The Lawyer's Guide to Working with AI
What to use it for, what to ignore, and where it quietly fails. A free resource for legal professionals in the common law jurisdiction.
What to use it for, what to ignore, and where it quietly fails.
A free resource for legal professionals in the common law jurisdiction.
Before we begin
This guide is not about whether you should use AI. That question has already been answered — by your clients, by your competitors, and by the direction of the profession.
This guide is about something more specific: how to use AI without being used by it.
Because the risk for lawyers is not that AI will replace them. The risk is subtler than that. It is that AI will make it harder to tell the difference between a lawyer who thinks well and one who merely produces well-formatted output.
The lawyers who will thrive in the next decade are not those who use AI the most. They are those who understand it well enough to know exactly when to trust it, when to question it, and when to set it aside entirely.
That is what this guide is for.
AI will not replace lawyers. But lawyers who know how to work with AI will replace those who don’t.
This guide covers five things:
- What AI actually is — explained in terms that matter for legal practice
- Where AI genuinely adds value in legal work
- Where AI quietly fails — and how to catch it
- Three questions every lawyer should ask before using AI output
- What good AI practice looks like in day-to-day legal work
Section One: What AI actually is
A functional explanation — not a technical one
You do not need to understand how a combustion engine works to drive a car. But you do need to know that it runs on petrol, not diesel — because putting the wrong fuel in has consequences.
The same logic applies to AI in legal practice. You do not need to understand the mathematics. But you do need to understand the following.
AI generates the most probable next word. Not the correct one.
When you ask an AI tool a legal question, it does not look up the answer in a database. It does not reason through the problem the way a lawyer does. It predicts, based on enormous amounts of text it has been trained on, what a plausible response to your question would look like.
This is why AI output sounds confident even when it is wrong. The model is not hedging. It is completing a pattern. And patterns that look like authoritative legal writing are exactly what it has been trained to reproduce.
What this means in practice
An AI tool can produce a response that reads like settled law but contains a case that does not exist, a statutory provision that was amended three years ago, or a principle that applies in one jurisdiction but not yours. It will present all three with identical confidence.
This is not a bug. It is how the technology works. Your job, as the lawyer, is to know this and to verify accordingly.
Why it sounds so good
AI has been trained on vast quantities of well-written legal text — judgments, textbooks, legal journals, practitioner guides. This means it has absorbed the style, structure, and register of good legal writing. It knows how a contract clause should sound. It knows how a skeleton argument is formatted. It knows the vocabulary of the law.
What it does not know is whether the substance behind that style is correct, current, or applicable to your specific facts.
Section Two: Where AI genuinely adds value
AI is not useful for everything. But for certain tasks in legal practice, it is genuinely, materially useful — and dismissing it entirely is as much a professional risk as adopting it uncritically.
The following are tasks where AI consistently performs well, with notes on how to get the most out of it for each.
First-draft documents
AI is well-suited to producing a first draft of standard documents: NDAs, employment contracts, terms and conditions, service agreements, consent letters. It will not produce a final document but it will produce a working starting point faster than starting from a blank page.
How to use it well: Give AI your specific parameters upfront. Jurisdiction, governing law, party type, key commercial terms. The more specific your prompt, the closer the first draft will be to something usable. Then review every clause — not for style, but for substance.
Summarising long documents
AI is very good at condensing lengthy materials into structured summaries: long judgments, due diligence documents, disclosure bundles, expert reports. This is one of the highest-value uses in practice — it saves hours without requiring the same level of verification as, say, legal research.
How to use it well: Ask AI to summarise by issue, not just chronologically. For a judgment, ask for the key findings on each ground of appeal. For a contract, ask for the obligations and limitations on each party. Structured prompts produce structured output.
Research starting points
AI can be a useful tool for getting oriented on an unfamiliar area of law — understanding the general framework, the key statutes, the main issues involved in the case. Think of it as a research assistant who has read widely but must never be trusted on citations without independent verification.
How to use it well: Use AI to map the landscape, then verify every case and statutory reference through your primary legal databases (Westlaw, Lexis, Manupatra, SCC Online). Never cite anything from AI output without checking the primary source.
Structuring legal advice
AI is good at generating structure: the logical skeleton of a legal opinion, the issues to address in a memorandum, the headings for a due diligence report. It will not fill in the substance correctly — but it can ensure you have not missed a structural element.
Generating clause options
In contract work, AI can generate multiple versions of a clause — e.g. a limitation of liability clause with varying risk allocation — which a lawyer can then evaluate and adapt. This is particularly useful in negotiation preparation, where having options quickly is more valuable than having one perfect draft.
Timeline and chronology building
AI can extract dates, events, and sequences from factual documents and organise them into a chronology. This is useful in litigation, arbitration, and regulatory matters where chronological clarity matters and the underlying documents are voluminous.
Section Three: Where AI quietly fails
This is the section that matters most. Because AI’s failures are not always obvious. They do not come with error messages. The output still looks polished, confident, and well-formatted. That is precisely what makes them dangerous.
