For a small UK law firm, match the tool to the task: a UK-content research tool such as Lexis+ with Protégé for research, a business-plan assistant or Copilot for drafting and client emails, and a contract tool such as Genie AI for review. Check where data is stored, and keep a supervision record for every AI draft.
The SRA does not recommend products and judges firms on outcomes, so the safe approach is a short shortlist, a data answer in writing for each tool, and a named fee-earner who reviews every output. A tool whose vendor cannot say where client files are stored drops off the list, however good its demo.
Matching each fee-earner task to a tool and a price
Prices come from vendors' own pages as of November 2026. UK invoices are in pounds, so check the figure in your own account. Where a vendor sells by quote, the table says so, because an unpublished price is something to budget around.
| Task | Best-fit option | Price evidence | Check before use |
|---|---|---|---|
| Legal research on UK law | Lexis+ with Protégé (formerly Lexis+ AI). Alternatives: Vincent by vLex (part of Clio), Thomson Reuters products | Quote only | Run five questions you already know the answer to |
| Letters, attendance notes, file summaries | Microsoft 365 Copilot Business, or ChatGPT Business or Claude Team | Copilot $21 a user annual; ChatGPT Business and Claude Team $20 a seat annual, $25 monthly, two seats minimum | Training default, storage location, retention |
| Client emails and updates | Copilot in Outlook, or an assistant on redacted text; drafting built into your practice system | Included with Copilot; practice-system tiers vary | Wording that promises nothing |
| Contract review and drafting | Genie AI, Spellbook, or an assistant with your playbook | Genie Free $0, Pro $59 a month annual ($75 monthly); Spellbook by demo | Playbook fit, storage and sub-processors |
| Meeting and attendance notes | Teams recap with Copilot or Teams Premium; Otter | Otter Pro $8.33 a month annual, $16.99 monthly | Consent, where recordings live |
| Large-matter document review | Harvey, Legora | Quote only; aimed at larger teams | Whether volume justifies it |
The pattern across the table is that cheap tools handle writing, quote-only tools handle authority, and the risky part of each is the same: what happens to the client material you load. Read whether solicitors can use ChatGPT without breaching confidentiality before anyone in the firm pastes a client name into an assistant.
UK data location: what each vendor actually says
Data location is a practical question, not a magic word. A UK location can simplify your answer to a client or a data-protection adviser, but it does not replace contract terms about training, retention and sub-processors. Here is what the vendors' own pages say, with the gaps flagged.
| Vendor | What the page says | What to ask |
|---|---|---|
| Microsoft 365 Copilot | Advanced Data Residency covers Copilot and Copilot Chat, lists the UK as a local region, and applies to Business Basic, Standard and Premium. It is a separate add-on that must cover every paid licence in the tenant, and the tenant's default geography must be the UK | Price of the add-on for your headcount, and whether your tenant is eligible |
| ChatGPT | OpenAI's business pricing comparison marks data residency "No" for Business and "Yes" for Enterprise, with the UK among ten storage regions. Its help article offers the choice to eligible API customers and new Enterprise and Edu workspaces. In-region model processing covers three regions, none of them the UK | If you need UK storage, ask OpenAI's sales team what an Enterprise workspace would cost at your size |
| Google Workspace | Data regions, from Business Standard up, keep data in one of two named regions, one of them Europe. No UK option is listed | Whether the Europe region meets your own data-protection advice |
| Genie AI | Documents are stored in UK and EU data centres on Amazon Web Services, never used to train models, with zero-data-retention agreements for third-party model providers. ISO 27001 certified; no SOC 2 report | Which model providers sit behind your plan |
| Lexis+ with Protégé | The UK product page says primary and back-up locations are in Europe and that customer data is not used to train public AI models. It does not name the UK | Whether your data stays in the UK, plus retention and sub-processors |
| Spellbook, Otter | The pages reviewed state no storage region. Spellbook cites SOC 2 Type II and zero-data-retention agreements that stop its model providers training on your data | Storage region, retention, sub-processors and deletion |
If a vendor cannot name a region, treat that as a finding. Whether a transfer outside the UK is acceptable depends on your data-protection obligations and the mechanism the vendor uses, so ask your data-protection adviser before you rely on any answer here. For the broader checks on suppliers, AI software for solicitors: data, supervision and supplier checks goes through the contract points.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowResearch: test a UK research assistant on questions you can mark
Research is the one task where a general assistant should not be your authority. LexisNexis describes Lexis+ with Protégé as combining its case law, legislation, secondary sources and Practical Guidance with conversational search, drafting, summarising and analysis of uploaded documents, and it connects to document management systems such as iManage, SharePoint and NetDocuments. Vincent by vLex, now part of Clio, draws on a library of over a billion documents from more than 100 countries, so check how deep its UK coverage goes in your practice areas. None of these publishes a UK price, so the test below is how you decide whether the quote is worth paying.
