gabrielmoreira Persuasive Legal Writing
This skill distills the techniques that separate serviceable legal writing from writing that actually moves people — judges, tribunals, opposing counsel, clients. The principles come from close study of Justice Kagan's opinions and dissents, the advocacy of David Boies and Ted Olson, and the analytical framework of Ross Guberman. They apply to any legal document that needs to persuade.
The skill is organised into two parts: Architecture (how to structure and sequence your arguments) and Craft (how to write the sentences and paragraphs themselves). Use both. Architecture without craft produces outlines; craft without architecture produces beautiful paragraphs that don't land.
Part One: Architecture
Open with a Frame, Not a Summary
The opening paragraph is the most important paragraph in any legal document. Its job is not to summarise — it is to frame. The reader should finish your opening having already absorbed your conclusion, even before you argue for it.
There are two proven approaches, and which one you choose depends on your role:
The Syllogistic Opening (for opinions, submissions, and analytical documents)
State the issue, state the opposing position, and then state the principle that defeats it — all in two or three sentences. The reader experiences the conclusion as inevitable. Kagan does this in Florida v. Harris: issue, lower court holding, governing standard — done. The holding sounds unreasonable before she has argued a word against it.
When drafting: write your opening last, after you know your strongest argument. Then compress that argument into a frame that makes the other side's position sound untenable on its face.
The Binary Opening (for advocacy and persuasive correspondence)
Force the reader to choose between two starkly different characterisations of the dispute. Boies and Olson do this by juxtaposing "hypothesized uncertainty" against deprivation of a constitutional right "for years." The imbalance is baked into the framing before any argument begins.
When drafting: identify the strongest contrast between what your client suffers and what the other side claims. Compress it into a single sentence. Make the reader feel the asymmetry.
Common pitfalls to avoid:
- Starting with procedural history. Nobody cares about the procedural history in paragraph one. Move it later.
- Starting with "This case concerns..." followed by a neutral restatement of both sides. That is summarising, not framing. You are allowed to have a point of view. State it.
- Front-loading caveats and qualifications. Confidence first; nuance later.
Sequence Arguments by Persuasive Force, Not by Logic
Many lawyers organise arguments in the order they occurred to them, or in the order the legal elements appear in a statute. This is almost never the most persuasive sequence.
Instead:
Lead with your most powerful argument. The one that, if accepted, wins the case outright. Put it first, develop it fully, and make it feel dispositive before moving on.
Place your second argument so it feels like independent confirmation. The reader should think: "Even if I wasn't convinced by the first argument, this one gets there too."
End the argument section with a forward-looking consequence. Show the reader what happens if they rule against you. Make it concrete and alarming. Kagan does this in her Winn dissent — she shows that the majority's rule gives government a simple playbook to "end-run" judicial review of religious funding. Boies and Olson end by reframing the entire case around the limits of democratic power.
Use Structural Signposting That Feels Conversational
Guide the reader through your argument, but do it in a way that sounds like you are talking to them, not filing a document at them. Kagan's technique in Kloeckner is the model: "Begin with...", "Turn next to...", "Now just put them together." The reader follows willingly because the transitions feel natural.
Avoid mechanical transitions like "Turning to the second element" or "With respect to the third factor." These signal that you are checking boxes, not building an argument.
Close with a Parting Thought, Not a Formulaic Conclusion
Most legal writers end with "For the foregoing reasons, the Claimant respectfully requests..." This is a wasted opportunity. The ending is what the reader carries away.
End your argument section — the last substantive paragraph before any formal prayer for relief — with a sentence that reframes your entire case at its highest level of generality. Make it the sentence the judge remembers when summarising your position to colleagues.
Part Two: Craft
Razor-Sharp Clarity
Every sentence should do exactly one job. If a sentence is doing two jobs, split it. If it is doing no job, cut it.
Specific techniques:
Short sentences for conclusions; longer sentences for reasoning. The punch lands harder when the sentence is short. "That is not a remotely sufficient justification" hits harder than a compound sentence that buries the same point.
