AI for UK law firms: a practical guide to what's worth using

AI is worth using in a UK law firm where the output is checked by a person and the job is repetitive: answering and qualifying enquiries, keeping clients updated, summarising, first drafts of routine correspondence and searching the firm's own material. It is risky wherever it produces legal authority or advice nobody verifies. The SRA permits AI; the solicitor stays accountable for everything it does.

That is the short answer. The rest of this guide is the longer one, written for the partner or practice manager who has had four vendor emails this week and wants a way to sort them. We build software for the front of a law firm, so we have a view, and we will say where it colours ours.

What does 'AI for law firms' actually cover?

The phrase covers products that have almost nothing in common apart from a language model somewhere inside. The most useful way to sort them is by the job they do and by what they touch, because that decides the risk, the price and who in the firm should own the decision.

The SRA's own Risk Outlook on AI uses a similar sort, borrowed from LawtechUK: risk identification and prediction, administration (including chatbots that gather information from potential clients and triage for urgency), profiling, search, and text generation. It also notes that a single system can do several of these at once. We find a slightly different cut more useful in practice, based on what the tool works on.

• Tools that work on documents. Drafting, review, due diligence, research and summarising bundles. This is the category Harvey, Legora and CoCounsel sit in. The input is the artefact of legal work, and the output is usually more legal work product.
• Tools that work on the relationship. Answering the phone, qualifying an enquiry, booking the call back, covering a fee-earner in court, keeping a client informed about where their matter is. The input is a person trying to reach the firm, and the output is a record and a next step.
• Tools that work on the firm itself. Time capture, billing, reporting, practice management features with AI added. Mostly incremental improvements to software you already own.
• General-purpose assistants. ChatGPT, Copilot, Gemini, Claude and the rest, used by individuals for whatever they choose. Often the largest actual use in a firm, and the least governed.

Each of those has its own risk profile. A document tool that invents an authority can end up in front of a judge. A front-door tool that invents an answer can end up in a client's ear. A general-purpose assistant used on a personal account can end up holding client data the firm never agreed to share. Treating them as one category called 'AI' is how firms end up with a policy that is too strict for the harmless uses and too loose for the dangerous ones.

Is AI allowed in UK law firms?

Yes. The SRA says plainly that it regulates outcomes rather than technologies, and that it does not specify which systems firms should or should not use. In May 2025 it went further and authorised Garfield.Law, the first purely AI-based firm to provide regulated legal services in England and Wales. We cover what that decision did and did not say in the first SRA-approved AI law firm.

What does not change is accountability. The SRA's announcement said that named regulated solicitors remain responsible for all the system's outputs and for anything that goes wrong. The Code of Conduct obligations on competence, confidentiality and not misleading clients apply to work done by software exactly as they apply to work done by a trainee. None of this is legal advice, and your COLP should be part of any decision about a specific tool.

Where is AI worth using in a law firm right now?

The test we use is simple. A use is worth it when the task is frequent, the output is easy for a person to check, and the cost of a mistake is small or caught before it reaches anyone outside the firm. Run that test and a short list falls out.

• Answering and qualifying enquiries. The work is repetitive, it happens when nobody is free, and the output is a summary a fee-earner reads before calling back. In our experience missed calls are the most common leak in a small firm, and this is the use where we build. See AI chatbots for law firms for what such a system can safely say.
• Covering a fee-earner who is in court or in a meeting, and keeping clients updated on where their matter stands. Clients mostly ring to ask what is happening. That question has an answer on the file, and a person should decide what gets shared.
• Summarising long material the firm already holds, as a first read that a lawyer then checks against the source.
• First drafts of routine correspondence, where the fee-earner edits before anything is sent.
• Internal search across precedents and know-how, provided the tool shows where each answer came from.

Notice what is not on the list: producing case law without a human checking every citation, telling a client their prospects, calculating a limitation date for a client to rely on, or anything that acts in the outside world without somebody approving it. Those are not impossible uses, but they are where the risk concentrates, and they need a much heavier process around them.

For a day-to-day view from the fee-earner's desk, see AI for solicitors. For a firm of fewer than ten people, AI for small law firms is the shorter version of this guide with a budget attached.

What are the risks of AI in law firms?

Four risks come up in every serious conversation. They are not equally likely, and they do not apply equally to every category of tool.

• Made-up output. Language models produce fluent text whether or not it is true. In Ayinde v London Borough of Haringey [2025] EWHC 1383 (Admin), the Divisional Court dealt with fictitious cases put before the court, and in the linked Al-Haroun matter, eighteen of forty-five citations referred to cases that did not exist. We go through the judgment in AI hallucinations and law firms.
• Confidentiality. Client information pasted into a public tool on a personal account has left the firm's control. The October 2025 judicial guidance tells judges not to enter private information into public AI tools, and the same logic applies to anyone handling a client's affairs.
• Data protection. Enquiries often carry special-category data: health, family circumstances, alleged offences. The ICO's guidance on AI and data protection expects a lawful basis, transparency and, for higher-risk processing, an impact assessment.
• Misleading people. A caller who thinks they are speaking to a person, and is not, has been misled. Disclosure has to be built in, not left to a prompt.

