You wouldn’t hand a new paralegal a matter and say "deal with it." You'd tell them who the client is, what you need, and how you want it written up.
Artificial Intelligence (AI) is no different. Give it three words and it gives you three words' worth of thinking back.
Many of you reading probably already use AI to summarize calls and meetings. But if you’ve only ever typed “summarize this call,” you’ve only seen a fraction of what it can do. You might get a tidy recap that reads well but misses the deadline you agreed on, the next steps, and the details a colleague would need when picking the file back up.
Getting a useful result starts with what goes in — and a few simple principles can help you get better at writing effective prompts for your AI.
This guide covers how to write better AI prompts for legal work, with practical examples built for lawyers’ workflows.
⚠️ One thing before we start. AI can get things wrong. Always read and check what it produces before you rely on it. A good prompt makes the output sharper, not bulletproof.
What is an AI prompt?
The words you type into an AI tool to get a result are known as a "prompt." You can think about prompts as a conversation starter.
How you begin that conversation — the questions you ask, the context you give, the instructions you set — has a big impact on the quality of what comes back. Ask something vague and you get something vague straight back.
What can lawyers do with AI prompts?
AI makes headlines for the big tasks like drafting briefs, digging up case law, and reviewing a contract in seconds. This is powerful, but it usually comes with a big price tag, and not every firm is ready to commit to Harvey or Legora just yet.
The good news is you don’t need to. Some of the most useful AI wins are in the unglamorous paperwork around the legal work that nobody brags about. The file note after a call, the summary of a client meeting, the follow-up email you’ll write ten of this week.
Five minutes here, ten there. It sounds like nothing until you add up every call and meeting across your firm. Picture handing your AI a conversation you just had and getting back the file note and the follow-up email you’d have written yourself, if only you’d had the time.
So, how do you write a prompt that gets you that?
How to write better prompts
This skill gets called "prompt engineering," but it's not as intimidating as it sounds. There's no code and no qualification required. It just requires you to be clear about what you want.
Think of it the way you'd brief a new paralegal — someone who knows the law but knows nothing about your firm, your matters, or how you like things written. The more precisely you set out the task, the better the work comes back.
A good prompt has five parts: Persona, Task, Context, Format, and Constraints. The easiest way to see how they work is to take one weak prompt and build it up.
Writing a prompt to summarize your phone calls
Let’s apply this to an AI that is able to summarize your phone calls.
Start with the prompt almost everyone reaches for first:
"Summarize this call."
It tells the AI nothing about what you care about, so it guesses. Here's how to fix that:
1. Persona — tell it who to be
Generative AI won't be winning an Oscar any time soon (give it a year). But ask it to play a role and it'll commit to the part — taking on the tone, the vocabulary, and the priorities of whoever you tell it to be.
So tell it what role to take.
"You're a paralegal at a family law firm."
This takes the AI from writing like a generic chatbot to writing like someone who works in legal.
2. Task — tell it exactly what to do
Be specific about the job.
"Summarize this client call. Pull out what the client asked for, what we agreed to do, and any deadlines mentioned."
Now it knows what to look for instead of guessing what your "summary" means.
3. Context — tell it what's going on (and why)
Share the situation and the reason behind that task. The AI takes the why and runs with it.
"The summary goes straight onto the client file, so write it like a file note: factual and specific, sticking to what was actually said."
4. Format — tell it how to lay the answer out
Tell it exactly how the answer should look — headings, bullet points, length, the lot. Left to its own devices it'll pick a shape, and this is unlikely to be the one you'd have chosen.
"Use these headings: Summary, Action items, Key dates. Keep it under 150 words."
5. Constraints — tell it what to avoid
It's just as useful to say what you don't want. A summary that faithfully captures the weather, the "how are the kids" chat, and three rounds of "no, you go ahead" isn't a file note, it's a transcript.
"Include action items, deadlines, and anything relevant to the matter. Leave out small talk, scheduling back-and-forth, and anything not relevant to the case."
Put it together
Stack those five parts and the generic "summarize this call" becomes a real instruction:
Customizing your AI summaries in VXT
VXT is legal AI that turns conversations into completed legal paperwork. During calls and meetings, it takes notes so you can focus on your client while keeping an accurate record of everything said. Afterwards, it summarizes and saves those notes, along with a time entry, to the right matter in your legal software, and drafts a follow-up email based on the conversation.
That summary is where your prompt comes in. You can stick with VXT’s default, or take everything you’ve learned above — your firm’s persona, your format, your constraints — and sharpen the output into something that fits exactly how you work. You can go further still, using call tags to apply different prompts to different call types, so an intake and a follow-up each get handled their own way.
You don’t strictly need VXT for any of this. With a secure, legal-grade AI tool, you could paste in a transcript, add a prompt like the one you built, and get a solid file note out the other end.
The catch is you’d do it every single time. Export the transcript, copy your prompt, paste it in, wait, then copy the result back into the right matter — for every conversation. VXT runs your prompt the moment a conversation ends, saving the note and a time entry to the correct matter and drafting a follow-up email. It’s the same output without any of the extra steps.
A prompt is a brief
You wouldn't hand a new paralegal a matter and say "deal with it." You'd brief them: here's the client, here's what we need, here's how we write it up.
A prompt is that same brief for AI. Write it once, and every conversation comes back the way your firm actually works, with the right details pulled out, in the right format, ready to drop into the file or turn into a follow-up email. The summary stops being something you fix and starts being something you file.
That’s the whole point of getting your prompts right. And VXT is where you can put those prompts to good use.
.jpg)