Choose five questions from the last year where you know the answer: a limitation point, a notice requirement under a standard form, a procedural deadline, a recent change in guidance and a point of statutory interpretation. Score each answer on four lines, one point each:
- It found the authority you found.
- It found something relevant you missed.
- Every cited source resolved when you opened it.
- It stated the rule for the right UK jurisdiction.
Twenty points available. Under 14, do not pay. Between 14 and 17, ask for a longer trial on live matters. Eighteen or more, negotiate the quote. A single unresolved citation costs the whole question, because in court work one invented authority is enough. The routine in how to stop AI inventing case law turns this into a daily habit.
Drafting and client communication: prompts, outputs and fixes
This is where a business-plan assistant or Copilot earns its price. The pattern that works is facts in, structure specified, gaps marked, and a fee-earner reading the result as if a trainee wrote it.
An attendance note from your own jottings
Turn these jottings into an attendance note. Headings: attendees, instructions, advice given, next steps, deadlines. Use only what is below and mark gaps [CONFIRM]. Jottings: client [first name], wants to sell the freehold of a rented flat; told her the lender's consent is needed; she will send the lease by Friday; I will request the title by Monday; fee estimate to follow in writing.
Illustrative output: a five-heading note, with "advice given" recording that lender consent is required, a Friday deadline for the lease, a Monday action for you, and [CONFIRM] against the fee estimate. What you fix: check that "advice given" says only what you said, add the time and date, and delete any sentence that records advice you did not give. The assistant may add "client understood the risks", which you cannot assume.
A client update that promises nothing
Draft a 130-word email to [first name]. Facts: title documents requested on Monday; the seller's solicitors usually take up to ten working days; I will call when they arrive. Tone: calm, plain. Do not predict a completion date or say the sale will complete.
Before: "Everything is on track and should complete in four weeks." After: "We requested the title documents on Monday. The seller's solicitors typically take up to ten working days, and I will call you as soon as they arrive. I cannot give a completion date yet." The second version ties every statement to a fact you gave. A client reads "on track" as a promise.
A probate letter before and after
Before: "We write further to our previous correspondence regarding the above estate and would be grateful for your urgent response." After: "We are writing about the estate of [name of deceased]. To close the account we need a letter confirming the balance at the date of death. Please reply by 14 March; the enclosed form has the details." The assistant's useful habit is putting the request, the deadline and the next step in the first three lines. Your edits: confirm the date, the enclosure and the exact wording of the request.
Keep the draft and your sent version of the first ten letters side by side. The edits you make again and again (a softer opening, the deadline moved up, a reference line added) become the standing instructions you give the assistant next time, and the number of edits per letter tells you whether the tool is saving time at all.
Five phrases to cut from AI-drafted client emails
Assistants write reassuring prose, and reassurance is where firms get into trouble. Search every draft for these and replace them with a fact, a date or a next step.
- "On track". Replace with what has happened and what is outstanding.
- "Should be straightforward". Nothing in a file is straightforward until it completes; say what you have seen so far.
- "I am confident that". Replace with the basis for the view, or delete.
- "We will complete by". Give a range only if you control every step, otherwise say what the date depends on.
- "You do not need to worry". Clients hear it as a guarantee. Say what you are doing about the risk.
