Drafting with AI — Contracts, Briefs, and Correspondence
Deliberate Academy Editorial Team
Reviewed for accuracy and professional relevance
You're 4 lessons in — don't lose your progress.
Sign up free to save where you are and earn a verified certificate when you pass.
- Construct a structured legal drafting prompt that captures parties, commercial terms, governing law, and required provisions to reduce the review burden
- Apply the six-point drafting review checklist to an AI-generated legal document before it leaves your desk
- Explain the three key limitations of AI in legal drafting — commercial context, relationship and strategy, and jurisdiction-specific requirements
- Generate clause alternatives for negotiation using AI with specified commercial parameters for each variant
Drafting is the part of legal practice where AI offers the most immediate, most visible, and most widely used efficiency gains. First draft generation, clause alternative production, plain-English translation, and tone adaptation are all tasks where AI can compress hours of skilled work into minutes. The critical skill is not producing the draft — it is reviewing it with the rigor that legal drafting demands, because the AI does not know your client, does not know the deal, and does not understand what the output will be used for.
AI as a Drafting Accelerator
First draft generation. For standard transaction documents — NDAs, service agreements, consultancy terms, employment contracts, board minutes — AI can produce a solid first draft from a structured prompt in under a minute. The draft will be technically competent in structure, will include the standard provisions the document type requires, and will be a workable starting point. The time it saves is real: a first draft that previously took a junior lawyer two hours to produce from a precedent now takes five minutes of prompt engineering.
The caveat is equally real. A generic first draft is a starting point, not a finished document. It does not reflect the commercial reality of the specific deal, the risk appetite of your client, the negotiating dynamic with the counterparty, or the specific obligations your client has agreed to in heads of terms. The difference between a first draft and a deal-ready document is the layer of qualified legal judgment applied to it.
Clause alternative generation. When a counterparty challenges a provision, AI can generate alternative clause formulations that shift the commercial position incrementally. "Draft three alternative versions of this limitation of liability clause: one that maintains our current position, one that makes a modest concession on the cap amount, and one that narrows the carve-outs from the cap" is a prompt that produces useful alternatives quickly. The selection and negotiation of which alternative to deploy is still a judgment call requiring legal and commercial context.
Plain-English adaptation. Translating complex legal provisions into client-accessible language is a skill that consumes significant lawyer time in client-facing practice. AI handles this translation task well, provided the legal content it is translating from has been verified. A client-facing explanation of a complex indemnity regime, drafted by AI from a reviewed and accurate legal summary, can be a significant time-saver. An AI-generated plain-English explanation produced from an AI-generated legal summary is a compounded risk.
Use this structured review checklist for every AI-generated legal draft before it leaves your desk. Check: (1) every defined term is used consistently throughout the document; (2) every cross-reference is accurate; (3) the governing law and jurisdiction clauses are correct for this matter; (4) any exclusions, carve-outs, or limitations reflect the commercial deal as agreed; (5) there are no blank spaces, placeholder brackets, or template instructions that have not been removed; (6) the document has been checked against any heads of terms, term sheet, or commercial summary for consistency. AI drafts frequently contain cross-reference errors, retained template text, and defined terms that are introduced but never used — these are the most common and most professionally damaging errors to send to a counterparty.
Prompt Engineering for Legal Drafting
The quality of an AI-generated legal draft is largely determined by the quality of the prompt. A general prompt — "draft an NDA between two companies" — produces a generic document. A specific prompt produces a more useful one.
A structured legal drafting prompt includes: the document type and purpose; the parties (described by type, not by name); the key commercial terms that must be reflected (duration, scope of confidentiality, permitted disclosures, remedies); the governing law; any specific provisions that must be included or excluded; the tone (formal commercial, plain English for SME client); and the jurisdiction.
The more specific the prompt, the more of your drafting judgment is captured in the instruction layer rather than the review layer. A well-constructed prompt that includes all material commercial terms reduces the review burden because the AI has more to work with. A generic prompt increases the review burden because the AI has produced a generic document that needs substantial adaptation.
A paralegal uses the prompt 'draft an NDA between two UK companies' to generate a first draft for a new client matter. The partner reviews the document for structural completeness and sends it to the counterparty. What is the most significant risk in this workflow?
Select one answer.
Limitations of AI in Legal Drafting
Commercial context. AI does not know what was negotiated during the deal process, what concessions were made in heads of terms, what the relationship history between the parties is, or what the client's actual risk appetite is. It will draft to market-standard positions — which may be materially different from the position this specific deal requires.