Address the reader directly when explaining complex statutory or regulatory provisions. "Begin with regulation X, which governs..." is clearer and more engaging than "Regulation X provides that..." This is Kagan's technique and it works because it converts passive exposition into guided reasoning.
Use syllogism to make the conclusion feel inevitable. State premise one. State premise two. State the conclusion with "Ergo" or "It follows that" or simply present it as obvious. Kagan does this explicitly in Kloeckner and implicitly in most of her openings. The reader feels they reached the conclusion themselves.
Vivid, Concrete Examples
Abstract legal principles do not persuade. Concrete examples do. There are three types of example, and each serves a different purpose:
The Narrative Hypothetical
Build a short story that makes the reader feel the principle before you name it. Kagan's binoculars hypothetical in Jardines is the model: a stranger stands on your porch with super-powered binoculars and peers into your home. You feel the invasion viscerally. Only then does she say: "Has your visitor trespassed? Yes. Has he invaded your reasonable expectation of privacy? Yes, of course."
The technique: describe the scene, let the reader react emotionally, then name the legal conclusion. The conclusion feels obvious because the reader already reached it.
The Recruited-Intuition Analogy
Find an analogous situation where the reader's intuition is already settled, then pivot: if you agree with me there, you must agree with me here. Kagan's bailout analogy in Winn is the model — nobody would be fooled by relabelling an appropriation as a tax break. Boies and Olson's "nationals of Chinese descent" analogy works the same way.
The technique: choose an analogy where (a) the answer is obvious and (b) the structural parallel to your case is tight. If the analogy is loose, it will backfire.
The Parade of Horribles
Stack concrete examples that show the absurd consequences of the opposing position. Boies and Olson: if you accept this argument, you must also accept that the State can prohibit marriage by senior citizens, by felons in separate prisons, or by couples using contraception. The accumulation makes the opposing position look untenable.
The technique: pick three examples, each more uncomfortable than the last. Use parallel construction so the list builds rhythmically.
Parallel Construction
Parallelism is not decoration — it is a structural tool that makes complex arguments scannable and contrasts impossible to miss.
Streamline lists. When you have multiple independent grounds for a legal conclusion, cast them in parallel grammatical form. Each item should start with the same part of speech and follow the same syntactic pattern. This makes a complex, multi-ground argument feel organised rather than sprawling.
Draw contrasts through antithesis. When you want to expose an asymmetry or double standard, use parallel structure to make the reader see it. The Boies/Olson sentence about heterosexual individuals being "authorized" while gay and lesbian individuals are "prohibited" works because the parallel structure itself enacts the inequality. The sentence does not argue that there is a double standard — it shows one.
The test: read your parallel construction aloud. If the rhythm breaks, the parallelism is off. Fix the grammar until it flows.
Quote Sparingly and Strategically
Judges complain that lawyers quote too much and too often. The solution is not to stop quoting — it is to quote differently.
Merge short quoted snippets into your own sentences about your own dispute. Do not drop a block quote and expect it to argue for you. Instead, weave two or three key words from the authority into a sentence that applies the principle to the facts at hand. Boies and Olson: "the inevitable inference that the disadvantage imposed on gay and lesbian individuals by Prop. 8 is born of nothing more than naked animosity." The quoted fragments ("inevitable inference," "born of," "animosity") come from Romer, but the sentence is about Prop. 8.
Quote your opponent's words back at them. When the other side uses language that, in context, undermines their own position, quote those exact words and reframe them. Boies and Olson quote the Intervenors' phrase "official legal promotion" and then show that it is a euphemism for denying a fundamental right. The reader sees through the euphemism because you showed them the real language underneath.
Reserve block quotes for language that is genuinely devastating. If the quoted passage does not make the reader's eyes widen, paraphrase it instead.
Voice: Personal, Warm, and Grounded
Persuasive legal writing sounds like a person talking, not an institution filing. This does not mean being casual — it means being human.
Word choice matters. Prefer concrete, Anglo-Saxon words over abstract Latinate ones. "Chutzpah" over "audacity." "Topsy-turvy" over "inverted." "Fool's errand" over "futile endeavour." These are Kagan's choices and they work because they are vivid and unpretentious.
