AI for law firms · Guide
AI first drafts in law firms: what suits AI and how to review it
AI first drafts in law firms work best for routine letters, file notes, chronologies and document summaries built from the firm's own precedents. Every draft needs review by a lawyer qualified to verify it, because the lawyer who uses it stays responsible. Court documents need the most care: the NSW Supreme Court bars Gen AI from generating affidavit content.
- By
- Elias
- Published
- Last reviewed
- Reading time
- 7 min read
Which legal documents suit an AI first draft?
AI first drafts suit documents that are routine, built from information already on the file, and quick for a lawyer to check: correspondence, file notes, chronologies and summaries. They suit advice and court documents far less, and court rules restrict Gen AI for some documents. The test is simple: can the responsible lawyer verify every line faster than writing it?
The regulators point the same way. The joint statement on the use of AI in Australian legal practice from the Victorian Legal Services Board + Commissioner, the Law Society of NSW and the Legal Practice Board of Western Australia suggests limiting AI to tasks that are “lower-risk and easier to verify”, such as drafting a polite email, and not using it for higher-risk tasks such as analysing an unfamiliar legal concept.
| Draft | Fit for AI | What the reviewer checks |
|---|---|---|
| Routine letters: acknowledgements, requests for documents, cover letters | Good | Names, dates, amounts, tone, and that nothing is promised |
| File notes and attendance notes from the lawyer’s notes or a transcript | Good | That the note records what was actually said and agreed |
| Chronologies and document summaries | Good | Every date and quotation against the source document |
| Standard agreements from firm precedents | Moderate | Every variable, defined term and cross-reference |
| Letters of advice | Limited: structure and plain-English editing only | That the analysis is the lawyer’s own |
| Written submissions | High care | Every authority, plus any verification the court requires |
| Affidavits and witness statements (NSW Supreme Court) | Not for content | That Gen AI did not generate the content, and the required disclosure is included |
What does NSW Practice Note SC Gen 23 allow?
Practice Note SC Gen 23, which has applied in the NSW Supreme Court since 3 February 2025, allows Gen AI for chronologies, indexes and witness lists, for summarising or reviewing documents and transcripts, and for preparing written submissions, subject to conditions. It prohibits Gen AI generating the content of affidavits, witness statements and character references, and requires leave before Gen AI is used in an expert report.
Three provisions matter most for drafting:
- Submissions (paragraphs 16–17). Where Gen AI was used, the author must verify in the body of the submissions that all citations, legal and academic authority, case law and legislative references exist, are accurate and are relevant. That verification must not be carried out solely by a Gen AI tool.
- Evidence (paragraphs 10–13). Gen AI must not generate the content of affidavits, witness statements or character references, though work merely preparatory to drafting is allowed. The document must disclose that Gen AI was not used to generate its content.
- Protected material (paragraph 9A). Material subject to non-publication or suppression orders or the Harman undertaking, or produced on subpoena, must not be entered into a Gen AI program unless the practitioner is satisfied it will stay in a controlled environment with confidentiality restrictions, be used only for that proceeding, and not be used for training any large language model.
Other courts set their own rules. The Federal Court of Australia has its own Generative AI Practice Note (GPN-AI), and the requirements differ in the detail, so check the current practice note of every court you file in.
How does drafting from your firm’s precedents work?
Drafting from precedents means the AI tool first retrieves your own approved templates, clauses and the relevant matter documents, then drafts from them, rather than relying on what a general model absorbed in training. The technique is called retrieval-augmented generation. It makes drafts read like your firm’s work and lets the tool show which precedent each paragraph came from.
It narrows the risk of invented content; it does not remove it. The model can still misread a clause or blend two sources. Three practical rules help:
- Curate the precedent bank. Only current, approved versions go in. Retire superseded precedents and tag each one by jurisdiction, so a Victorian letter is never drafted from the NSW version.
- Give the tool only what it needs. The Law Society of NSW’s guide to responsible use of AI (updated January 2026) recommends the principle of least privilege: the tool gets the data it needs and nothing more. Draft from the matter the lawyer is working on, not the whole document system.
- Prefer tools that show their sources. The same guide suggests preferring tools that let users interrogate the sources behind an output.
