ChatGPT for lawyers is useful for first drafts, summaries, attendance notes and plain-English letters, as long as a fee earner reviews the draft. It isn't safe for legal research on its own, because it invents cases. Client information should only go into a business account that doesn't train on your data.
Here is where generative AI helps inside a UK firm, where solicitors run into trouble with it, and what the SRA and Law Society expect.
Use it for drafting and summarising material you supply. Don't rely on it for unchecked authorities.
Use a business account, write an AI policy, and keep a fee earner signing off drafts before they leave the firm.
What ChatGPT is good for in a law firm
It is most useful on tasks a fee earner can check in a few minutes:
- First drafts. A letter before action, a client update or a first cut of a lease clause, written from a precedent you paste in.
- Summaries. A 60-page disclosure bundle or a long email chain turned into a chronology you can check against the source.
- Attendance notes. A call transcript turned into a note in your firm's format, with the actions pulled out.
- Plain English. Explaining a completion statement or a TA6 answer to a first-time buyer.
- Spotting what's missing. Checking a contract against your playbook and flagging clauses that depart from it.
In each of these tasks the facts come from you, so ChatGPT is rewriting, condensing or comparing material you've supplied. That is where it tends to be most reliable.
Is ChatGPT any good for legal research?
Not on its own. Large language models predict plausible text, and a plausible case name with a plausible citation is what they often produce when they don't know the answer. The SRA noted in its 2023 Risk Outlook on AI that these systems don't have a concept of truth.
The courts have already dealt with the fallout. In Ayinde v Haringey and Al-Haroun v Qatar National Bank (June 2025), the Divisional Court addressed submissions citing cases that didn't exist or didn't say what they were quoted as saying. In Al-Haroun, the solicitor had relied on research done by his client, and the court referred him to the SRA. The judgment states bluntly that free tools like ChatGPT "are not capable of conducting reliable legal research", and that checking authorities remains the lawyer's professional responsibility.
Specialist tools are better, though not infallible. Stanford researchers tested AI legal research tools in 2024 and found Lexis+ AI and Ask Practical Law AI gave incorrect information more than 17% of the time, and Westlaw's AI-Assisted Research more than 34%. The vendors disputed parts of the methodology, but the practical takeaway holds. AI research tools only give you a starting point, and cited authorities still need checking in Westlaw, Lexis or BAILII before going into advice or submissions. I compare the research tools and the wider UK market in the best AI for lawyers.
ChatGPT, Copilot or a legal AI tool?
Most UK firms end up choosing between three routes:
- ChatGPT Business or Enterprise. Strong at drafting and summarising, and cheap per seat. It knows nothing about your precedents or matters unless someone sets that up.
- Microsoft 365 Copilot. Works inside Outlook and Word, and keeps data in your Microsoft 365 tenant. Many firms already pay for it and barely use it, because fee earners were never shown practical ways to work with it.
- Legal AI like Harvey, Legora or Lexis+ AI. Built for legal work and, in Lexis's case, grounded in a proper research database. Strong products, though mostly aimed at large firms, and they still require a proper rollout.
For a 10 to 50 person firm, I think the tool matters less than the setup. The firms getting real value from any of them have connected the model to their own precedents, written down how specific tasks should run, and trained fee earners to review the output.
What the SRA and the Law Society expect
Neither bans ChatGPT. Both apply the same professional obligations they always have to a new tool.
- You stay responsible. The Law Society's October 2026 practice note on using AI says professional responsibility can't be handed to an AI system. Even if AI produced the first draft, the fee earner who sends it remains accountable.
- Supervision. The SRA expects firms to supervise AI systems and how staff use them. For a small firm that means a written AI policy your COLP owns. The SRA's own position is set out in our SRA AI guidance summary.
- Confidentiality. Client information should only go into business accounts with no training on your data. OpenAI says ChatGPT Business, Enterprise and API data isn't used for training by default, whereas personal accounts are different.
- Billing and transparency. The practice note says firms shouldn't bill for time a task would have taken without AI, and clients should be told how AI is used on their matters.
OpenAI's own usage policies also rule out tailored legal advice without a licensed professional involved. That aligns with how firms ought to use it anyway, strictly as a drafting aid for a solicitor.
How to set ChatGPT up for a 10 to 50 person firm
Most firms I speak to either ban ChatGPT altogether or leave staff using personal accounts with no shared rules. Neither works well. A better rollout looks like this:
- Pick one business AI account for the whole firm. If you're on Microsoft 365, Copilot may already be licensed.
- Write a short AI policy covering approved tools, permitted data, review responsibilities and client disclosure. Our AI policy template for law firms is free to copy.
- Give it your own material. A model working from your precedents and house style drafts far better than one guessing from the open internet.
- Build a review step into the workflow, especially where any authority is cited.
- Start with one workflow, like attendance notes or first drafts, and track the hours saved.
- Train people on real matters from that workflow rather than abstract demos. I've compared the AI training providers for UK law firms.
That's what we set up with AI for law firms. We configure AI inside Outlook, Word and your practice management system, whether that's Clio, Actionstep, LEAP, iManage or NetDocuments, and draft an AI policy your COLP can adapt and own. Client data stays in your own Microsoft 365 tenant. The two most common starting points are first drafts from your own precedents and contract review against your playbook.
The first Task Kit is ready within 7 days, backed by a full refund if you don't see an ROI within 90 days. Pricing is a flat monthly fee based on firm size, which we go through on an intro call before confirming the exact figure in your AI opportunity audit. Firms that want intake, drafting and matter updates joined up move on to an AI operating system for law firms, rolled out one team at a time.