The judgment in Ayinde also said something that changes who in the firm owns this. It said practical and effective measures must now be taken by those with leadership responsibilities, naming managing partners and heads of chambers, and that courts can be expected to ask whether those responsibilities were met. That makes AI governance a partner's job, not the IT supplier's. AI governance for law firms has a one-page policy you can adapt.

How do you choose AI tools for a law firm? A five-question filter

Most vendor demos are built to impress. These five questions are built to find out what happens on a bad day, and a vendor who is worth talking to answers each one concretely.

• What does it do when it does not know? A good answer is a refusal and a handoff to a person. A bad answer is a confident paragraph.
• Who checks the output before it reaches a client or a court? If the answer is nobody, the tool is acting on the firm's behalf and needs a much higher bar.
• Where does client data go, how long is it kept, and can the firm change that? Ask for the retention default in writing.
• Does it tell people it is AI? For anything client-facing, ask to hear the exact words and whether they can be switched off.
• What does it cost when usage doubles? Per-seat, per-minute and per-matter pricing behave very differently as a firm grows.

The document-AI vendors are large and well funded, and they are often the first names a partner hears. We explain one fairly in what Harvey AI is and who it is for. The short version is that document tools and front-door tools solve different problems, and a firm can reasonably want one, both or neither.

What does 'agentic' AI mean for a law firm?

An agent is a system that takes steps rather than just producing text: it looks something up, fills in a form, sends a message, books a slot. The word is being stretched in marketing to cover almost anything. The question that matters is which steps it may take without a person approving them. We set out a way to classify that in AI agents for law firms.

Garfield.Law is a useful reference point here. According to the SRA, its system is not autonomous, only takes a step where the client has approved it, and cannot propose case law at all. That is what the regulator was comfortable authorising: a narrow job, approval at each step, and the riskiest capability removed rather than managed.

Does AI work differently for different practice areas?

The tools are the same; the stakes are not. A conveyancing enquiry and a call from somebody leaving an abusive relationship both arrive on the same line, and the second needs a very different handling rule. Family, immigration, crime and personal injury all carry more special-category data and more distressed callers than commercial work does. We look at the most sensitive case in AI for family law firms, and the same principles carry across.

A 90-day plan for introducing AI in a UK law firm

This is the sequence we would follow in a firm of five to fifty people. It is deliberately slow at the start, because the first fortnight of any deployment teaches the firm more than the demo did.

• Days 1–10: find out what is already in use. Ask every fee-earner and every member of support staff, without blame, which AI tools they use and on what. The answer is almost never 'none'.
• Days 10–20: adopt a one-page policy that sorts uses into allowed, allowed with review, and not allowed. Name an owner. Tell staff which tools are approved and on which accounts.
• Days 20–30: pick one job with a clear before-and-after measure. Missed enquiries, response time to new callers, or hours spent on a named routine task are all good choices.
• Days 30–60: run it with a named reviewer who reads every output for the first two weeks, then a sample after that. Keep a log of every error and every near-miss.
• Days 60–90: compare against the baseline, decide to extend, change or stop, and write down why. Update the policy with what the pilot taught you.

A worked example, with numbers invented for illustration only: a hypothetical eight-fee-earner firm counts its unanswered calls for two weeks before starting and finds 60, of which it believes a third were new enquiries. It then runs an AI front door for four weeks and measures how many new enquiries became booked consultations. Whatever the result, the firm now knows something it can act on, which is more than most firms know about their own phone line. Our safe-AI checklist for client intake covers the governance questions for that specific pilot.

Where WasilX fits

We build for the relationship side, not the document side. Cam, our AI Engagement Coordinator, takes calls on the firm's line at any hour, says it is AI, asks the firm's own qualifying questions, books the call back and sends the firm a summary. A personal assistant for each fee-earner covers their calls while they are in court and drafts matter updates for the fee-earner to approve. We do not read or summarise your documents and use no case law, so we do not draft legal work or give legal answers. If that is the gap in your firm, we are happy to talk it through.

Frequently asked questions

Can UK solicitors use ChatGPT? Nothing in the SRA's rules bans a particular tool. The obligations are about outcomes: keep client information confidential, check anything that goes to a client or court, and stay competent. A common approach is to allow general assistants for non-client work and to require an approved, business-account tool for anything involving client information. Your COLP should set the line.

Does the SRA approve AI tools? No. The SRA says it does not recommend individual providers or software. It authorises firms, not products. Its 2025 authorisation of Garfield.Law approved a firm whose service is delivered through AI, with named solicitors accountable for the output.

What is the safest first use of AI in a law firm? One where a person checks the output before anyone outside the firm relies on it, and the data involved is not the firm's most sensitive. Enquiry handling that ends in a summary for a fee-earner, and internal summarising checked against the source, both fit that description.

Who in a law firm should own AI? A partner, usually alongside the COLP. The Divisional Court in Ayinde said managing partners and others with leadership responsibilities must take practical measures to make sure everyone providing legal services understands their duties when using AI.