Acknowledge the reader's experience. When you have just taken the reader through a complex statutory analysis, it is fine to say: "If you need to take a deep breath after all that, you're not alone." This is Kagan in Kloeckner. It builds rapport and signals that the complexity is the statute's fault, not yours.
Pivot from logic to consequences. After you have made the doctrinal argument, show the reader what happens in the real world if the other side wins. Kagan does this in every major dissent — she zeroes in on the practical harm. This is where you stop being a lawyer and start being an advocate.
Use dialogue with the reader. Ask a question, then answer it. Kagan's Jardines concurrence is built on this: "Has your visitor trespassed? Yes, he has. And has he invaded your reasonable expectation of privacy? Yes, of course, he has done that too." The question-and-answer rhythm pulls the reader into the argument as a participant, not a spectator.
Part Three: Avoiding AI-Sounding Legal Prose
Even when applying all the techniques above, legal writing can still sound machine-generated if it falls into certain structural patterns that real lawyers do not use. These patterns are specific to legal and analytical writing — they go beyond the general AI tells (like "delve" or "tapestry") and infect otherwise competent prose. They all share a common flaw: the writer steps outside the argument to comment on the argument, announcing its structure, counting its parts, labelling its method, or disclaiming its scope. Real legal writing stays inside the argument. The reader should experience the reasoning, not a guided tour of the reasoning.
Do not announce the structure of your own argument
Bad: "The argument is therefore limited in four respects. First, it does not romanticise search. Secondly, it does not suggest that all curation is manipulative or unlawful. Thirdly, it does not claim that existing law is helpless. And it does not treat 'source plurality' as a ready-made legal standard."
This reads like a numbered list disguised as a paragraph. No one talks like this. Real lawyers make their points without counting them out loud. If you need to state limitations, weave them into the argument naturally or just state them without the inventory system. You might instead write: "The argument does not romanticise search, and it does not suggest that all curation is manipulative. Existing law has tools — after April 2025, broader and sharper ones. 'Source plurality' is useful as a diagnostic, but it is not a rule."
Do not use meta-commentary to signpost, preview, or narrate the paper's own strategy
Bad: "The continuity point comes first."
Also bad: "The contrast deserves to be worked through."
Also bad: "Section 227 matters in a complementary way."
Also bad: "But if the paper is organised around hallucination alone, it reduces the problem to ordinary inaccuracy."
These all do the same thing: they talk about the paper instead of making the argument. "The continuity point comes first" tells the reader you are about to make a point instead of making it. "The contrast deserves to be worked through" announces that something interesting is coming without delivering it. "Section 227 matters in a complementary way" is throat-clearing with no content — just say what section 227 does. And a paper should never refer to itself in the third person ("if the paper is organised around..."). If you are writing the paper, just make the argument; do not step outside it to discuss editorial choices with the reader.
The same applies to sentences that narrate the paper's internal architecture:
Bad: "Even so, claims about preference formation and dependency can quickly become paternalistic or empirically loose. They are better treated here as explanatory mechanisms that intensify the three primary harms rather than as the core legal wrongs."
This is the AI explaining to the reader how it has categorised its own arguments — which concepts are "explanatory mechanisms" and which are "core legal wrongs." A real person would just use the concepts where they belong without announcing their structural role. If preference formation intensifies the harms, say so when discussing the harms. Do not give the reader an organisational chart of the argument.
Bad: "The doctrinal fit is strongest where there is a trader, a commercial practice, and a transactional decision in the statutory sense."
This is the AI evaluating the strength of its own argument rather than making it. A real lawyer would apply the law to the facts: "The Act applies where there is a trader, a commercial practice, and a transactional decision. All three are present here." Do not rate your own doctrinal fit — demonstrate it.
Do not write axiomatic one-liners that sound like logic textbook entries
Bad: "If the baseline is romanticised, the argument built on it fails."
This is too compact, too self-contained, too pleased with itself. It sounds like a theorem, not like a person reasoning. Expand it, ground it in the specific facts, or fold it into the surrounding argument so it does not sit there like an epigram. You might instead write: "The problem with romanticising the baseline is that everything built on it inherits the distortion." Better still, show the distortion rather than announcing it.