Every failure mode below is one that a lawyer using AI without awareness could miss — and that a client, a court, or a regulator could later hold them responsible for.
| Where AI fails | What to do instead |
|---|---|
| Hallucinated cases and citations | Verify every case reference through a primary legal database. Do not cite anything you have not read. |
| Outdated law | AI training data has a cutoff date. Always check that the statute, regulation, or case you are relying on is current. |
| Jurisdiction confusion | AI frequently blends English, Scottish, US, and Commonwealth law in a single response. Specify your jurisdiction explicitly in every prompt, and verify. |
| Overconfidence on unsettled points | Where the law is genuinely uncertain, AI will often present one position as though it were settled. Check whether the point is actually contested. |
| Missing what is not there | AI can tell you what a contract says. It cannot reliably tell you what a well-drafted contract should say but doesn’t. Spotting omissions requires a lawyer. |
| Commercial and factual context | AI does not know your client, their industry, their risk appetite, or their commercial priorities. These shape every legal judgment. AI cannot make them. |
| Negotiation and leverage | Assessing what a counterparty will accept, what they need to walk away with, and where the real pressure points are — this requires human judgment and relationship knowledge. |
| Privilege and confidentiality | Inputting client-specific facts into a third-party AI tool may raise privilege and confidentiality issues. Know your firm’s AI policy and your professional obligations before you input anything client-related. |
The Mata v Avianca warning
In 2023, a US court sanctioned lawyers who submitted a brief containing AI-generated citations to cases that did not exist. The lawyers had not verified the citations. The court was not sympathetic.
This case is now the most-cited example of what happens when lawyers treat AI output as a final product rather than a starting point. It is not a cautionary tale about AI. It is a cautionary tale about professional responsibility — which has not changed simply because the tool has.
Section Four: Three questions to ask before using AI output
You do not need a complex checklist. You need three questions — applied consistently, every time you are about to rely on something AI has produced.
01 — Is this legally accurate or does it just sound accurate?
AI output sounds authoritative because it is trained on authoritative sources. That does not mean any given piece of output is correct. Before you use a legal proposition, a case summary, or a statutory interpretation produced by AI, ask yourself: have I verified this against a primary source, or am I trusting the tone?
Apply this especially to: case citations, statutory provisions, jurisdiction-specific rules, and any statement about what the law ‘requires’ or ‘permits’.
02 — What has AI assumed that I haven’t told it?
AI fills gaps. If your prompt does not specify the governing law, AI will assume one. If you do not specify the parties’ relationship, AI will infer it. If you do not specify the commercial context, AI will generalise. Every assumption AI makes without your instruction is a potential gap in the output.
Before using a draft, ask: what would AI have needed to know to get this exactly right — and did I tell it?
03 — What would I miss if I relied on this without reading it carefully?
The question is not whether to read AI output. You must always read it. The question is what to look for when you do. The most common errors are not random — they cluster around citations, jurisdiction-specific rules, omissions, and overstatements of certainty.
Read AI output as you would read a draft from a junior: with appreciation for the effort and with your own professional eye firmly engaged.
Section Five: What good AI practice looks like
Good AI practice in legal work is not a set of rules. It is a disposition — a way of approaching the tool that keeps professional judgement at the centre of every task.
Here is what it looks like in practice.
AI is infrastructure, not a replacement for judgment
Think of AI the way you think of a legal database. Westlaw does not give you legal advice — it gives you raw material that you transform into advice through your training, your judgment, and your knowledge of the client. AI is similar. It gives you faster raw material. What you do with it is still entirely yours.
You set the task. AI executes it.
The lawyer who gets the most from AI is not the one who asks the broadest questions. It is the one who knows exactly what they need, frames the task precisely, and evaluates the output critically. This is the skill of working with AI well — and it is a legal skill, not a technical one.
You are responsible for everything you produce
AI does not sign the advice. You do. AI does not owe a duty to your client. You do. AI cannot be disciplined by the SRA, the Bar Standards Board, or the Bar Council of India. You can.
This does not mean you should not use AI. It means you should use it in a way that you would be comfortable defending to a client, to a court, or to a regulator.
Build a consistent workflow, not ad hoc experiments
The risk of learning AI by trial and error is not that you will make mistakes. It is that you will make inconsistent mistakes — sometimes getting it right, sometimes not, without knowing which is which. A consistent workflow — knowing which tasks you use AI for, how you prompt it, and how you verify the output — is what converts a useful tool into a reliable one.
Stay current
AI tools are changing fast. The tool that hallucinated regularly six months ago may be significantly more reliable now. The tool that seemed safe may have introduced new risks. The professional obligation to keep your knowledge current now extends to the tools you use to practise.
This guide covers the map. The AI Bar teaches you the terrain.
If you want to move from knowing AI’s limits to working within them confidently — with structured training, real legal tasks, and worked examples from practice — that is what The AI Bar is built for.
Practical AI modules for lawyers, built for common law practitioners in the UK, Singapore and India. Self-paced. On-demand. Built by lawyers who use these tools in active practice.
theaibar.in — Where lawyers learn to practise with AI.