Structuring legal AI prompts with clarifying questions for better results

We have all heard that AI is going to reshape the legal profession. But after spending significant time integrating it into my day-to-day work, I have realized a fundamental truth: AI isn’t a search engine. It’s a junior associate. And just like working with a junior associate, the quality of the output is entirely a reflection of the quality of your instructions.

If you want partner-ready drafts, you need a structured framework. In this short video, I break down the four pillars I use every day to get precise, usable results:

⚖️Role: Anchor the perspective (e.g., “Act as senior commercial counsel”).
📝 Task: Use precise legal verbs (“Draft,” “Extract,” “Redline”).
📖 Context: Define the jurisdiction, the client's side, and your risk appetite.
🧠 Judgment: Demand structural constraints and verification safeguards.

The golden rule I always follow? Trust it to accelerate the draft, but distrust it enough to verify every word.

I would love to hear how other practitioners are prompting their workflows. What’s working for you, and what are you still refining? Let's discuss in the comments.

#LegalAI #LegalTech #ArtificialIntelligence #InHouseCounsel #Lawyers #GenerativeAI

There are dozens of AI prompt frameworks that lawyers are being told to use. But giving good instructions is already a core legal skill. We don't need another acronym. We just need to draw on those same principles:

1️⃣ Role

Tell the AI who it is and what expertise to draw on. Being specific is good, but like all things legal, there’s a balance with being too specific as the data might be too limited. In my example, I’ve referenced professional athletes, but within the broader category of high net worth individuals.

2️⃣ Purpose

Identify the client and state what they're trying to achieve. Try to provide as much context as possible. It can also help to refer to any notes you may already have.

3️⃣ Task

State the specific deliverable that we need to prepare. If the draft email is going to a client or a supervisor, state that. It’s helpful also to include any examples, checklists, risk matrices etc. that should be followed. Same as if you were giving a junior a precedent to use instead of making them start from scratch.

4️⃣ Key facts

Set out the info that isn’t publicly available or that the AI should focus on. In this case, we are asking them to consider 2 specific options instead of every contract extension variant.

5️⃣ Constraints

Here we build in guardrails. For example, the format, word count, tone, house styles, things to avoid etc. The more key constraints we can include, the more relevant the final output will be.

6️⃣ Iterations

This bit is something I like to do whenever I give instructions to people, and I’ve found it helpful when prompting AI. It’s best to think of prompting as an iterative process instead of something you need to get perfect in one go.

NB: This guide is tool-agnostic. Follow your organisation's rules for AI use.

What’s your prompting approach? Do you just give instructions like you would to a human, or do you have a system that works better?

----------

I'm building an e-library with the 110+ how-to guides I've written over 6 years. If you want early access when it launches, sign up to the free newsletter on my website / profile.

Every legal team hears that AI will change contract work. But we also hear the horror stories about relying on it too much. Where is the practical advice for your daily work?

Join our next session to get real, honest guidance.

What you will learn:
🛠️ Where AI works and where it fails: The truth about AI in drafting and review.
❗ Risk framework: How to easily spot mistakes in AI content.
⭐ Smart prompting: Techniques to get reliable results every time.

We will show you these strategies live in LawVu Draft using real contract examples.

Link in comments

How I DON'T use AI as an in-house lawyer

I use AI. A LOT. But there are plenty of times when I don’t bother. It’s not a philosophical stance. Sometimes AI is simply not the fastest, safest, or most useful tool for the job.

A few examples:

1. When I already know the answer.

If the question is straightforward and I can answer it in 30 seconds, I do not need to spend 5 minutes writing a prompt, checking the output, and cleaning it up.

Sometimes the fastest tool is still your own brain.

2. When the facts are too thin.

AI is great at organizing information.

It’s great at understanding information.

It’s great at summarizing information.

It is not great at magically knowing the missing facts.

The number one rule about AI prompts is CONTEXT. If someone sends me “Can we do this?” with no context, I usually need to ask better questions first. Otherwise AI is just helping me analyze a mystery box.

Bonus: You CAN ask AI to draft follow up questions that help you fill in the missing facts.

3. When the issue is mostly political, not legal.

A lot of in-house work is less about the contract or the risks and more about:

Who owns this relationship?

Who promised what?

Who is already annoyed?

Who needs to be brought along before we respond?

The point? AI is not great at reading the room.

4. When I need company history.

Sometimes the answer depends on something that happened six months ago in a meeting, a prior exception, a sensitive customer issue, or a business preference that is not written anywhere useful.

AI does not know that unless I tell it.

And if I have to explain the entire backstory, sometimes I might as well just do the analysis myself.

5. When the work needs my voice.

Ok, yes, I use AI to draft my work emails. I tend to be a warm, casual person and AI is very helpful in assisting me meet the appropriate business tone.

But my LinkedIn posts? I always take the first pass.

Creative writing (I dabble 🙂)? I don’t use AI for anything except as an easier to use thesaurus. Not even editing. Not even outlines. It’s too easy to lose my voice which is death in creative writing.

For me, AI is most useful when the work is messy, repetitive, long, scattered, or mentally draining. It is less useful when the work is quick, context-heavy, relationship-sensitive, already clear, or deeply personal.

#LegalAI #InHouseCounsel #AIOverreliance

In-house lawyers, a quick, practical AI tip - this is something I do for basically all my prompts and have saved it as a shortcut so that I can apply it on autopilot.

At the end of your prompt, add:

Ask me 3 clarifying questions before you generate the output.

It adds a little bit more effort to each workflow, but I find it improves the output about 80% of the time by asking me to add context I forgot or raises a perspective I hadn't thought of.

Usually only 1-2 of the 3 clarifying questions are helpful, but are often very helpful. Plus most LLMs will skip straight to the output where you leave questions unanswered. Occasionally the LLM will get confused and add clarifying questions into the output as opposed to the chat box, which is unhelpful, but rare.

I have not added it to any system prompts though because sometimes I use a simple, zero-shot prompt where risk is low and speed is the priority (like asking for a table of countries by Eurovision wins - no clarification necessary, I want to know it now, and it's low stakes if the table is wrong).

I have played around with skewing the clarifying questions to make them more relevant to me or the legal profession, but I've found it's cleaner and more effective just to let the LLM clarify what it's most confused about.

Overall, my experience is it adds to effectiveness while having neutral or positive impact on risk.

Always keen to hear how lawyers are using AI and catch up for coffees if you are stuck for practical uses. Let me know if you have a standard add on to your prompts that you use frequently? Do you have a suggestion to improve my prompt add on?

[I do not use AI to write any of my posts.]