Do not use the "Nor does it... It does, however..." negation scaffold
Bad: "That study does not prove that answer systems generally displace search. Nor does it establish dependency, closure, or any settled behavioural tendency across contexts. It does, however, support a narrower point: if..."
This mechanical pattern — stacking negations and then pivoting with "It does, however" — is one of the most recognisable AI constructions in legal writing. A real lawyer would write something like: "This does not mean that answer systems generally displace search. But it does support a narrower point..." The fix is usually to flatten the structure: say what the thing does not show, then say what it does show, without the formal three-part pivot of negation, further negation, concession.
Do not use pedantic throat-clearing before definitions or reframings
Bad: "For present purposes, 'knowledge' need not be treated philosophically. The more useful formulation is practical:"
This comes across as condescending — as if the writer is worried the reader might start doing epistemology unless warned off. It also sounds like a lecture, and an almost arrogant one. If you need to define a term practically, just define it: "What matters here is whether the user can find out X" or simply use the term in context and let the usage do the defining. Do not announce that you are being practical instead of philosophical. The reader can tell.
Other AI patterns to watch for in legal writing
Formulaic conditional framing. Sentences like "If X, then Y follows" stated as standalone axioms rather than woven into the argument. Real reasoning develops; it does not announce logical rules and then apply them.
Symmetrical negation pairs. "It is not X. It is Y." is fine once. When the same structure appears repeatedly across a document — "This is not a question of A. It is a question of B" in one paragraph, "The issue is not C. The issue is D" three paragraphs later — it becomes a formula the reader can see through.
Overly tidy limitation lists. When you find yourself writing "This argument does not claim X, does not suggest Y, and does not assume Z," ask whether you actually need to disclaim all of those things. Often the disclaimer is defensive rather than useful. State your actual claim clearly enough that the reader does not need to be told what it is not.
AI tell-tale phrases in legal writing
Beyond structural patterns, certain phrases and word choices are dead giveaways of AI authorship even when they appear in a single sentence. These are not about structure — they are about diction and register. A human would never choose these words in this way.
Narrating your own language choices.
Bad: "The phrase is useful as a title. In what follows, plainer language is preferable: answer-based intermediation, synthetic responses, and the conditions of comparison and verification."
Nobody announces that they are about to switch to plainer language. If plainer language is better, just use it. The reader does not need a commentary track explaining your editorial decisions. Write the plain version and move on.
Forced or precious idioms.
Bad: "earns its keep, but only if used carefully"
AI reaches for colloquial idioms to sound human, but the effect is the opposite — it sounds like someone performing casualness rather than being casual. A real person would write "plays a very important role" or "matters here" or just say what the thing does. The test: if the idiom draws attention to itself rather than to the point being made, cut it and use ordinary language.
Other AI-favourite idioms to avoid: "does the heavy lifting," "the devil is in the detail," "cuts both ways," "moves the needle," "worth its weight," "punches above its weight." These are not wrong — they are just the idioms AI gravitates toward, and readers have started to recognise them as tells. Watch especially for idioms combined with meta-commentary about the paper's own strategy, e.g. "Personalisation sharpens the concern but should not do the heavy doctrinal lifting in this paper." That sentence is doing nothing except narrating the paper's architecture using a forced idiom. Cut it entirely.
The "narrower concern" retreat.
Bad: "Every interface steers in some sense; that, by itself, proves nothing. The narrower concern is steering whose operation or commercial logic is not reasonably apparent to the user."
Also bad: "The point is not that all such design is unlawful. The concern is narrower: answer-based steering is often experienced as assistance rather than intervention, and is therefore harder for consumers to discipline through ordinary judgment."
This is one of the most characteristic AI moves in analytical legal writing: state a broad proposition, then immediately retreat into "the narrower concern is..." or "the concern is narrower." It makes the writer sound like they are perpetually backing away from their own point. A real person states the concern directly without the hedging preamble. The first example should read: "Every interface steers consumers in some sense; the concern here is steering whose operation or commercial logic is not reasonably apparent to the user." The second should read: "The point is not that all such design is unlawful. Rather, it is that answer-based steering is often experienced as assistance rather than intervention, and is therefore harder for consumers to be aware that they are being steered."
The fix is usually simple: drop "the narrower concern is" and just state the concern. If you need to distinguish your point from a broader one, do it with "rather" or "but" or "what matters here is" — not by announcing that you are narrowing.
Evaluative grading of concepts before engaging with them.
Bad: "That is a useful diagnostic idea. It would be a poor legal rule if left at that level of generality."
Also bad: "The Explanatory Notes are useful because they make clear that the test is objective."
This is the AI as professor, marking the work before responding to it. Real lawyers do not rate concepts as "useful" or "poor" in this detached, evaluative way — they engage with them directly. The second example is especially revealing: the writer is explaining to the reader (and to itself) why it is citing the Explanatory Notes, rather than just citing them. A real person would write: "The Explanatory Notes make clear that the test is objective." Full stop. The word "useful" and the "because" clause are the AI's internal reasoning leaking onto the page.
Instead of telling the reader a concept is useful, show why it matters. Instead of calling it a poor legal rule, explain what goes wrong if you try to apply it. The grading language ("useful," "valuable," "helpful," "important diagnostic," "a poor X if...") is a consistent AI pattern that signals the writer is commenting from outside the argument rather than reasoning from inside it.
Vague attribution to unnamed positions or sources.
Bad: "That maps onto answer systems more directly than some accounts assume."
What accounts? Who assumes? This is the AI gesturing at a broader scholarly conversation that may or may not exist, to make the argument sound more situated than it is. A real person would either name the accounts they are distinguishing from, or just make the point without the comparative: "That applies directly to answer systems." If you cannot name the "some accounts," you probably do not need to reference them.
Hollow axioms that sound authoritative but say nothing concrete.
Bad: "It is not a free-standing fairness charter."
Bad: "though it must be kept in its place"
These are the AI producing sentences that have the cadence and confidence of a real legal point but contain no actual content. What does it mean to say a statutory provision is "not a free-standing fairness charter"? What place must the concept be "kept in"? These phrases sound like they are doing work, but if you try to extract a concrete legal proposition from them, there is nothing there. They are rhetorical filler dressed up as restraint.
The test: can you state what the sentence actually means in plain terms? If the answer is just "this provision does not do everything" or "do not rely on this too heavily," either say that directly or cut the sentence entirely.
Self-assessing the "right use" or scope of your own argument.
Bad: "The right use of section 229 here is modest. It supports the proposition that answer architecture can matter even when conduct does not fit neatly into a single misleading statement. It does not license a roving standard of informational virtue."
This is several AI patterns at once. "The right use... is modest" is the writer evaluating their own argument's ambition. "It supports the proposition that..." is the writer narrating what a statutory section does within the paper rather than applying it. And "it does not license a roving standard of informational virtue" is a grandiose negative — the AI producing an impressive-sounding phrase ("roving standard of informational virtue") to describe something it is not arguing, which no real person would think needed disclaiming. The whole passage is a lot of words that add up to very little.
If section 229 supports a point, make that point and cite section 229. Do not write a three-sentence preamble explaining how modestly you intend to use it and what you are not claiming. That is the AI being cautious and performatively restrained — qualities that feel responsible but read as obviously machine-generated.
Patronising, disciplinary tone toward legal concepts.
Bad: "though it must be kept in its place"
Bad: "earns its keep, but only if used carefully"
Bad: "The right use of section 229 here is modest."
These all treat legal provisions or concepts as things that need to be disciplined or managed by the writer — kept in their place, made to earn their keep, used modestly. Real lawyers do not adopt this schoolteacher posture toward the law. They apply it or they don't. If a provision has limited relevance, say what it does and move on. Do not talk about it as if it might misbehave if you are not careful.
Labelling your own rhetorical devices.
Bad: "That question is not rhetorical; it is the test that section 229 imposes."
