Document review is how lawyers read, code and assess documents for relevance, privilege, risk and key terms, whether for discovery, due diligence or a contract negotiation. AI document review does the first pass in minutes: it sorts, summarises and flags clauses so lawyers spend their hours on judgment instead of reading. Australian courts have accepted technology assisted review since 2016, but every AI output still has to be verified by a lawyer.
If your firm still reviews documents page by page, you are paying senior rates for work a machine can triage. This guide explains what document review is, how the AI version works step by step, how accurate it really is, and what Australian courts and regulators now expect. It closes with a checklist for choosing a legal AI platform built for Australian law firms.
What is document review in law?
Document review is the stage where lawyers examine a set of documents and decide what each one means for the matter. In litigation that means relevance, responsiveness to discovery and privilege. In transactions it means risk, obligations and missing terms.
The most common types of legal document review are:
- Discovery review: coding emails, files and chats for relevance and privilege before production.
- Due diligence: checking leases, supply agreements and corporate records for change of control, assignment and liability terms.
- Contract review: comparing a draft against your firm's playbook before you negotiate. See AI contract review.
- Employment review: checking contracts and offer letters against awards, the National Employment Standards and restraint law. See employment contract review.
- Regulatory and internal investigations: finding what a regulator or board needs to see, fast.
Manual vs AI document review: why the maths changed
Manual, linear review is slow because a person must open every document. A common industry benchmark is 40 to 50 documents per hour per reviewer.
In McConnell Dowell v Santam, the Supreme Court of Victoria's first approval of technology assisted review, the party had about 4 million documents. De-duplication cut that to roughly 1.4 million. At 50 documents an hour, that is still about 28,000 reviewer hours before a single privilege call.
AI changes the order of work. Instead of reading everything, lawyers teach or instruct the system, let it rank and summarise the set, and then review what matters first. For an SME firm, the same logic applies to a 60-page commercial lease or a stack of 30 employment contracts: the machine reads, the lawyer decides.
How AI document review works: the 6-step process
- Collect and ingest. Upload the documents (PDFs, Word files, emails, scans) into the review platform. Poor OCR on scanned pages is the most common cause of missed text, so check scans first.
- Cull and de-duplicate. Remove duplicates, filter by date and custodian, and set aside documents that cannot be relevant. This step alone often removes most of the volume.
- Set the review criteria. Write the coding protocol or contract playbook: what counts as relevant, which clauses are must-haves, which positions are acceptable. Clear criteria are the single biggest driver of AI accuracy.
- Run the AI first pass. The system classifies documents, extracts parties, dates and amounts, summarises each file and flags risky or missing clauses against your criteria.
- Lawyer second pass and privilege review. Lawyers check the flagged items, resolve close calls and make every privilege decision. Privilege is a legal judgment and should never be delegated to a model.
- Quality control and delivery. Sample the AI's "not relevant" pile, confirm every citation and quote against the source document, then produce the set, the redline or the advice.
On LegalScout, steps 3 to 4 run as playbook-aligned review with a risk dashboard, gap analysis and redline comparison. Drafting the response then moves to AI legal document drafting grounded in your own precedents, and repetitive intake can be handed to agentic AI associates.
How accurate is AI document review?
For large discovery sets, courts have accepted that machine-assisted review is reliable. In McConnell Dowell, Justice Vickery relied on Irish authority finding that technology assisted review is at least as accurate as manual review for large data sets.
Generative AI is a different risk. It can summarise and analyse brilliantly, but it can also invent things. By August 2025, more than 20 Australian matters had involved AI-generated fake citations or errors in court documents. In the Dayal matter, the Victorian Legal Services Board varied a solicitor's practising certificate after he tendered an AI-generated list of authorities with inaccurate citations.
Accuracy depends on four things, not the AI alone:
- The tool: answers grounded in your uploaded documents and Australian sources, with a citation to the exact clause or page.
- The documents: clean text, complete attachments, consistent formats.
- The instructions: a matter-specific playbook beats a vague prompt every time.
- The lawyer: a human check on every output that leaves the firm.
The practical rule: use AI to find and flag, use lawyers to decide and sign. For case law, use a research tool that cites real Australian authorities, such as AI legal research for Australia, and still open every case you rely on.
AI document review in Australia: the rules you need to know
- Technology assisted review is accepted. McConnell Dowell [2016] VSC 734 and the Victorian Practice Note SC Gen 5 (January 2017) endorsed it for large discovery. The Federal Court's Technology and the Court Practice Note (GPN-TECH) also recognises it.
