home  /  productivity & workflows  /  drafting
practice · ai for legal practice

Drafting.

The output reads well immediately. That is the benefit and the hazard, and they arrive together.

begin here

Where is your firm?

Start a conversation with the AI Adoption Concierge, already scoped to drafting. Pick a starting point, or describe your situation directly.

AI Adoption Conciergedrafting · orientation, not legal or ethics advice
Tell me what your firm drafts most often and whether you have good precedent for it. I'll help you design the workflow — precedent-grounded beats generated, nearly always.

Drafting is where most firms feel the gain first, because the change is immediate and visible: a first draft of a routine document appears in a fraction of the time it took to produce. The saving is real and it is concentrated at a specific point — the blank page, which is where a surprising share of drafting time actually goes. The hazard is specific too. Generated prose is fluent by construction, and fluency is what reviewers use as a proxy for correctness. A draft that reads like a competent lawyer wrote it receives lighter scrutiny than a rough one, exactly when the errors are subtle: a defined term used inconsistently, a standard clause that does not fit this deal, a provision that reads sensibly and allocates risk the wrong way.

mechanisms

What drafting assistance is good and bad at.

Strongest at the top, weakest at the bottom — and the bottom is where the money is.

Structure and boilerplate

Getting a conventional document shaped correctly. Reliable and a genuine time saver.

First drafts of routine documents

Standard agreements, letters, routine filings. Large saving where the form is well-established.

Rewriting for tone or length

Turning something dense into something a client can read. Consistently good.

Consistency across a long document

Defined terms and cross-references drift. Needs a deliberate pass, and reviewers rarely make one.

Deal-specific provisions

It will produce something conventional. Conventional is often wrong for this deal.

Risk allocation

The commercial substance. This is the lawyer's judgement and it does not transfer.

methodology

What the evidence shows — and what we examine.

How firms make drafting assistance work.

Ground it in your own precedentSupplying the firm's form and having it adapted beats generation from nothing, by a wide margin.
A structured review passDefined terms, cross-references, party names, dates — checked deliberately, because fluent text hides these.
Flag the substantive clausesMark which provisions carry the commercial risk and review those as if drafted by a stranger.
Capture what workedA prompt that produced a good draft is firm knowledge. Save it or lose it.
what's at stake

What drafting practice decides.

Whether the saved time turns into capacity or into a correction later.

realised time saving internal document consistency whether risk is allocated correctly consistency across the team what juniors learn about drafting exposure on executed documents

Fluent drafts get lighter review.

Reviewers use polish as a proxy for care. Generated text is polished before it is correct, which inverts the signal — and the errors that survive are the quiet ones about substance.

common questions

Drafting — practical questions.

Is it better to generate from scratch or adapt a precedent?

Adapt a precedent, in almost every case. Supplying your own form and asking for it to be adapted to these facts keeps the firm's positions, its risk allocations and its accumulated judgement in the document, and it narrows the task to something the tool is genuinely good at. Generating from nothing produces a document reflecting whatever was conventional in the training data, which is a different firm's risk appetite at best. The gap in output quality between these two approaches is larger than the gap between most tools.

What review does a generated draft need?

The same review as a junior's draft, plus a specific mechanical pass. The mechanical pass — defined terms used consistently, cross-references pointing at real provisions, party names and dates correct throughout — matters because these are exactly the errors fluent text conceals and human drafters make less often. Then substantive review on the clauses that carry the commercial risk, done as though a stranger drafted it. The failure pattern is a reviewer who reads for quality of prose, finds it good, and stops.

Should clients be told a draft was AI-assisted?

Practice varies and the safe reading is that it depends on the engagement, the jurisdiction and what the client has been told previously. The relevant obligations are around communication and fees rather than any specific AI rule: a client is generally entitled to know material facts about how their matter is being handled, and if AI use materially affects the work or the bill, that starts to look material. Some firms address it once in the engagement letter, which is cleaner than deciding matter by matter. Check what your own jurisdiction has said.

How do we stop everyone drafting differently?

Treat prompts as firm knowledge rather than personal technique, because otherwise you get as many drafting standards as you have lawyers. When someone finds an approach that produces good drafts of a document type the firm produces often, that belongs in a shared library with the firm precedent it works against. This is unglamorous and it is where the durable gain lives — individual experimentation produces individual gains that leave when the person does.

related

Related specialization areas & resources.

Make drafting assistance actually pay.

Describe what your firm drafts most often. The Institute will help you build the workflow.

AI adoption conciergeorientation · not legal or ethics advice
Tell me what your firm drafts most often and whether you have good precedent for it. I'll help you design the workflow — precedent-grounded beats generated, nearly always.