If you ask a question and it is the statutory test, the reader will understand that from context. You do not need to tell them you are not being rhetorical. Explaining your own rhetorical choices — "that was not rhetorical," "the analogy is deliberate," "the point is not academic" — is the writer stepping outside the text to annotate it. Let the writing speak for itself.
Disclaiming your own analogies.
Bad: "The analogy should not be pressed too far. Answer systems are not review publishers. Even so, the fake-reviews regime is important because it shows that..."
This is the AI hedging before and after every analogy, afraid of being criticised for overreaching. A real person would fold the qualification into the sentence naturally: "While answer systems are not review publishers, the fake-reviews regime is relevant as it shows that..." One sentence, not three. State the qualification and the point together. Do not build a little retractable bridge every time you draw a comparison.
Scoping paragraphs that exist only to acknowledge limits.
Bad: "The limit of the existing framework should also be squarely stated. Chapter 1 of Part 4 is strongest where the answer system is tied to a trader's commercial practice and a sufficiently proximate transactional decision. It is weaker where the complaint is broader... That does not mean the law says nothing. It does mean the fit is thinner, and that thinner fit is better acknowledged openly than buried in caveats."
This entire paragraph adds nothing. It is the AI performing intellectual honesty — showing the reader that it knows its argument has limits — without advancing the argument. If the law's fit is thinner in some contexts, that will become apparent when you apply the law. You do not need a standalone paragraph to announce that you are about to be honest about limitations. Cut these paragraphs entirely or, if the limitation is genuinely important, state it in a single sentence where it naturally arises.
The paper talking about itself — its format, its timing, its title.
Bad: "Data protection law is relevant, but not central, to this paper."
Bad: "For a July 2026 conference paper, that makes the regime imminent rather than fully embedded."
Bad: "The role of competition policy should be kept narrower than the title might tempt one to make it."
These are among the most obvious AI tells in academic legal writing. The paper should never refer to itself as "this paper," discuss what is "central" to it versus peripheral, comment on its own timing ("for a July 2026 conference paper"), or warn the reader about what its title might "tempt" them to expect. A real author simply writes about the subject. If data protection is not the focus, do not discuss it at length — the reader will infer its role from how much space you give it. If a regime is not yet in force, say when it comes into force and move on. If competition policy has a narrow role, give it a narrow treatment. The reader does not need to be told that you are keeping things narrow; they can see the narrowness from the page.
The underlying problem is the same one that runs through all of Part Three: the AI steps outside the argument to manage the reader's expectations about the argument. Real writers stay inside.
The impersonal "one" construction to describe the state of the law or the argument.
Bad: "The consequence is that one can no longer write as if UK consumer law were still a weak, court-dependent disclosure regime."
"One can no longer write as if" is a very AI way to describe a legal development. A real person would say what the law now does: "UK consumer law is no longer a weak disclosure regime" or "Since April 2025, UK consumer law has broader enforcement tools than the old court-dependent model." State the change directly; do not describe it as a shift in what "one" can or cannot "write."
Other tell-tale phrases to watch for:
- "In what follows" — nobody writes this in legal documents
- "The more useful formulation is..." — evaluative and lecturing
- "For present purposes" — almost always unnecessary padding
- "It is worth noting that" — just note the thing
- "To be clear" — if you need to say this, the preceding sentence was not clear enough; fix that sentence instead
- "This is not to say that" — defensive and AI-characteristic; rephrase as a direct statement of what you do mean
- "The contrast deserves to be worked through" — say nothing; just work through it
- "X matters in a complementary way" — say what X does, not that it "matters"
- "They are better treated here as..." — narrating your own editorial choices
- "The doctrinal fit is strongest where..." — evaluating your own argument's strength rather than making it
- "maps onto X more directly than some accounts assume" — vague attribution; name the accounts or drop the comparison
- "It is not a free-standing X" — hollow axiom; say what it actually does or does not do
- "though it must be kept in its place" — patronising; just state the limits plainly
- "The right use of X here is modest" — self-assessing your own argument's ambition
- "It does not license a roving standard of..." — grandiose negative disclaiming something nobody was arguing for
- "X is useful because..." — AI showing its internal reasoning for citing something; just cite it and state the point
- "That question is not rhetorical" — labelling your own rhetorical device
- "The analogy should not be pressed too far" — disclaiming before or after every comparison; fold the qualification into the sentence
- "X is relevant, but not central, to this paper" — the paper scoping itself
- "For a [date] conference paper, that makes..." — the paper situating itself temporally
- "The role of X should be kept narrower than the title might tempt one to make it" — the paper discussing its own title
- "the decisive question is functional" — axiomatic labelling; just ask the question or make the point
- "that thinner fit is better acknowledged openly than buried in caveats" — AI performing intellectual honesty rather than just being honest
Applying These Techniques
When you are asked to draft or improve a legal document, work through this checklist:
Opening: Does the first paragraph frame the dispute, or merely summarise it? Rewrite it until the reader absorbs your conclusion before your argument begins.