- NSW sets strict generative AI rules. Practice Note SC Gen 23 has applied in the NSW Supreme Court since 3 February 2025, and the District Court adopted it. Affidavits, witness statements and character references must reflect the person's own knowledge, not AI-generated content. Where generative AI helps prepare submissions, citations must be verified, and that verification cannot be done by a generative AI program.
- Confidentiality is your duty. The NSW note warns that material entered into public AI tools may be used to train models. That makes data hosting a professional responsibility question, not an IT preference. Read LegalScout's approach to Australian cloud hosting and Privacy Act compliance.
AI document review tools compared (2026)
There is no single best tool. The right one depends on your matter volume, budget and where your client data must live.
- eDiscovery and TAR platforms (Relativity, Everlaw, Logikcull): built for very large litigation sets, predictive coding and production workflows. Often more than an SME firm needs.
- Enterprise legal AI (Harvey, Legora): broad generative AI for large firms and in-house teams. Compare costs in our legal AI pricing guide and our Legora alternative breakdown.
- Practice management add-ons (such as Clio Duo): handy summaries inside software you already run, with lighter review controls.
- Generic chatbots (ChatGPT and similar): fast, but not designed around privilege, firm playbooks or Australian data residency. Risky for client documents.
- LegalScout: a private legal AI platform for Australian SME firms. Contract and employment contract review with playbooks, risk dashboards and redlines, hosted in AWS Sydney with zero data retention by model providers and no training on your data. Connects with InfoTrack.
For the full field, see our ranking of the best legal AI tools in Australia.
How to choose AI document review software: a 7-point checklist
- Where is the data hosted? Ask for the region in writing. Australian hosting keeps client files under Australian law.
- Is anything retained or used for training? You want zero retention by model providers.
- Does every answer cite its source? No citation, no reliance.
- Can it follow your playbook? Generic risk flags are not your firm's positions.
- Does it fit your workflow? Check integrations with InfoTrack, Actionstep or Clio.
- Is the security real? Look for encryption at rest and in transit and enforced MFA. See enterprise security.
- Is the price built for your size? Check LegalScout pricing against enterprise quotes.
Where LegalScout fits
LegalScout is built for the firms most review tools ignore: SME law firms, sole practitioners and in-house legal teams. The same platform handles family law financial calculations, and law students use it to learn review the way practice now works. Everything sits behind the LegalScout Trust Centre.
Frequently asked questions
1. Can AI review legal documents accurately enough for a law firm to rely on?
Yes, for the first pass, if a lawyer verifies the output. Australian courts have accepted technology assisted review for large discovery since 2016. Generative AI adds speed but can invent citations, so use a tool that links each finding to the source clause and check every result before it reaches a client or a court.
2. Is AI document review allowed in Australian courts?
Yes. The Supreme Court of Victoria endorsed technology assisted review in McConnell Dowell v Santam (2016), and the Federal Court's technology practice note recognises it. In NSW, Practice Note SC Gen 23 permits generative AI with limits: no AI-generated content in affidavits or witness statements, and verified citations in submissions.
3. What is the best AI for legal document analysis for a small Australian law firm?
The best fit for a 2 to 20 lawyer firm is a private platform hosted in Australia that follows your playbook and cites its sources. Enterprise tools like Harvey and Legora suit large firms. LegalScout is built for Australian SME firms, with AI contract review, drafting and research in one place.
4. Is an Australian-hosted legal AI safer than global tools?
For client confidentiality, usually yes. Australian hosting keeps documents under Australian law and makes Privacy Act 1988 obligations easier to meet. Also check that the model provider keeps no copy of your data and never trains on it. LegalScout is hosted in AWS Sydney with zero data retention; see the Trust Centre.
5. What is the best way to analyse an employment contract or offer letter with AI?
Use a tool that checks the document against the relevant modern award, the National Employment Standards and enforceable restraint terms, not a generic chatbot. Upload the contract, review the flagged gaps, then confirm each issue yourself. LegalScout's employment contract review runs these Fair Work checks automatically.
Try AI document review on your next matter
Upload a contract and see the risk dashboard, gap analysis and redline in minutes. Start a free trial, explore AI contract review, or book a demo with the team behind Australia's private legal AI platform for SME law firms.