Relationship and strategy. In litigation drafting — witness statements, skeleton arguments, pleadings — the strategic framing of the case, the selection of points to emphasize and points to reserve, and the judgment about what the tribunal or court needs to understand are decisions that require the advocate's professional knowledge of the case, the judge, and the proceedings. AI can produce a structured first draft of a skeleton argument from a fact summary, but the advocacy judgment embedded in the final document must come from the lawyer.
Jurisdiction-specific requirements. Legal drafting requirements vary significantly by jurisdiction. Court filing requirements, mandatory notice periods, specific formality requirements for particular document types — AI may not reflect these accurately for a given jurisdiction, particularly for less common jurisdictions or for recent legislative changes made after the model's training cutoff.
Never send an AI-generated legal document to a client or counterparty before a qualified lawyer has reviewed every substantive provision. The AI draft is a starting point for lawyer work, not a finished legal product. The professional and reputational risk of sending a document that contains an incorrect cross-reference, a retained template clause, or a provision inconsistent with the agreed commercial terms is disproportionate to the time saved by skipping the review. Build the review step as a fixed step in your drafting workflow — not a discretionary one depending on time pressure.
First-Draft NDA Review — In-House Legal Team, Technology Scale-Up
Context
An in-house legal team at a growing technology company was managing a high volume of NDAs for partnership and vendor onboarding. The team began using AI to generate first drafts from structured prompts to reduce turnaround time. One draft NDA reached a counterparty before the standard six-point review had been completed, due to a miscommunication about which version had been signed off.
Action
The counterparty's solicitor raised two issues: a limitation of liability carve-out that the counterparty had explicitly refused to accept in the prior negotiation, and a governing law clause that defaulted to a jurisdiction the parties had agreed to move away from. Both errors reflected the AI drafting to market-standard positions rather than reflecting the agreed commercial terms. The in-house counsel pulled the email and sent a corrected version with an explanation. She then introduced a version-control workflow: AI drafts were watermarked as unreviewed until each of the six checklist points had been signed off by a qualified lawyer.
Outcome
The counterparty accepted the correction without escalating the issue. The in-house team's post-incident review concluded that both errors were consistent with how AI drafting tools behave — they produce technically sound market-standard documents that do not reflect deal-specific carve-outs unless those are explicitly included in the prompt. The revised workflow required the drafter to append a short prompt addendum capturing any non-standard commercial terms before generating each draft, reducing the gap between the AI output and the agreed deal.
A trainee solicitor uses AI to draft a service agreement for a new client engagement. The AI produces a well-structured document that includes all the expected clauses. The trainee reviews it for obvious errors and sends it to the client. What is the most significant gap in this review process?
Select one answer.
Exercise
Your Task
Take a standard document type you draft regularly — an NDA, a service agreement, or a similar transaction document. Write a structured prompt that includes all the elements from this lesson: document type and purpose, parties described by type, key commercial terms, governing law, any specific provisions to include or exclude, tone, and jurisdiction. Generate a first draft, then apply the six-point review checklist: consistent defined terms, accurate cross-references, correct governing law, commercial terms matching the agreed deal, no retained template text, and no blank spaces. Record every error the checklist catches that a quick read-through would have missed.
Your reflection
Did you complete this exercise? What did you find? (Saved locally in your browser)
Try It: AI-Graded Practice
The exercise above is self-assessed. The exercise below is graded automatically, so you can get direct feedback on whether your rewritten drafting prompt actually captures the structure this lesson teaches.
- AI first draft generation for standard transaction documents — NDAs, service agreements, employment contracts — compresses hours of junior lawyer drafting work into minutes, but the draft reflects market-standard positions, not the specific deal your client agreed to.
- Clause alternative generation is a practically useful AI drafting application — specifying the commercial parameters of each alternative (maintain position, modest concession, narrowed carve-out) produces useful variants quickly for negotiation purposes.
- A structured drafting review checklist — consistent defined terms, accurate cross-references, correct governing law, commercial terms matching the agreed deal, no retained template text — addresses the most common and most professionally damaging AI draft errors.
- The quality of a legal drafting prompt determines the quality of the output — a prompt that includes parties, commercial terms, governing law, required and excluded provisions, and tone reduces the review burden by capturing more of the lawyer's judgment in the instruction layer.
- AI in litigation drafting — skeleton arguments, pleadings, witness statements — can produce structural scaffolding, but the advocacy judgment embedded in the final document must come from the lawyer who knows the case, the tribunal, and the strategy.