Sequence: Is the strongest argument first? Does the argument section end with a consequence, not a whimper?
Examples: Is every abstract principle grounded in at least one concrete example, analogy, or hypothetical? If not, add one.
Parallelism: Are lists grammatically parallel? Are key contrasts structured as antithesis?
Quotations: Are you quoting snippets merged into your own sentences, or dropping block quotes? Restructure any block quote that is not genuinely devastating.
Voice: Read the document aloud. Does it sound like a person or an institution? Warm up the diction. Add one moment of direct address or acknowledged complexity.
Ending: Does the last substantive paragraph reframe the case at its highest level? If it starts with "For the foregoing reasons," rewrite it.
Repetition audit: After making all edits, re-read the document as a whole and check whether the same word, phrase, or syntactic construction has been introduced more than once. This is the most common failure mode when applying these techniques across a document: a good fix in one paragraph becomes a tic when the same fix appears three paragraphs later. Specific traps to watch for:
- Using the same lead-in phrase for different points (e.g., "The claim is specific:" in one paragraph and "They are specific:" in the next).
- Overusing a single rhetorical structure. "It is not X. It is Y." is powerful once or twice. Five times in the same paper and the reader notices the formula rather than the argument.
- Replacing different hedging phrases with the same confident alternative. If you cut "more modest" in one place and "more concrete" in another, vary the substitutions rather than reaching for the same word. The rule: every edit must be tested not only against the sentence it changes, but against every other edit in the same document. Read the document once more after all edits are complete, looking specifically for patterns you have inadvertently created.
AI-prose audit: After all substantive edits, read the document one final time looking specifically for the patterns in Part Three. Check for:
- Structural tells: numbered limitation lists disguised as paragraphs, meta-commentary announcing structure or narrating the paper's strategy ("The point comes first," "The contrast deserves to be worked through," "They are better treated here as...," "The doctrinal fit is strongest where..."), the paper referring to itself in the third person, standalone axioms ("If X, then Y fails"), the "Nor... It does, however..." scaffold, pedantic throat-clearing before definitions.
- Phrase-level tells: narrating your own language choices ("In what follows"), forced idioms ("earns its keep," "does the heavy lifting"), the "narrower concern" retreat ("the concern is narrower:"), evaluative grading of concepts ("That is a useful diagnostic idea"), meta-commentary about the paper's own strategy ("should not do the heavy doctrinal lifting in this paper"), and the specific phrases listed in the tell-tale phrases section ("for present purposes," "it is worth noting that," "to be clear," "this is not to say that").
- Self-referential tells: the paper talking about itself ("this paper," "central to this paper"), situating itself temporally ("for a July 2026 conference paper"), discussing its own title, labelling its own rhetorical devices ("that question is not rhetorical"), disclaiming its own analogies ("should not be pressed too far"), and standalone scoping paragraphs that exist only to acknowledge the argument's limits. The most important single principle: if a sentence or paragraph is about the paper rather than about the subject, cut it. These patterns survive other editing passes because they feel "organised" or "precise" — but organised is not the same as human, and precision is not the same as natural.
This is not a mechanical checklist to apply rigidly — it is a set of lenses to look through when revising. The best legal writing makes all of these things feel effortless and invisible. That only happens through revision.