Good legal writers use AI as an amplifier, not an author
AI doesn't make you better by default. It magnifies discipline when guided wisely and exposes laziness when left unguarded. That's the line every legal writer has to walk.
For many senior lawyers, AI feels like long-earned relief. For Gen Z, it can feel like a shortcut that steals the climb. Both instincts are valid-and both can be productive if we set the right rules.
The split: speed vs. meaning
Older professionals see AI as another tool in a long chain-from typewriters to spreadsheets. Results matter. If the work is accurate and on time, good.
Many younger professionals were trained to avoid AI in school, then told to use it at work. That reversal feels like a values shift: effort used to equal virtue; now automation wins. Surveys show young adults are heavy users and still anxious about AI's impact on thinking and jobs-dependence mixed with distrust.
Pew Research Center reports track this tension, and it's shaping how new lawyers approach writing and research.
Why some people aren't impressed
Wonder requires comparison. If you've drafted dozens of wills, motions or briefs by hand, a strong AI-generated version feels like a minor miracle. If you've only seen fragments, it just looks like more text on a screen.
That gap drives the attitude divide. Veterans welcome relief; novices want the reps because the reps build judgment.
What AI changes-and what it can't replace
AI gives you clean sentences and quick structure. It doesn't give you judgment, fact sensitivity or the instincts built through hard edits and second thoughts.
There's another risk: endurance erosion. Perfect-looking first drafts can train shallow thinking. If you never fight through a paragraph, you never build the intuition to fix one.
Use this framework: where humans lead, where AI helps
- Human-only calls: issue selection, theory of the case, audience analysis, risk/privilege decisions, ethical boundaries, what facts and authorities actually control.
- AI-accelerated (with strict review): outlines, alternative framings, style/voice adjustments, first-pass summaries, analogies, headings, checklists, template generation.
- Never fully delegate: citations, quotations, jurisdiction-specific rules, numbers/dates/names, confidentiality redactions, settlement terms, anything that creates binding obligations.
A 30-minute "Socratic" drafting loop
- Define intent: purpose, audience, word count, must-include facts, banned arguments, tone.
- Feed substance: the record, controlling authorities, key quotes, contrary authority you must neutralize.
- Get a skeleton: ask for a bullet outline; reject fluff; force a tighter thesis and fewer points.
- Iterate: point out what's wrong, vague or irrelevant; demand alternatives; compress, then expand as needed.
- Source discipline: require pinpoint cites placeholders; flag every factual assertion for verification.
- Cold read: step away; read aloud; check logic, transitions, and client exposure.
- Red-team: ask AI to attack your argument; patch the holes; then you attack it again.
Prompts that show you're thinking
- "Draft a three-point thesis for [issue]. One sentence per point. No rhetoric. Show the logical chain."
- "Summarize [case] in 120 words: holding, rule, key fact, what distinguishes it from [contrary case]."
- "Give me two alternative framings that strengthen causation without overstating the record."
- "Rewrite this section for a time-pressed trial judge. 150 words. Short sentences. No adverbs."
- "List every factual claim in the draft that requires a cite. Output as bullets with placeholders."
Guardrails against polished nonsense
- Verify everything: run cases, quotes and stats through primary sources; never trust summaries.
- Citation integrity: check pincites, signal strength and jurisdiction; confirm definitions of terms of art.
- Fact hygiene: dates, amounts, parties, procedural posture-treat each like a landmine.
- Tone control: compare against your judge/client profile; cut confidence where the record is thin.
- Decision log: record key choices and why; this preserves authorship and helps later edits.
- Privacy: keep client data out of non-approved systems; use redactions or on-prem tools.
- Map your approach to a basic risk framework like the NIST AI RMF-lightweight, but consistent.
Train for judgment, not just speed
Keep the reps that build taste. For juniors, schedule short, "no-AI" drills weekly: write a tight rule paragraph, a case analog, a fact statement from a messy record. Then use AI to critique, not to generate.
For seniors, model the review process out loud. Show how you spot weak verbs, over-claims and logical gaps. Let the team see the moves behind the edits.
Policy your team can live with
- Permitted uses: outlines, alternatives, edits, summaries, style shifts.
- Prohibited uses: privileged facts in public tools, unsourced citations, final-form filings without human review.
- Attribution: log AI-assisted sections and verification steps in the file notes.
- Quality checks: mandatory cite check and fact check before partner or client review.
- Training: prompt fluency, verification, and risk basics are required, not optional.
- Audits: spot-audit samples monthly; celebrate clean processes, not just quick turnarounds.
The point that keeps you honest
Treat the machine as an amplifier, not an author. Use it to widen your range, not to replace your judgment. The future won't belong to whoever types fastest, but to whoever still knows why the words belong on the page.
Further learning
Your membership also unlocks:
