions. General AI tools can produce plausible but wrong case names, citations or holdings. Never add authority you have not read yourself.
Do not trust praise. If a tool tells you your essay is excellent, treat that as no information at all. In our test, every general tool praised a failing essay.
A quick self-test for your application
Take one paragraph of application and delete the case names. If what is left is still a specific argument about these facts, you have applied the law. If what is left is “therefore the claimant will succeed”, you have only stated it.
Is it allowed to use AI on your law essay?
It depends on your university and your module. Rules vary, and they are changing fast.
Check your module handbook and your university’s AI policy. Many UK universities now set out, assessment by assessment, whether AI may be used and how it must be declared. Some allow AI for brainstorming and proofreading but not for drafting. Some do not allow it at all.
Declare what you used. If AI is allowed, you will usually be asked to say how you used it. The fifth edition of OSCOLA, published by the Oxford Faculty of Law in March 2026, includes guidance on how to cite output from generative AI.² If you quote or rely on AI output, cite it properly.
Remember you are responsible for every word. If an AI tool gives you a case that does not exist and you include it, that is your error, not the tool’s.
You are not alone in using these tools. In HEPI’s 2026 survey of 1,054 full-time UK undergraduates, 94% said they use generative AI to support assessed work.³ Universities know this. That is why the rules are getting more specific, and why some are adding oral follow-ups to written work.
How is your lecturer’s AI grader different from ChatGPT?
More law schools are starting to use AI in marking. If yours does, it is worth knowing how it differs from the tool you might use yourself.
It uses your module’s rubric and brief. The lecturer sets up the assignment with the brief, the rubric and instructions for how feedback should be written. The AI scores each criterion against that, not against its own idea of a good essay.
It is trained for assessment. Eduface, for example, scores law work with a Law model, one of six discipline models, instead of a general assistant. In our test that was the difference between 5.5 and 7.1 on the same failing essay.
A lecturer approves every mark. The AI’s suggestion is a draft. Your lecturer reviews it, changes what they disagree with and approves it. Nothing reaches you before that. Feedback marked “Lecturer + AI” has been read by your lecturer.
You see why. Instead of a single number, you get a score and a written reason for each criterion, with comments anchored to the exact passage in your essay.
1
You submit through Moodle, Canvas, Blackboard or Brightspace
2
The AI drafts a score and comment for each criterion
3
Your lecturer reviews, edits and approves
4
You receive the approved mark and feedback in your VLE
Your lecturer, not the AI, decides your mark.
Some law schools also use oral follow-ups. With Eduface’s Academic Integrity tool, currently in beta, you may be asked critical questions about your own essay and answer them out loud, as in a short viva. If you wrote the essay, that is easy. If you did not, it usually shows.
If you want to know what good feedback on a law essay looks like, and why most AI tools do not give it, see the best AI feedback tool for law essays.
What should you do the night before a law deadline?
Read the question again. Then read your conclusion. Does the conclusion answer that question, in one or two sentences a client could act on?
Check your application paragraphs. Use the self-test above. Every rule needs a “because on these facts…” sentence.
Find your weakest counter-argument. If you have not addressed the obvious objection, add one paragraph. That is often worth more than polishing ten sentences.
Check your authority. Every case you cite: have you read it, is it cited correctly, and have you used it rather than just named it?
Then, if you want, ask an AI for weaknesses. Not for a mark.
Frequently asked questions
Can I use ChatGPT to grade my law essay?
You can ask, but do not trust the number. In our test, ChatGPT graded a failing law essay at 7.1 against the lecturer’s 4.4. General AI responds to fluent writing, not to the quality of your legal reasoning. Ask it for weaknesses and counter-arguments instead of a grade.
Why did an AI tool give my essay a high mark when my lecturer gave it a low one?
Usually because the essay was well written but the legal analysis was thin, typically in the application of law to facts or the conclusion. General AI tools reward polish. Law markers reward reasoning. Compare the lecturer’s criterion-level feedback with the self-check table above to see where the gap was.
Is using an AI essay grader on my law essay cheating?
Checking your own work is not usually cheating, but using AI to write or rewrite it may be, depending on your university’s rules. Check your module handbook, declare any use your university requires you to declare, and cite AI output in line with OSCOLA’s fifth edition if you rely on it.
Can I use Eduface to check my own law essay?
Eduface is built for lecturers and institutions. If your module uses it, you receive criterion-level feedback and your approved mark through your VLE. Your lecturer reviews every mark before you see it.
What is the most accurate AI grader for law essays?
In our independent test on a Dutch-language constitutional law essay, Eduface came closest to the lecturer’s grade: 5.5 against 4.4. General tools were 2.4 to 2.8 points too generous and CoGrader gave full marks. One essay is a small sample, which is why no AI grade should replace your lecturer’s.
References
1. Choi, J. H., Hickman, K. E., Monahan, A., & Schwarcz, D. (2023). ChatGPT goes to law school. Journal of Legal Education, 71(3). [ChatGPT sat four real University of Minnesota law exams, graded blind; it averaged a C+.]
2. Goudkamp, J. (ed.). (2026). OSCOLA: The Oxford University Standard for Citation of Legal Authorities (5th ed.). Oxford Faculty of Law and Hart Publishing. [Published 25 March 2026; includes guidance on citing output from generative AI.]
3. Higher Education Policy Institute. (2026). Student Generative AI Survey 2026 (HEPI Report 199), with Kortext. [94% of 1,054 full-time UK undergraduates use generative AI to support assessed work.]
4. Eduface. (2026). The Complete Guide to AI Grading Tools for Higher Education. [Independent student test of eight tools, including a constitutional law essay graded 4.4.]
Ask your law school about it
Eduface gives law students criterion-level feedback with a Law model, and a lecturer approves every mark. Book a demo or start free.