A hallucinated case citation has already reached a UK courtroom more than once, and the SRA has put its name to a warning notice about it. AI in a law firm is a real time-saver and a real liability at the same moment. The difference between the two is training.
The summary
Yes, lawyers can save hours a week with AI on research triage, first-draft correspondence, document review and attendance notes. The tools that fit are private, firm-controlled environments like Harvey, Lexis+ AI and CoCounsel, not a public chatbot with client data pasted into it. The risks are specific: fabricated citations, privilege, and the SRA holding the fee-earner accountable for the output. Train for those three and the upside is safe to take.
What can lawyers actually use AI for?
Most legal work is reading, drafting and checking, and AI is quick at the first two while leaving the checking firmly with you. Used well, it takes the blank-page time out of a task and hands you a first draft to correct rather than write from scratch.
| Everyday task | How AI helps |
|---|---|
| First-draft correspondence and attendance notes | Turns your bullet points or a call recording into a clean first draft you then check and sign off. |
| Legal research triage | Summarises long materials and points you at the parts worth reading in full, so you verify faster. |
| Document and bundle review | Runs a first pass over disclosure or a data room to surface the clauses and dates that matter. |
| Contract mark-up and clause comparison | Compares versions and explains the commercial effect of each in plain English. |
| Chronologies and case summaries | Builds a first-draft timeline from the papers for you to correct against the file. |
Which AI tools suit lawyers?
The tool matters far less than where your data goes. For anything touching a live matter, the answer is a private, firm-controlled environment, never a personal chatbot account.
| Tool | What it is used for | What to watch |
|---|---|---|
| Harvey | Legal drafting, research and review inside a private, firm-controlled environment | Still needs a fee-earner to check every citation and conclusion. |
| Lexis+ AI | Legal research grounded in a known, licensed content set | Verify against the primary source; grounding lowers error, it does not remove it. |
| CoCounsel (Thomson Reuters) | Reviewing, summarising and drafting against your own documents | Confirm it has read the correct, current version of each document. |
| Microsoft Copilot | Email, Word and Teams admin inside your firm's Microsoft tenant | Your data governance settings decide what it can see, so set them first. |
| ChatGPT or Claude (Team or Enterprise) | General drafting and thinking on non-confidential matters | Never paste client or matter data into a consumer account. |
Three frameworks for training lawyers on AI
Generic AI training does not stick in a law firm, because a conveyancer and a litigator do not have the same day. These three frameworks are how we make it land.
1. Role-based upskilling
Role-based upskilling starts from the actual work of a team, not from a feature list. We look at what a department does every week, find the tasks where AI genuinely saves time, and train on those with real matter types rather than toy examples.
Act as an expert legal-sector trainer. Analyse the weekly workflow of our conveyancing team and identify the top 5 tasks where generative AI can cut manual effort by 30%. Give a 4-week practical training outline with tool recommendations, hands-on exercises using our real matter types, and measurable KPIs to track adoption.2. Managing change and resistance
The resistance in law firms is rarely about the technology. It is fee-earners worried the tool will make junior roles redundant, or nervous they will get the prompt wrong in front of a partner. The fix is psychological safety and a simple mental model, taught before any feature.
We are rolling out an AI drafting tool but our fee-earners fear it will replace junior lawyers and worry about getting prompts wrong. Design a change-management communication plan and a 1-hour introductory workshop agenda. Focus on teaching them to delegate to AI like a supervising partner rather than technical prompting, and outline how to handle the most common objections.3. Security, governance and ethics
The governance framework is where a law firm earns the right to use AI at all. Everyone needs to know the difference between a public and a private tool, exactly what client and matter data is never safe to paste, and how to spot a fabricated citation before it reaches a client or a court.
