AI legal research is genuinely useful. It's also easy to use badly. This guide isn't about whether to use it, you probably should. It's about how to use it in a way that holds up professionally, because the difference between good and bad AI research practice comes down to a few specific habits.
Start with a precise question
AI research tools are better than search engines at understanding natural language, but they still respond to how clearly you frame the question. "What's the law on dismissal?" returns something generic. "What are the procedural requirements for fair dismissal under the Employment Rights Act 1996, and what constitutes a fair reason under section 98?" returns something useful.
The more specific you are about the area of law, the jurisdiction, and what you're actually trying to establish, the better the answer. Think about how you'd frame a question for a trainee who knew the law but didn't know your client's situation.
Always check whether the source is live
This is the most important habit. An AI tool that retrieves from live UK sources, legislation.gov.uk, BAILII, the SRA's website, gives you something you can verify. An AI tool that answers from training data gives you a plausible answer based on what the law probably said at some point in the past.
You can't always tell which one you're dealing with from the answer alone. Both produce fluent, confident responses. The difference is visible in the citations. Click through. Does the link go to a real page? Does the passage cited say what the AI says it says? Is the legislation version shown as currently in force?
This check takes about a minute and catches most problems before they become your problem.
Know where AI research is reliable and where to be careful
For well-settled areas of UK law, the current text of a statutory provision, standard procedural requirements under the CPR, established SRA regulatory obligations, AI research tools trained on UK sources are reliable and fast. They're genuinely better than manual database searches for straightforward questions because they synthesise rather than just retrieve.
The areas where you should apply more scrutiny: very recent decisions (an AI tool may lag a few days or weeks behind BAILII), highly contested areas where the outcome turns on fine judicial interpretation, and anything involving retained EU law or devolved legislation where coverage is patchier.
For those questions, AI research is a useful starting point, not an endpoint.
Save it to the matter
Research that exists only in a chat window isn't really part of the matter. If you're ever asked how you arrived at a piece of advice, you need to be able to show the research basis. That means saving research outputs, with their citations, to the matter record, not just using them and closing the tab.
This also protects you. The SRA's guidance on AI is clear that solicitors remain responsible for the research they rely on. If the research is in your matter record with verifiable citations, you've done what the guidance requires.
The professional judgment step doesn't change
AI gives you the research. You apply it to your client's specific facts. That part is yours.
An AI tool can tell you that section 15 of the Landlord and Tenant Act 1985 imposes an obligation to keep installations for the supply of water in repair and proper working order. It can cite the cases that have interpreted that obligation. It can't tell you whether that applies to your client's particular lease, whether the landlord has a defence, or what the realistic outcome of a Part 1 claim would be in your local court.
The judgment doesn't transfer to the tool. What the tool does is get you to the judgment step faster, with better information, and with citations you've actually verified.
Writford's Research mode retrieves from live UK legal sources and saves research to the matter with one click, so the verification step and the filing step are both quick.