A step-by-step guide to the three workflows we build with professional teams: matter intake, first drafts in your own voice, and automated court list checking.
We guarantee you don’t need a technical background, but you need to keep your judgement on every output.
Free PDF · no email required
One rule was true long before AI and it’s still as important today – if your signature is on the document, you are responsible for every word of it. That was true when a junior associate gave you a draft, and it applies equally with an agent. We built this guide with that caveat in mind – set up the following four things first, before running any workflow.
Keep original client documents untouched in your normal filing system. Copy working documents into a second folder used only for AI work, and strip names, dates of birth, addresses and ID numbers as you copy. Convert files to markdown (plain text) where you can – because AI tools read md files faster, more accurately and with less token usage than a PDF.
Add one short text file at the root of each matter folder – to help steer your tool. Include who the client is, which court, the matter number, your position, your strategy. The AI reads the map first, so it stops hunting through every doc, your prompts stay two lines long, and the context stays accurate.
NSW Supreme Court Practice restricts what may be entered into gen AI unless the info stays within a controlled environment, is used only for the proceeding, and is not used to train the model. Check your AI tool’s enterprise data settings before any client material is entered, and de-identify regardless. Your duty of confidentiality remains.
SC Gen 23 commenced on 3 February 2025 and applies to NSW proceedings. Affidavit and witness statement content must not be generated with AI, expert reports require the court’s leave, and every citation must be verified manually. Other jurisdictions have issued their own guidance, so check the current rules for your court before you file anything!
The manual version: 45 to 60 minutes. Per matter, every matter, by someone whose time you would prefer billed.
Built as a workflow: about 5 minutes. Plus a compulsory 5 min lawyer review.
Write your manual process down. Every step, in plain language: what you open, what you look for, where you type it. That document becomes the instruction file (it’s called a “skill” if you’re using Claude) the AI follows the same way every time. Pick one matter type your firm sees all the time, similar matters teach the pattern fast.
Have the AI read the originating documents. The statement of claim and any orders. It extracts the type of proceeding, the court, the matter number, the parties, the key facts and the amounts claimed.
Encode your deadline rules. For example, in NSW a defence is due 28 days after service of the statement of claim (UCPR r 14.3), and a Fair Work unfair dismissal application must be lodged within 21 days of the dismissal taking effect. Build your matter type’s rules and the public holiday calendar into the skill and verify the current rules.
Log everything into your practice management system. One consistent home for this all – the matter record, the deadlines as calendar tasks, the extracted details in their proper fields.
Instruct it to stop and ask. Anything it could not find, or is unsure of, gets flagged as a question. AI likes to be confident – not necessarily correct. Ensure uncertainty is never stated as fact.
End with a client request list. Employment contract, payslips, correspondence: whatever the matter type needs, generated automatically so the file starts complete.
Generic AI drafting reads like AI slop. Senior people rewrite it from scratch, so it was a total waste of time.
Trained on your writing – so it reads like you. Still requiring sign off, but only a light touch edit.
Gather 10 to 20 of your own past documents of one type. Filings, advices or letters you were happy with, in your voice.
Have the AI analyse how you write. How you present facts, build an argument, structure an idea. Go down to the minute detail: the punctuation you prefer, phrases that are automatically AI ‘tells’ – “It’s not X it’s Y”, and the words you never use (quietly). Your professional judgment gets written into the skill, which helps the output to read like you.
Keep the matter brief in the folder. With the map file from the ground rules in place, your prompt can be two lines: what kind of document, and your strategy. “Argue 1, 2, 3 and 4. Number 4 matters most, because of X.”
Ask for a plan before a draft. Iterate the plan back and forth until you’re happy with it, then ask for the full draft. Lawyers hold an unfair advantage here – your entire profession is describing exactly what you want, in precise language, to someone with no interest in your problem!
Verify every authority yourself. Check each citation against AustLII or Jade. SC Gen 23 requires manual verification of every citation, and verification by an AI program is not how to do this! A well-built skill tells you which authority it missed and then asks you to supply one, as opposed to inventing one to please you.
Read every word, then sign. In NSW, affidavit and witness statement content must not be AI-generated at all. For everything else, apply the general ground rule – if it has got your signature on it, it’s your responsibility.
What you’re building: an agent in your Chrome browser that checks every active matter against the published court lists while you watch, then fills in a spreadsheet with the result. Setup is one go, and should take roughly 30 minutes.
WHAT YOU NEED FIRSTInstall Claude in Chrome. Open the Chrome Web Store, search “Claude”, and check the publisher says Anthropic before you click Add to Chrome. Sign in with your Claude account. Pin it so you can find it.
Give it permission for the court site only. When Claude asks which sites it can work on, add the court list page (e.g. the NSW Online Registry) and your Excel/Google Sheet. Nothing else. And leave the confirmation prompts switched on – that way Claude asks before it acts.
Open the list page, then open Claude beside it. Navigate to the published lists exactly the way you normally would. Click the Claude icon and a side panel opens next to the page. Claude can now see what you see.
Record the check once. In the side panel, choose Record a workflow, then do the check yourself for one matter, slowly: which page, which search field, where the matter number goes, what a listing looks like when you find one. Claude watches and learns your exact steps. Stop the recording and name it “court list check”.
Test on five matters. Type: “Run the court list check for the first five matters in [your sheet link]. For each one, write into the sheet: listed or not listed, which court, which day, and the position on the list. If you cannot find a matter or you are not sure, write UNSURE.” Then watch it work. When it takes a wrong turn, correct it in plain language (“the search field is the one on the left”) the same way you’d correct a new starter.
Fix, then scale. Whatever it got wrong on five, tell it, and then run five again. Give it the full list only once five come back all correct. Debugging at full volume is miserable – you don’t want to do that!
When your court publishes its lists, run it. Open the court list page, open the side panel, and say: “Run the court list check for every matter in [sheet link].” The agent works through the list while you watch, filling the sheet as it goes, and pacing itself like a person.
Review the sheet like a lawyer. UNSURE rows first, then spot-check a handful of the rest. Run your normal manual check in parallel for the first month – a missed listing means a missed appearance, so this one is too critical to hand over on day one. Keep the report per matter – which court, which day, which position on the list.
Pick one repetitive task this week and write down exactly how you do it, every step, in plain language. That document is your first skill. Start small and low-risk, build confidence, then take on the bigger processes. AI replaces tasks. Lawyers keep the strategy, the client relationships and the final say.
Neoma is a Sydney-based AI capability uplift and workforce reskilling consultancy. We train teams to build and run workflows like these themselves, and we measure adoption so leaders can see the return. Neoma is a member of the Anthropic Partner Network and the Tech Council of Australia, and currently delivers AI capability programs across both the NSW Government and the private sector.
Free PDF · no email required
Gemma Toogood · Co-Founder · 0478 890 064 · partnerships@neoma.com.au · neoma.com.au