A sample first pass on a supplier agreement
For a contract tool, test it on an agreement you have already reviewed. Say a small manufacturer sends a supplier agreement and you run it against your firm's short playbook: payment within 30 days, liability capped at the contract price, retention of title clear, either party able to terminate on 30 days' notice and a governing-law clause that matches the client's location.
| Clause | Illustrative tool output | What you check by hand |
|---|---|---|
| Payment | Flagged: 60 days against a 30-day playbook | Whether late-payment interest is excluded in the schedule |
| Liability | Flagged: cap at 12 months of fees, with an uncapped indemnity | Read the carve-outs; the cap may not cover the indemnity at all |
| Retention of title | Not flagged | Whether the clause survives mixing of goods and covers proceeds of sale |
| Termination | Flagged: supplier only, 90 days | Whether the client has a separate right for breach |
| Governing law | Flagged: different law from the playbook | The jurisdiction clause as well; they can differ |
As ever, the tool is good at matching your playbook and weaker on the interplay between clauses. Use it as a scanning aid, record which flags it raised and which you added, and over a month you learn what it reliably catches in your type of contract and what it never does.
SRA supervision checks that make AI use defensible
The Law Society's generative AI guidance is clear that a solicitor's duties to the court and the client apply to work done with AI, whether it was used by the solicitor personally or by anyone under their supervision. It ties misuse to several Code of Conduct requirements: not misleading clients or the court (paragraph 1.4), maintaining competence (paragraphs 3.2 and 3.4) and supervising work for the client (paragraph 3.5(b)). It also describes the 2025 High Court ruling in the Ayinde case, on invented citations, and an April 2026 decision in which a judge referred two solicitors to the SRA after AI-assisted court documents contained inaccurate case citations.
The SRA's risk outlook on AI adds that firms remain accountable for AI outputs, must protect confidentiality and privilege, and should make sure clients are suitably informed about AI's role. In practical terms, that becomes six checks:
| Check | What you do | Evidence to keep |
|---|---|---|
| Approved tools only | A written list of AI tools staff may use, with plan names | The list, dated, with the contract terms |
| Named reviewer | Every AI draft is read by a fee-earner before it leaves | A line in the file note naming the reviewer |
| Citation verification | Every authority pulled and read from a research service | A log of what was checked and where |
| Client information | A sentence in the client care letter about AI assistance | The signed letter |
| Confidentiality | No client identifiers in consumer tools; training off or contract in place | Vendor answers on training and storage |
| Competence and training | A short session for every fee-earner and support staff member | Attendance and a one-page crib sheet |
The firm-wide framework sits in the SRA AI guidance checklist for small firms and the pre-use questions in what solicitors should check before using AI. For the policy itself, take advice from your compliance officer or insurer, because this tutorial is not legal advice.
A supervision note you can copy
AI use note (attach to file)
Matter: [client matter number]
Task: attendance note / client email / clause comparison
Tool and plan: [name, plan, training setting]
Input: redacted facts only (yes/no)
Reviewer: [fee-earner initials], date reviewed
Authorities cited: none / list, each pulled from [service] on [date]
Changes made: [summary of edits]
Client informed of AI assistance: yes / no (care letter dated)
Filled in for a lease-extension query, the note takes under two minutes and gives you a record if anyone asks how the output was checked. The effort is small and the value shows when a client query or an insurer's question arrives months later.
A budget for a six-person firm in wills, probate and conveyancing
Take an illustrative firm with four fee-earners and two support staff. Monthly volumes: 40 client emails, 25 attendance notes, 12 probate letters, 6 contract or lease reviews and 15 research queries.
| Job | Minutes by hand | Minutes with a draft | Saved |
|---|---|---|---|
| 40 client emails | 10 each, 400 | 4 each, 160 | 240 |
| 25 attendance notes | 20 each, 500 | 8 each, 200 | 300 |
| 12 probate letters | 30 each, 360 | 15 each, 180 | 180 |
| 6 contract or lease reviews | 60 each, 360 | 35 each, 210 | 150 |
| 15 research queries | 30 each, 450 | 20 each, 300 | 150 |
| Total | 2,070 | 1,050 | 1,020 (17 hours) |
At an illustrative internal cost of $90 an hour, 17 hours is $1,530 a month. The spend: Copilot Business for six users at $21 is $126 a month on annual billing, and Genie Pro for one reviewer is $59, so $185 a month before any research quote. If the firm adds a research subscription at several hundred dollars a month, the research row saves 150 minutes, which is worth $225. That is the number to hold against the quote; research tools justify themselves by authority and accuracy, not by minutes saved. The time saved in research queries also assumes you verify every authority, so the minutes with a draft include checking.
