Artificial intelligence is making legal writing cheaper, faster and less embarrassing. Its first achievement won’t be a flowering of legal thought. It’ll be the elimination of drudgery: typos, ghostly cross-references, inconsistent uses of supposedly mandatory "shalls" and defined terms that mutate from section to section.

The slog of curing those problems has long subsidized law practice. A brief takes many days because lawyers research, check citations, compare authorities, revise prose and discover that the draft doesn’t say what they thought. A deal document takes time because lawyers redline precedents, reconcile provisions and hunt inconsistencies. Some of that time buys judgment. Much of it buys competence at tasks that a machine can perform in minutes. Clients are noticing.

This won’t eliminate lawyers’ work. It’ll expose what the work was always supposed to be: thinking.

The divide won’t be between lawyers who use AI and lawyers who don’t. AI will soon be routine. It’ll be between lawyers who use it as a dictation machine and lawyers who use it as an adversarial editor. The latter group will set constraints, supply authority, challenge assumptions, reject fluent nonsense and argue with the machine until the language is exact.

A few will accept the first draft because it sounds lawyerly. That’ll be malpractice by complacency: quieter than the old kind because nothing on the page will give it away.

The great cleanup

Let’s be candid: Most contracts aren’t finely calibrated instruments. They’re more like junk drawers with defined terms. Provisions copied from ancient forms sit beside new language. Temporal phrases conceal conditions. “Shall” commonly does the work of duty, futurity, entitlement and permission. Lawyers stick to forms not because they’re good but because they haven’t yet caused a memorable disaster.

Survival is hardly proof of clarity. It’s often proof of luck.

This is where AI’s advantage is hardest to dispute. Tell a system that every “shall” must impose a duty on the subject of its clause, and it can flag suspect usages throughout a 100-page agreement. Ask it to identify conditions introduced by “unless,” “subject to,” “upon” and “in the event of,” and it can apply the same test from beginning to end. It won’t tire at 11 p.m. It won’t miss page 87 because page 86 was tedious.

AI changes the economics of skepticism. It becomes feasible to ask of every clause: What work is this word doing? Is this a duty, condition, discretion or surplusage? Does the defined term mean the same thing everywhere? Does an exception swallow the rule? Does a limitation in section 12 undermine a remedy in section 47?

Those questions aren’t glamorous. That’s the point. Lawyers skip them because answering them manually can be exorbitantly expensive. AI makes systematic skepticism cheap. It needn’t understand the deal’s commercial logic to perform a first pass. It need only apply disciplined tests and flag what deserves human attention.

The litigation dividend—and risk

Litigation has more built-in discipline. Opposing counsel will attack the brief. A citation that doesn’t support its proposition can be exposed. A missing element can become the centerpiece of a reply. AI will still raise the floor: cleaner issue statements, tighter transitions, better organization and useful synthesis from a mass of research.

But fluency isn’t neutral in an adversarial system. It’s a weapon aimed wherever its user aims it. A polished brief can hide a weak argument from a hurried reader just as easily as sloppy prose can expose one. And AI’s signature failure—confident, well-formed fabrication—isn’t theoretical. Lawyers have already been sanctioned for nonexistent authorities.

The deeper danger is subtler. We’ve long treated bad writing as evidence of bad thinking. A confused sentence suggests a confused proposition. An incoherent structure suggests an incoherent argument. A mangled citation suggests inadequate checking. Those inferences have never been perfect, but they’ve been useful.

AI is already breaking that connection.

A lawyer who hasn’t thought a problem through can now produce a document that looks thoroughly considered. Clean prose, correct formatting, respectable citations and orderly structure will no longer tell us much. The machine can supply all four. A polished brief may become less a sign of competence than a mask for its absence.

That doesn’t make polish worthless. It means lack of polish will stop doing diagnostic work.

When every side can produce a competent summary of governing law, basic synthesis becomes cheap. The advantage shifts to whoever spots the procedural wrinkle, inconvenient fact, overlooked statutory phrase, adverse precedent that controls or distinction that changes everything.