Where the tool runs matters as much as how it drafts. The joint statement says lawyers cannot safely enter confidential, sensitive or privileged client information into public AI chatbots, and should carefully review the contractual terms of any commercial AI tool used with client information. Our comparison of private AI vs ChatGPT for law firms sets out the options.
Why must every AI draft be verified?
Every AI draft must be verified because language models produce fluent text that can be wrong, and the lawyer who relies on it remains responsible. The joint statement is blunt: no tool based on current large language models can be free of hallucinations. Verification is part of drafting, not an optional last step.
Australian practitioners have already paid for skipping it. In Dayal [2024] FedCFamC2F 1166, a Victorian solicitor tendered a list and summary of authorities, generated with AI, that contained inaccurate citations and summaries. On 19 August 2025 the Victorian Legal Services Board varied his practising certificate so that he could no longer practise as a principal or handle trust money, and required two years of supervised practice with quarterly reporting (VLSB+C statement). The regulator acted against the practitioner, not the software vendor, so the checking has to happen inside the firm.
Nor can AI mark its own homework. The Law Society of NSW guide warns that solicitors should not rely on generative AI to verify sources produced by AI, and SC Gen 23 says citation verification must not be done solely with a Gen AI tool.
What do supervision duties mean for AI drafting?
Supervision duties apply to AI drafts in the same way they apply to a junior’s work. Rule 37.1 of the Australian Solicitors’ Conduct Rules requires the solicitor with designated responsibility for a matter to exercise reasonable supervision over everyone working on it. The Law Society of NSW guide adds that principals must take reasonable steps to ensure the practice’s lawyers comply with their obligations, and says AI deserves the same care as a legal assistant or paralegal.
Two consequences follow for a firm’s drafting process:
- The reviewer must be able to verify the content. The joint statement says lawyers using AI to prepare documents “must be able and qualified to personally verify the information they contain”. A graduate who drafts advice in an unfamiliar area with AI cannot verify it; the supervising lawyer must.
- Supervision failures are judged too. In Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731, a document filed for the applicant contained footnotes referring to reports and papers that were mostly non-existent or incorrectly cited. The junior solicitor who prepared it had used Google Scholar as the source of the citations. The Law Society of NSW guide notes the court found that, among other things, a failure to ensure her work was adequately supervised contributed to the error. Whatever the tool, the supervision duty is the same.
Costs follow the same logic. The Law Society of NSW guide notes that legal costs must be fair, reasonable and proportionate under section 172 of the Legal Profession Uniform Law, and that under time-based billing, time saved through AI should be reflected in the bill.
What should a review checklist for AI first drafts include?
A review checklist turns “check the draft” into specific steps a supervisor can audit afterwards. Use it for every AI first draft that goes on the file or leaves the firm, and record who completed it.
- Sources. Open the precedent and matter documents the draft relied on.
- Facts. Check names, dates, amounts, addresses and matter references against the file.
- Authorities. Open every case, section and rule in an authoritative source, not by asking an AI tool. Confirm it exists, says what the draft says and is still current.
- Quotations. Compare every quotation with its source, word for word.
- Omissions. Look for conditions, caveats or deadlines the draft left out.
- Carry-over. Make sure no other client’s names or details survived from a precedent.
- Instructions and tone. Check the draft matches the client’s instructions and the firm’s style.
- Court rules. For court documents, apply the relevant practice note, such as the citation verification SC Gen 23 requires in NSW Supreme Court submissions.
- Record. Note on the file that AI was used, for what, and who reviewed it.
- Time. Record time that reflects the work actually done.
How does the Drafting Assistant, by Pylon Digital, handle review?
The Drafting Assistant, by Pylon Digital prepares first drafts of letters, file notes and summaries for a lawyer to review, and builds review into the workflow rather than leaving it to habit. Nothing leaves the firm without human approval, and a full audit log records each draft, who reviewed it and what was sent.
It is one of our AI agents for professional firms. Where drafting draws on confidential client material, firms that cannot use public tools can pair it with Private AI & Knowledge Assistants, which stores client data in fully GDPR-compliant data centres.
This is general information, not legal advice.