Why generic AI fails in legal practice
ChatGPT and Claude on the public web are powerful — but every prompt risks waiver of legal professional privilege if it contains client identifiers, draft advice, or matter facts. Most firms we speak to have already had a partner type something into a public model that should never have left the firm.
A custom GPT for your law firm runs on infrastructure where the prompt and the response stay inside your tenant. We deploy on Azure OpenAI (Australian regions), a private LLM in your own VPC, or an Anthropic Bedrock instance in ap-southeast-2 — and turn off training on your data.
What Australian law firms typically build
Precedent Drafting Assistant
Pulls from your firm's library of 5,000+ approved clauses. A junior asks 'draft a deed of release for an employment dispute, NSW jurisdiction' and gets a first cut citing only your precedents.
Matter Q&A
Indexed across your matter management system (Affinity, Practice Evolve, Leap, Smokeball). Partners ask 'what was agreed at the 14 March mediation?' and get an answer with the source document linked.
Case-law Summariser
Reads AustLII, Federal Court, and NSW Caselaw judgments and returns a 200-word summary in your firm's note-up style. Reduces 4 hours of reading to 20 minutes.
Client Intake Triage
Front-of-website assistant that captures matter details, conflicts info, and urgency — then drops a structured intake form into your CRM with the priority pre-filled.
Privilege and confidentiality controls
- Data residency — deployed in Sydney or Melbourne Azure regions. No data leaves Australia.
- No training — your prompts and responses are never used to train a foundation model. Confirmed in writing under DPA.
- Per-matter access — the GPT can only retrieve documents the user has permission to see in your matter management system.
- Audit log — every query is logged with user, timestamp, retrieved documents, and response. Exportable for compliance review.
- Privileged-mode flag — partners can mark a session as privileged so it's stored separately and excluded from analytics.
Where the time savings actually come from
Across 14 Australian law firm engagements we've measured, the consistent wins are: first-draft brief preparation drops from 90 minutes to 18, case-law note-ups drop from 3 hours to 35 minutes, intake form completion at the top of the funnel rises from 22% to 71%, and after-hours client enquiries get a useful triaged response instead of an empty inbox.
Solicitor-on-record stays in the loop. Every output the GPT produces is marked 'draft for solicitor review' and routed to a human before it goes to the client. The model accelerates work — it doesn't replace the practising certificate.
Frequently asked questions
Will using a custom GPT breach the Legal Profession Uniform Law or the Solicitors' Conduct Rules?
No, provided the firm retains supervisory responsibility, the GPT's outputs are reviewed by a solicitor before being sent to the client, and the data architecture preserves confidentiality. We deploy under a configuration that has been reviewed by partners at firms in three states.
Can it cite Australian case law correctly?
Yes. We connect to AustLII and the firm's internal authorities library so citations are pulled from authoritative sources, not hallucinated. The GPT is hard-blocked from inventing case names.
How is this different from Harvey AI or Lexis+ AI?
Harvey and Lexis+ are excellent off-the-shelf products with strong US/UK case-law coverage. A custom GPT is the right choice when you need (1) tight integration with your specific document management system, (2) Australian jurisdictional coverage, and (3) full control over the data pipeline. We sometimes recommend Harvey instead — we'll tell you when.
What's the cost compared to a paralegal?
A typical 20-fee-earner firm pays $2,400–$4,800 per month for a custom GPT covering precedent drafting, matter Q&A, and intake. That's roughly the equivalent of 4–8 hours of paralegal time per week — but the GPT is available 24/7, doesn't need annual leave, and scales linearly with usage.
Ready to build your custom GPT?
Get a free 30-minute scoping call. We'll map your use case, data sources, and ROI before you commit.
Start the Conversation