Knowing the law is getting cheaper. Knowing what to do with it isn’t.

The new craft

Legal writing instruction has spent decades attacking visible defects: passive voice, throat-clearing, dangling modifiers, inflated diction, unnecessary repetition. Those lessons remain sound. But AI will increasingly enforce them automatically. Avoiding obvious defects will cease to distinguish a good lawyer from an ordinary one.

The scarcer skills will be judgment and tenacious questioning.

What must this document accomplish? Which authorities control? Which facts can be conceded? Which must be developed? What proposition should remain narrow? What risk is the client actually allocating? Which arguments should be abandoned, even though they’re available?

Those aren’t computer questions. They’re legal questions.

The best litigators won’t ask for “a motion to dismiss.” They’ll ask the system to map each element against pleaded facts, identify the strongest contrary authorities, strip argumentative words from the statement of facts, test jurisdictional assumptions, and propose the narrowest viable ground for dismissal. Then they’ll attack the result.

They won’t ask, “Can you make this better?” They’ll ask, “What’s wrong with this?”

Transactional lawyers will work similarly—perhaps more urgently—because transactional drafting still often begins with precedent, rather than first principles. A lawyer won’t merely request a confidentiality agreement. The instruction will specify the business risk, parties, information, permitted uses, fitting remedies and obligations that survive termination.

Then the system will be made to attack the draft—revealing unworkable definitions, incompatible provisions, obligations masquerading as recitals, variable wordings for the same idea, remedies that collide with liability caps, provisions that seem to allocate risk without doing so.

That last pass is often skipped—not because lawyers are careless but because exhaustive review of a long agreement is tedious enough that it rarely happens twice. AI makes it cheap enough to do every time.

That changes the economics of carelessness. It makes some sloppiness a choice as opposed to a cost of doing business. The machine will be quick. The lawyer will have to be exact.

The end of the mask

AI won’t make every lawyer good. It’ll make it harder to tell, at a glance, who isn’t.

The old tells will fade. Typos will disappear. Tables of contents will behave. Cross-references will work. Defined terms will be checked. “Shall” will stop doing five unrelated jobs in a single document. A lawyer who hasn’t thought things through will no longer advertise that fact in the grammar and layout.

Universal polish won’t necessarily reward better judgment. It may simply make bad judgment harder to detect. Clients, courts, partners and hiring committees have relied on visible defects as crude proxies for competence. The proxies were imperfect. But they were something.

Now, the machine can remove the evidence.

The result could be salutary. We may hope that lawyers will spend less time manufacturing documents and more time confronting the problems those documents are supposed to solve. But there’s a darker possibility: AI may give mediocre lawyers an unprecedented ability to conceal mediocrity.

So the central question isn’t whether AI will write better legal prose. It almost certainly will. The question is whether lawyers will use that prose to express better thought—or to disguise its absence.

Legal documents aren’t literature. They’re instruments. A contract allocates risk. A motion seeks an order. A brief frames a dispute. A legal opinion tells a client what can safely be done. Elegance matters less than legal function.

For centuries, legal writing has mixed substance with debris. AI will strip away much of the debris. But once that’s gone, the profession will face an uncomfortable test: What remains?

If every lawyer can produce clean prose, clean prose will cease to impress. If every lawyer can generate a plausible first draft, the first draft is cheapened. And if every lawyer can produce a competent summary of the law, well, you get the idea.

The real work will be judgment.

And judgment doesn’t announce itself with good grammar. It appears in the choice of argument, the recognition of controlling authority, the treatment of inconvenient facts, the allocation of risk, the refusal to make an unnecessary concession, the decision to say less, and the ability to see what everyone else has missed.

AI may bring about the end of bad legal writing. But it won’t bring about the end of bad lawyering. It may make bad lawyering look like impeccable judgment.

Bryan A. Garner. (Photo by Karolyne H.C. Garner)

Bryan A. Garner is the author of The Winning Brief, Garner’s Modern English Usage and Legal Writing in Plain English.