Law Essay Flagged as AI: IRAC, Quoted Authority and OSCOLA Footnotes

A law essay hands a detector an unusual amount of prose that was written to a shape. IRAC and CLEO fix the order of your paragraphs and, in the way they are taught, the opening words of them. Case treatment arrives in a house pattern. Then a large share of what is left is quotation you must not touch, because changing the words of a quoted section of an Act is not editing, it is misquoting.

Two things follow, and they land in different places.

On the AI Writing Report, the percentage is worked out over qualifying prose and nothing else. Turnitin's pages that define that report never mention quotations, footnotes or endnotes at all, so nothing published tells you your block quotes sit outside the number. On the Similarity Report, quoted and cited material does have documented handling, and across the pages discussed here that is the only place your citation style is named at all: citation recognition is documented as trained on APA, MLA, Turabian and IEEE numbered citations. OSCOLA is not on that list. Neither is Bluebook.

So the two halves of a law essay need two different responses, and treating them as one problem is how people end up rewriting a quotation that never needed a word changed.

Start with the denominator, because a law essay's is unusual. The FAQ defines the scored material:

"This qualifying text includes only prose sentences, meaning that we only analyze blocks of text that are written in standard grammatical sentences and do not include other types of writing such as lists, bullet points (short non-sentence structures), or other non-sentence structures."

And, on the size of the number relative to your file:

"As noted previously, this percentage is not necessarily the percentage of the entire submission. If text within the submission is not considered long-form prose text, it will not be included."

That leaves an obvious question for anyone writing OSCOLA or Bluebook, where a large share of the word count can end up below the line in footnotes: are footnotes prose? Is a three-page block quote from a judgment prose? The documentation does not say. I checked rather than guessed.

What I counted. On 27 August 2026 I opened the three Turnitin pages that define the AI Writing Report (Using the AI Writing Report, Turnitin's AI writing detection capabilities FAQs, and the AI writing detection model release notes) in a browser and read the rendered body text: 7,218 + 31,712 + 9,957 = 48,887 characters. Searched across all three: `quotation` 0 · `quote` 0 · `footnote` 0 · `endnote` 0 · `citation` 0 · `verbatim` 0 · `statute` 0 · `legal` 0. The string `cite` returned three hits on the release-notes page, all of them inside the word "excited". Sensitivity check on the same text, so you know the search was working: `prose` 13 · `qualifying` 15 · `bibliograph` 4.

The one exclusion that is published is the reference list. An August 2023 release note says "Bibliographies are now excluded when processing the AI writing report", and the sentence after it tells you existing submissions need resubmitting to benefit. A bibliography is not a footnote, and the release note says nothing about footnotes.

Which discipline you are in does not appear either. The word `law` returns zero across those 48,887 characters. `subject area` returns three times, all on the FAQ page and all describing the training sample: once as "academic writing across geographies and subject areas", then twice in a repeated list of "less common subject areas such as anthropology, geology, sociology, and others". The nearest thing to a statement about genre sits inside the definition of qualifying text, which describes "a longer piece of written work, such as an essay, a dissertation, or an article, etc."

Nothing published carves out an exception for legal writing. That cuts both ways: there is no special leniency for a case note, and there is no special scrutiny either.

IRAC is not a stylistic preference you drifted into. Touro Law Center's study-skills page states it plainly: "IRAC stands for the 'Issue, Rule, Application, Conclusion' structure of legal analysis. An effective essay follows some form of the IRAC structure where it is organized around an 'issue', a 'rule', an 'application', and a 'conclusion' for each and every issue and sub-issue identified as a legal problem." The same page calls it "a supporting scaffold (or training wheels)".

Read the instructions on that page next to what Turnitin publishes about text that produces false positives, and the collision is not subtle.

Turnitin's own account of false-positive-prone text:

"Sometimes false positives (incorrectly flagging human-written text as AI-generated), can include content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas."

Read the sentence that follows it before you take the report to a tutor: "If our indicator shows a higher amount of AI writing in such text, we advise you to take that into consideration when looking at the percentage indicated."

That is a description of a taught legal answer, written by the vendor, without anyone in the conversation having mentioned law. How Essay Structure Affects Your Turnitin AI Score is the general version of this.

