ChatGPT for Lawyers: What It Can (and Can't) Safely Do

Lawyers can use ChatGPT for low-risk, non-confidential tasks — brainstorming arguments, plain-English explanations, reformatting text, or first-draft scaffolding you fully verify. It is not safe to rely on for legal advice, case citations, or anything involving client or matter data: general-purpose ChatGPT can fabricate cases that do not exist, defaults to United States law, does not reliably cite Australian or New Zealand authority, and its consumer versions are not built for the confidentiality and privilege duties lawyers owe. For real legal work, a purpose-built tool trained on Australian and New Zealand law — with citations, lawyer-led review, and a confidential, sovereign data posture — is the safer choice. Quillio is built for exactly that.

Can lawyers use ChatGPT?

Yes — for the right tasks, with the right precautions. No — as a source of legal advice or for confidential matter work. That is the short, honest answer, and the rest of this guide unpacks it: what ChatGPT can safely do for legal work, where it falls short, how to reduce the risk when you do use it, and the purpose-built alternative for real matters.

It is worth acknowledging the genuine appeal up front, because the point here is not that AI has no place in legal practice. ChatGPT is quick, it writes fluently, it is available the moment you need it, and it can be a real help with the blank-page problem. The issue is narrower and more practical: a general-purpose chatbot is being asked to do a specialist job, and for an Australian or New Zealand lawyer that gap shows up in jurisdiction, citations and confidentiality.

What ChatGPT can safely do for legal work

Used carefully on non-confidential, low-stakes work — where you verify everything before it touches a matter — ChatGPT can be a useful drafting and thinking aid. Reasonable uses include:

• Brainstorming arguments and counter-arguments to pressure-test your own thinking.
• Explaining a concept to yourself in plain English before you research it properly.
• Restructuring or tightening prose you have already written.
• Generating a checklist or outline scaffold you will rewrite in your own words.
• Summarising a document you are actually allowed to paste — that is, non-confidential or rigorously de-identified material.

The rule of thumb is simple: ChatGPT is a drafting and thinking aid, never an authority. Every output needs independent lawyer verification before it goes anywhere near a client matter. This is general-purpose AI used carefully — not a legal tool, and not a substitute for your own professional judgement.

Where ChatGPT falls short for lawyers (the real risks)

These are the limits of a general-purpose tool used for a specialist job. They are not a claim that the technology is bad — they are the reasons you cannot point it at a live matter and trust the result.

Hallucinated cases and citations

Because generative models predict plausible text, they can invent realistic-looking case names, citations and quotes that do not exist. The canonical example is the United States matter Mata v. Avianca in 2023, where lawyers were sanctioned after filing a brief containing fabricated, ChatGPT-generated case citations. Courts and regulators across several jurisdictions, including Australia, have since issued guidance on the careful use of generative AI in legal work — confirm the current guidance that applies to you before relying on any AI output.

United States-centric law, not AU or NZ

General-purpose models default to United States legal concepts and terminology. They do not reliably know current Australian or New Zealand statutes, the differences between jurisdictions (NSW versus Victoria, or Australia versus New Zealand), or recent amendments. For local practice, that is a fundamental mismatch rather than a rough edge.

No reliable citations to authority

Even when the substance is broadly correct, ChatGPT often cannot point you to the actual source you can cite. For legal work that must be grounded in authority, the inability to produce a verifiable, correct citation is disqualifying on its own.

Confidentiality, privilege and professional duty

Pasting client or matter data into a consumer chatbot can put confidentiality, privilege and conduct-rule obligations at risk. This is the single biggest reason firms hold back. The duty is yours to meet: this page does not give legal advice, and you should look to your firm's obligations and your regulator's current guidance to decide what is appropriate. In Australia, the OAIC has advised, as a matter of best practice, against entering personal information — particularly sensitive information — into publicly available generative AI tools. In New Zealand, the New Zealand Law Society has cautioned lawyers to address privacy, confidentiality and privilege risks before using generative AI on client work.

Is ChatGPT safe for lawyers? (confidentiality and data)

Consumer ChatGPT, enterprise tiers and API access handle data differently, and those terms vary by tier and change over time — so the safest approach is to check the provider's current, published terms rather than assume. But the lawyer's real question is not "is the vendor secure?" It is "does this meet my confidentiality, privilege and data-handling duties?" For live client matters, a general-purpose consumer chatbot is usually not the right venue, whatever the vendor's posture.

That is precisely why purpose-built legal AI exists. If you want to go deeper on the trust questions, see our guide to AI safety, confidentiality and data sovereignty for law firms.

How to use ChatGPT more safely for legal work

If you do use a general-purpose tool, a few disciplines make a real difference. This is the practical, genuinely useful part — not a sales pitch:

• Never paste client-identifying or privileged information. De-identify rigorously, or do not use it at all.
• Treat every factual and legal claim as unverified until you independently confirm it against primary sources.
• Never rely on an AI-supplied case citation without pulling and reading the actual judgment.
• Use it for structure and language, not for the law itself.
• Know and follow your firm's AI policy and your regulator's current guidance.

Done perfectly, this still leaves you with a workaround. It does not give you Australian or New Zealand-grounded drafting, reliable citations to authority, or a confidential home for matter data — which is the gap a purpose-built tool is designed to close.

What about Claude, "legal GPTs" and other general AI?

The same questions come up for Claude, for "legal GPT" custom assistants, and for any tool built on top of a general-purpose model. A custom prompt or a thin wrapper does not change the underlying model's training or your own data-handling duties — so jurisdiction fit, reliable citations and confidentiality remain the recurring gaps regardless of which general model sits underneath.

For the model-specific treatment, see our companion guide on using Claude for legal work.

The purpose-built alternative: Quillio

The shift that matters is from "a chatbot you prompt" to "an AI legal assistant built for the job". Instead of starting from a blank prompt, Quillio works from the actual matter and produces a grounded first draft the lawyer reviews — the precise thing a general-purpose chatbot cannot do safely. Quillio helps lawyers draft, review and refine; it does not replace the lawyer or automate legal advice. In practical terms:

These security claims should be confirmed in a firm's own compliance review. The aim is not to disparage general-purpose tools — if your need is non-confidential brainstorming or language polish, generic AI may suit you fine. For grounded drafting, citations and confidential matter data, a purpose-built legal AI is built differently for a reason.

When generic ChatGPT is fine — and when it isn't

A quick decision aid: it is fine for non-confidential brainstorming, language polish, explaining a concept to yourself, and scaffolding you will fully rewrite. It is not fine for drafting on a live matter, anything involving client data, relying on case law or citations, or producing client-facing or court-facing work without independent verification.

For the broader category view, see a purpose-built AI legal assistant for Australian law firms, and for the head-to-head, the Australian ChatGPT alternative for lawyers.