AI prompts for lawyers: a safer way to draft and review
Search for AI prompts for lawyers and you'll find endless lists of one-liners. Most are worse than useless for real legal work. Here's why — and what to use instead.
- Generic AI prompt lists encourage the exact behaviour that gets solicitors into trouble: trusting unverified output.
- A good legal prompt is grounded in your real documents and builds in a verification step — not a clever one-liner.
- Skills take this further: they turn a tested prompt into an installed tool that runs the same way every time.
"AI prompts for lawyers" is one of the most common things solicitors search for when they start with AI — and most of what comes back is a list of magic phrases promising to turn a chatbot into a legal assistant. The instinct is right; the product usually isn't. A list of 150 prompts you copy and paste is not a system, and for legal work it can be actively dangerous.
Why generic prompt lists fail solicitors
The problem isn't that the prompts are badly written. It's what they encourage. A prompt like "summarise the law on X" or "draft a clause that does Y" invites the model to produce confident, fluent output from its own general knowledge — exactly the conditions in which it invents authority, drifts into US law, or asserts "market standard" it cannot evidence. The prompt gets you an answer; it does nothing to make that answer safe to rely on.
That is the behaviour the court warned against in Ayinde v Haringey. A prompt list that produces unverified output and leaves the checking to you — under time pressure, with no structure — is a route to precisely the failure the profession is now alert to.
What a good legal prompt actually does
The prompts that work for legal practice share three features, and none of them is cleverness:
- They work from your documents. The model reviews the actual contract, drafts from your actual precedent, works from the real transcript — not from what such a document "usually" contains.
- They build in verification. A good prompt instructs the model to flag anything it cannot verify rather than asserting it, and to source what it does claim.
- They hold to your standard. Your house style, your positions, your format — consistently, not whatever the model defaults to that day.
The Lawyer's Prompt — book + 15 Claude skills
The tested legal prompts from the book, turned into installable Claude skills that run them for you — grounded in your documents, with the Ayinde verification discipline built in.
See what's included — £99 →From prompts to skills
This is the step beyond a prompt list. A skill takes a tested prompt and installs it into Claude as a tool — so instead of finding the right prompt, pasting it, and remembering to verify, you just ask, and the structured process runs every time. The Lawyer's Prompt is built on exactly this idea: it's not a prompt pack, it's the prompts turned into tools, with the discipline that makes them safe for legal work baked in. Read more in Claude for lawyers.
If you only take one thing from this
Don't judge AI prompts for lawyers by how clever they sound. Judge them by whether they work from real documents and build in verification. A prompt that skips those is a liability dressed up as a shortcut. See how to verify AI output for the method that makes any prompt safer.