Before you conclude that none of it can move, read further down the same Touro page. Two sentences after the "whether, when" advice it says this:

Some professors might not want to see this language – “the issue is whether.” You achieve the same result with other words – “Did” or “Can”, for example. Don’t get fixated on language. Follow your individual professor’s instruction and realize that either way, you achieve the same result: identification of the legal problem.

So the four words at the top of each issue are movable, and the page that taught them to you is the thing saying so. What costs something is the rest of it: the order of the four moves, and the instruction to run them for every issue and sub-issue. That is the part a marking scheme is usually checking, which is why "just vary your structure" is cheap advice in a discipline where the structure was your own choice and expensive advice here.

Everything above is about the AI number. Quoted authority is a different report with different rules, and the rules are worth knowing precisely, because law is the discipline where they bite hardest.

OSCOLA fixes the mechanics. The fifth edition says "Quotations must be faithful to the original", allowing only a change of quotation marks between single and double, with `[sic]` marking errors you must otherwise preserve. It also splits quotations by length: up to three lines they go inside your sentence in single inverted commas, and past three lines they become "an indented paragraph, with no further indentation of the first line", where you "do not use quotation marks". Citations go in footnotes: "When citing a source, place the reference in a footnote. Indicate footnotes with a superscript number."

Now put that against Turnitin's Similarity Report machinery.

The single inverted comma is worth its own paragraph, because OSCOLA mandates it and Word will silently decide which one you get. The Match Groups page lists seven recognised shapes and its single-quote entry is the curly pair, U+2018 and U+2019. The exclusion-filters page lists eight shapes and its single-quote entry is the straight apostrophe, U+0027 twice. Same help site, same feature area, different character at the one position a law student cares about. I read both pages on 27 August 2026 and pulled the code points rather than eyeballing the glyphs, because at body-text size the two glyphs look identical.

Chasing the correct glyph is the wrong job, given that the two pages do not agree on what it is. The thing you can actually check is the file you are about to upload. A template swap, an export to PDF, a co-author opening the document in a different word processor: any of those can change the mark after the last time you looked at it, and nothing on screen announces that it happened. The exclusion-filters page also states outright that quotation marks separated from the text by a space are not recognised, and a stray space is precisely what survives a copy-paste out of a case report.

One more sentence deserves reading twice: the exclusion filters are settings. Somebody at your institution decides whether "exclude quoted text" is on for your assignment, and that decision does not appear on the percentage they send you. Why your quoted and cited text still shows up in the Similarity Report covers the four Match Groups and the recognition model in full.

The practical rule for a law essay is short. Quoted statute, quoted ratio and quoted dicta are not candidates for rewriting under any circumstances. If they matched, the answer is attribution and formatting, not language.

Strip out the quotations, the citation apparatus and the parts of the skeleton the marking criteria require, and what is left is the application step. In IRAC that is the A. In CLEO it is the E. The Touro page calls it "the heart of the discussion".

Concretely, these are moves that only work if you actually did the reading, which is why they read as yours:

1. Name the specific facts, not the category. Touro's page sets its own two versions side by side. The weak one has the officer noticing that the suspect "fit the description". The strong one gives the reason: "because Dan was extremely tall at 6'4", was wearing a green and tan sweater with purple patches and pointy-toed alligator cowboy boots, fitting the description provided by the eyewitness to the robbery." Same law, same facts, and only one of them is doing analysis.
2. Say why this authority and not the obvious one. A sentence explaining that you are relying on a first-instance decision because the appellate treatment turned on a point that is not in issue here cannot be produced by anyone who has not read both.
3. Put the counterargument in and lose it on the merits. A paragraph that runs the other side's best point and explains what defeats it has a shape that a recitation of the rule does not.
4. Say where the authority runs out. Hedged, specific uncertainty ("the point has not been decided above first instance in this jurisdiction, so what follows is an argument by analogy") is both honest legal writing and unlike anything in the rest of the essay.
5. Drop the sub-issues that go nowhere, in one sentence each. The instinct under exam conditions is to IRAC everything at equal length. Uniform-length treatment of unequal issues is exactly the flat pattern the vendor describes, and it spends the same wordage on issues that are not worth the same.

None of this is a promise about a percentage, and anyone who gives you one is guessing. It is a description of which paragraphs in a law essay are yours to write differently, and which are not.