Create a clear, non-technical AI acceptable-use policy and a short quiz for a law firm training module. Teach the difference between public and private AI environments, what client and matter data is never safe to input, how to spot fabricated case citations, and how to fact-check every AI output against primary sources before it is used with a client or in court.Example AI prompts for lawyers
These are starting points, not finished answers. Every one of them ends with you checking the work.
| Task | Prompt you can adapt |
|---|---|
| Attendance note | Turn these notes from a client call into a formal attendance note in UK English, flag every action point and deadline, and list what I should follow up on. |
| Research triage | Summarise the key principles from the materials pasted below, set out the points for and against my client's position, and tell me what I still need to verify against primary sources. |
| Client care letter | Draft a plain-English client care letter explaining this issue to a non-lawyer, keep it under 400 words, and keep a neutral, reassuring tone. |
| Clause comparison | Compare these two versions of a limitation of liability clause, set out the differences in a table, and explain the commercial effect of each in plain English. |
| Plain-English rewrite | Rewrite this paragraph of legal drafting into plain English for a client, without changing its legal meaning, and flag anything that is genuinely ambiguous. |
What are the risks of AI for lawyers?
The risks below are not reasons to avoid AI. They are the exact things a good training day is built around.
| Risk | Why it matters | How training handles it |
|---|---|---|
| Fabricated case law | Made-up citations have reached UK courts and drawn judicial criticism and costs consequences. | Source-first verification: never cite what you have not read in the primary report. |
| Confidentiality and privilege | Pasting matter data into a public tool can breach the SRA Code and put privilege at risk. | Private environments only, with a plain data rule everyone can recite. |
| Over-reliance | The danger is delegating judgement, not just typing. | Human sign-off on every output before it leaves the firm. |
| Currency and jurisdiction gaps | Models miss recent law and mix up jurisdictions. | Keep a qualified lawyer in the loop and check the date and jurisdiction of everything. |
How should a law firm roll out AI?
The firms that get value do not buy a licence and hope. They start narrow, prove the time saved on one workflow, and widen from there with a policy already in place.
| Step | What it involves |
|---|---|
| 1. Set the guardrails first | Agree an acceptable-use policy and a private-tool decision before anyone is trained. |
| 2. Pick one workflow | Choose a single high-volume task, such as attendance notes, and train the team that owns it. |
| 3. Train hands-on | Run practical sessions on real matter types, with verification built into every exercise. |
| 4. Measure and widen | Track time saved and adoption, then extend to the next workflow once the first is safe. |
Want your fee-earners safe and quick with AI?
We train legal teams on the tools that fit a law firm, the SRA risks, and the habits that keep client data where it belongs. Practical, hands-on, and built around your real matters.
Take Our Free 2 Minute AI Skills Assessment
Harry Lang runs The Oxford AI School, teaching teams and businesses the basics of AI in a fun, easy and practical way. He can be reached at Harry@TheOxfordAISchool.com.
AI Training for Lawyers: FAQ
Can lawyers use ChatGPT for legal work?
For non-confidential thinking and drafting, yes, but never by pasting client or matter data into a consumer account. For live matters, a private, firm-controlled environment such as Harvey, Lexis+ AI or CoCounsel is the safe route, and every output still needs a fee-earner to check it.
Is AI allowed by the SRA?
There is no ban. The SRA has issued a warning notice on the misuse of AI, which makes clear that the fee-earner remains accountable for the output, that confidentiality must be protected, and that fabricated citations are a serious problem. Used inside those limits, AI is permitted and increasingly common.
Will AI replace junior lawyers?
It changes the work rather than removing the lawyer. AI does the first draft and the first pass; the judgement, the verification and the client relationship stay human. The junior role shifts towards supervising and checking AI output, which is a skill worth training for.
What is the biggest AI risk for a law firm?
Two, really: fabricated case citations reaching a client or court, and confidential material being pasted into a public tool. Both are training problems before they are technology problems, which is why the governance framework comes first.
How long does AI training for a law firm take?
A useful working level comes from a focused half-day or day on one or two workflows, not a term-long course. Most teams are saving time within a couple of weeks of practising on their own matters. That is the whole point of role-based training.