If the firm chooses ChatGPT Business or Claude Team for six users at $20 a seat annual, the assistant cost is $120 a month, with a training default of off. The same six-person firm would pay $150 a month on monthly billing. Add the Genie seat, and the stack is $179 or $225 depending on billing, in the same range as the Copilot route.
A realistic mistake: the lease summary that dropped a condition
Illustrative scenario. A fee-earner asks an assistant to summarise a commercial lease for a client call. The summary states: "Tenant has a break option at year five." The lease says the break is exercisable only if the tenant gives six months' notice, has paid all rent and has no outstanding breaches. The client hears "I can leave at year five" and plans accordingly. The error surfaces when the tenant is in arrears at year four.
The control that would have caught it is the supervision note plus a rule for summaries: any summary of a right, option or deadline is checked against the clause text before it goes to the client, and the clause number is quoted in the email. A summary that cannot cite its clause is not ready to send. For more on spotting what AI misses in agreements, read AI contract review for small firms.
A sixty-day rollout for a small firm
Days one to ten: choose the assistant plan, switch off training or set up the business workspace, write the approved tools list and obtain vendor answers on storage and sub-processors. Days eleven to thirty: train fee-earners in one hour, start the supervision note on every AI-assisted file, and measure minutes on emails and attendance notes. Days thirty-one to sixty: trial the contract tool on two live reviews, run the five-question research test with a vendor trial, and decide what to buy.
At day sixty, list what you pay, what each tool replaced and what you would cancel. If you would like help running that process for your firm, that is what my AI implementation consultation covers; get in touch when you are ready.
Questions UK firms ask before choosing
Does the SRA say which AI tools solicitors may use?
No. The SRA says it regulates on outcomes, does not specify technologies and cannot recommend individual providers. Your duties to clients and the court apply whatever you use, so the work is choosing tools you can supervise and showing how you checked their output.
Do I need UK data storage for client files?
Not as a rule in itself, but it makes transfers simpler to explain. Data-protection law governs moving personal data abroad, so ask your data-protection adviser. Whatever the answer, get the vendor's storage location, sub-processors and retention in writing before client material goes in.
Can a business-plan ChatGPT replace a legal research tool?
Not for authority. A general assistant can structure an argument, but it has no licence-grade UK case law behind it, and it can invent citations. Use it for drafting and thinking, and use a research service for any authority you rely on or cite.
What should I tell clients about AI use?
The SRA's risk outlook says firms should make sure clients are suitably informed of how AI is involved. A line in your client care letter is the simplest route: AI tools may assist with drafting under a solicitor's supervision, and client material is not used to train models.
Further reads
- ChatGPT for Solicitors: Five Safe Tasks and Three to Avoid — Five safe ChatGPT tasks and three to avoid.
- Can AI Draft a Client Care Letter That Meets SRA Rules? — Drafting client care letters that meet the rules.
- AI Training for Solicitors' Firms: A One-Hour Team Session Plan — A one-hour team session plan on AI use.
- SRA Transparency Rules: Use AI to Draft Price and Service Pages — Price and service pages drafted with AI.
- How Small Law Firms Use AI to Draft Letters and Routine Documents — Routine letters and documents drafted faster.
- Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — Per-seat or pay-as-you-go legal AI pricing.
- Can Lawyers Use AI Note Takers in Client Meetings? — Recording and consent rules for client meetings.
- AI Time Capture for Solicitors: Recover Hours You Never Billed — Recover the hours AI-assisted work never got billed.
- AI for Barristers: A Chambers Checklist for Casework and BSB Rules — Turn BSB guidance into a practical chambers policy and a per-matter check before any AI-assisted work is used.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Law Society generative AI guidance, SRA risk outlook on AI, LexisNexis UK, vLex, Genie AI, Spellbook, Microsoft Learn, OpenAI business pricing and help centre, Google Workspace and Otter pages (checked 1 November 2026).