Order of operations, in the order that avoids wasted work.

1. Check which report you are holding. The AI writing percentage and the similarity percentage are documented as "completely independent and do not influence each other". A law essay routinely has a high similarity figure for entirely proper reasons, and it says nothing about the other number. Turnitin AI Writing Report vs Similarity Report sorts this out in one pass.
2. Look at the highlights, not the number. Whether the highlighting sits in your application paragraphs or in your rule statements changes what you would do next, and the number does not tell you which.
3. Rule out the mechanical causes on the similarity side first. Quotation-mark shape, a space after an opening mark, footnote distance, whether the file was submitted as .docx or as a PDF. These cost minutes and can move a match between Match Groups without a word changing.
4. Do not touch quoted authority. Ever, for this purpose.
5. Then, and only then, look at the prose you actually wrote. That is the application, evaluation and conclusion material, and it is the only part where rewriting is legitimate at all.

If step five is going to happen anyway, the thing to protect is the citation apparatus. A law essay carries footnote numbering, cross-references such as "(n 27)", short-form case names, tables of authorities and pinpoint references, and all of them break quietly when text moves around them. HumanPen takes the file itself instead of pasted text, so a DOCX goes up and an editable DOCX comes back with structure, terminology, citations, layout and styles kept in place. Hand it a Turnitin or iThenticate AI report and it finds the flagged passages for you rather than making you mark them, then rewrites those and nothing else. Its smallest editable unit is one paragraph, and the expanded selection is put in front of you to approve before anything runs, which for a footnote-heavy document is the difference between re-proofreading four paragraphs and re-proofreading forty. When eligible, you can keep reducing AI for free.

Does Turnitin exclude quoted statutes and judgments from the AI writing percentage? Nothing published says so. Across the three pages that define the AI Writing Report, the words quotation, quote, footnote and citation do not appear at all. The only exclusion that is published for that report is the bibliography, added in an August 2023 release note. The exclusion filters that do handle quoted and cited material belong to the Similarity Report, and someone at your institution controls whether they are on.

Does OSCOLA get recognised as a citation format? The documented training set for citation recognition is "APA, MLA, Turabian, and IEEE numbered citations and references", and the guidance for anything else is that the report "may--but is less likely to--recognize those as citations". OSCOLA and Bluebook are not on the list. That is a statement about the Similarity Report's Match Groups, not about the AI percentage.

I did use an AI tool on part of the research. Is there a proper way to say so? Both systems have a rule for it now, and neither of them is a workaround. OSCOLA's fifth edition added §3.7.13, "Generative artificial intelligence": cite the model as the author, put the instructions you gave it in inverted commas, name the organisation that built it and give the date the material was generated. The manual's own example is `ChatGPT, response to 'Explain how artificial intelligence works', OpenAI (16 July 2023)`. The Bluebook added Rule 18.3 in its twenty-second edition in 2025, split across large language models, search results and AI-generated output, and it asks you to keep the output as a PDF; that one I know from the coverage when it came out rather than from reading the edition. Note what these rules are not: I searched all 85 pages of the OSCOLA PDF and `detect`, `Turnitin` and `plagiar` return zero. A citation manual tells you how to attribute a tool. Whether you were permitted to use it is your faculty's rule, and neither manual answers that.

My whole rule section is highlighted and I wrote every word. How? Sentences do not get scored one at a time. Turnitin's description is that a submission is cut into overlapping segments, each segment gets a value between 0 and 1, and each qualifying sentence "inherits the segment's score", with overlapping scores pooled. A long stretch written in one uniform register can come back as one continuous band. Turnitin flagged my whole methodology section walks through the same shape in a different discipline.

I checked one problem question on its own and got a terrible number. Short submissions behave differently in kind. The FAQ says that in documents of only a few hundred words "the prediction will be mostly 'all or nothing' because we're predicting on a single segment without the opportunity to overlap". A single 800-word problem answer is close to that regime. There is also a floor: no AI writing report is generated below 300 words of prose text.

Should I stop using IRAC? That is a question for your tutor and your marking criteria, not for a detector. What you can do without touching the skeleton is stop writing the application paragraphs in the same shape as the rule paragraphs, which is where the marks are anyway. Should You Change How You Write to Avoid Being Flagged? argues that question at length.