A solicitor at a five-partner practice shared something quietly revealing at a client meeting earlier this year. She had been using an AI tool to draft correspondence for three months, but nobody had told her which tools were approved, what the firm’s position was on client data, or what the SRA expected. She found out when a drafted letter went to a client containing a legal citation that simply did not exist.
That scenario is playing out across owner-managed UK legal practices. The tools arrive before the governance does, and the training rarely happens before something goes wrong.
What does “practical AI training” mean for a legal team?
Practical AI training for a legal team means giving your lawyers, fee-earners and support staff the knowledge to use AI tools within your professional obligations. It covers which tools are approved, how to write effective prompts, how to verify outputs, and what to do when something goes wrong. For a firm of five to fifty, this is internal sessions and written policies built around your actual work.
The Law Society’s 2024 guidance on generative AI says firms should “build AI competency” across their lawyers, covering the limits of tools, how to verify outputs, prompt design and bias risks. The emphasis is on practical capability, not theoretical awareness.
UCL offers a formal “AI for Lawyers” short course, priced between £1,950 and £2,950 per delegate. That benchmark tells you something useful: structured AI training for legal professionals is now a mainstream expectation in the profession. The goal for an owner-managed practice is to ensure that level of knowledge exists inside the firm, whether you build it through internal sessions, an external course for one or two people, or a combination of both.
For a practice of this size, a documented programme serves two purposes: it builds capability and it demonstrates to the SRA that you have systems and controls in place.
Why do UK legal regulators expect your team to be trained?
The SRA’s November 2024 Risk Outlook classified generative AI as a strategic risk, warning that poor supervision of AI-assisted work could lead to negligence, confidentiality breaches and regulatory action. The Bar Standards Board issued guidance in July 2024 stating that barristers using AI remain personally responsible for accuracy and must not feed confidential client information to tools where it may be stored or reused.
Three regulators have now put this in writing. The SRA says firms must ensure AI use does not compromise duties of competence, confidentiality and supervision. The BSB says AI does not change personal responsibility for the work product. The Law Society says building AI competency is a standard expectation for solicitors.
The US case of Mata v Avianca in 2023 sits behind much of this attention. Lawyers submitted a brief containing fabricated case citations generated by ChatGPT and were sanctioned by the court. The Law Society and BSB both cite this case as a reason that AI output cannot be treated as an authoritative source.
The ICO adds a further layer. Its 2023 guidance on AI and data protection requires organisations to train staff on what can and cannot be entered into AI tools, given the risk of pasting personal data and breaching UK GDPR. A Stanford HAI study from 2024 found that major language models produced incorrect legal citations in a significant proportion of tests, with error rates above 60% in some jurisdictions.
At a regulatory level, documented training is the mechanism by which your firm demonstrates it has appropriate systems and controls in place.
Where will your team actually meet AI in their daily work?
AI surfaces for legal teams in four places: drafting first-pass correspondence, summarising long contracts and judgments, research scaffolding to identify the relevant area of law, and handling administrative tasks such as attendance notes and meeting summaries. The Law Society identifies these as the practical near-term use cases. Training built around these specific tasks lands better than training about AI in the abstract.
A Thomson Reuters survey of legal professionals found that lawyers expect to save an average of 12% of their time using generative AI for drafting and research, with early adopters reporting up to 30% time savings on suitable tasks. That efficiency case is real. The saving only materialises when the team knows how to use the tools correctly and check the outputs.
Practical training should be built around those four touchpoints, with real examples from your firm’s work. For a conveyancing practice, that means a session on using AI to produce a first-pass NDA mark-up and then reviewing it clause by clause. For a commercial firm, it means using AI to summarise a long judgment and then verifying every case it references against Westlaw or Lexis.
A structured fact-check checklist, one that requires lawyers to verify all cases and statutes via an authoritative legal database before any document leaves the firm, should be part of every session from the first one.
What does a starter training programme actually look like?
For a firm of five to fifty people, a practical programme runs over four to six weeks and totals around four to six hours per fee-earner. The UK Government’s AI Playbook structures training around three groups: senior leaders, practitioners and support staff. In a small legal firm, those map to partners and the COLP, fee-earners, and secretarial or paralegal staff.
Here is what each group needs.
Partners and the COLP need two sessions of around sixty minutes each, covering regulatory exposure, what an acceptable-use policy must include, which tools the firm will approve, and how to assess AI vendor contracts for data residency and training-data terms. The Law Society advises that senior lawyers need enough technical understanding to ask the right questions of AI vendors, even if they are not developers themselves.
Fee-earners need three or four hands-on sessions covering prompt writing for the task types your firm handles, output verification, the fact-check checklist, and a thirty-minute case-study module. That module should cover Mata v Avianca (where fabricated AI-generated citations led to court sanctions), the Samsung incident (where staff pasted sensitive internal data into a public AI tool, triggering fears of data leakage), and what each incident means for your practice specifically.
Support staff need one or two sessions on safe use of AI for admin tasks: what can and cannot be entered into any tool, how to handle document summaries, and where the confidentiality rules apply just as much as they do for fee-earners.
On cost, an internal programme led by a partner or COLP has a low cash outlay. Across a thirty-person practice, the time investment totals roughly a working week in aggregate. Sending one or two people on an external course and having them cascade the learning internally is a cost-effective alternative for firms that want structured external content without the full cost per head.
What else needs to sit alongside the training?
Training on its own does not close the regulatory gap. The SRA and the ICO both expect supporting documentation: an AI acceptable-use policy that sets out which tools are approved, what can be entered into them, and where human review is mandatory. The NCSC advises that enterprise tools with contractual data-protection guarantees are the right default for any firm handling client information.
A workable policy covers four things: which tools are approved and how to access them securely; what client-identifying information cannot be entered into any public tool; the requirement to check all AI-generated content before it leaves the firm; and what to do if something goes wrong.
For firms with cross-border work, the EU AI Act is worth a mention. It classifies AI systems that assist in interpreting facts or law as high-risk when they influence judicial or administrative decisions, and providers are already updating their documentation accordingly. Your training should flag this trajectory for any matters involving EU clients.
Review the policy and training content quarterly. Update your approved-tool list, prompt templates and acceptable-use rules based on what your team is encountering in practice. The UK Government AI Playbook emphasises continuous iteration in AI adoption, and legal teams are no different. What works in month one rarely covers everything your team will encounter by month six.
If you are thinking about where AI fits in your firm and want to talk through what a practical rollout looks like, Book a conversation.
