The Digital Transformation Playbook

How to Make ChatGPT Your Legal Assistant Without Going to Jail

Kieran Gilmurray

The legal profession stands at the threshold of transformation as generative AI reshapes how lawyers work. But how do you harness this technology effectively? The secret lies in mastering prompt engineering—the art and science of crafting instructions that guide AI to produce valuable legal outputs.

TLDR:

  • Generative AI represents an unprecedented shift in how lawyers work, with Microsoft's legal team reporting 32% faster task completion and 20% better accuracy
  • Legal applications span contract review, dispute resolution, litigation support, M&A due diligence, regulatory compliance, and client communications
  • Start experimenting with these techniques on your own work to discover practical applications and efficiency gains


Microsoft and the Singapore Academy of Law just released a 30+ page guide on using GenAI in legal practice.

 Our AI generated conversation dives deep into the recently published guide "Prompt Engineering for Lawyers," extracting actionable insights for legal professionals eager to leverage AI without getting bogged down in technical complexity. We explore the four fundamental pillars of effective prompting: crafting clear goals, providing rich context, setting explicit expectations, and referencing specific sources.

The potential benefits are substantial. Microsoft's legal team reported tasks completed 32% faster with 20% better accuracy when using AI tools properly. From contract review and litigation support to regulatory compliance and client communications, we walk through real-world examples of prompts tailored for different legal scenarios. Each example demonstrates how thoughtfully constructed instructions can yield powerful results across practice areas.

Yet we don't shy away from critical ethical considerations. The discussion emphasizes that AI serves as a co-pilot, not autopilot—the lawyer always bears ultimate responsibility for the work product. We address essential questions about client confidentiality, data security, and the importance of verification when working with AI-generated content.

Whether you're already experimenting with AI tools or just beginning to explore their potential, this episode provides a practical foundation for developing this increasingly essential legal skill. Start applying these prompt engineering techniques today and discover how they can elevate your practice while freeing you to focus on the strategic work that truly requires your expertise.

Read my top ten prompts for legal professionals - here

Support the show


𝗖𝗼𝗻𝘁𝗮𝗰𝘁 my team and I to get business results, not excuses.

☎️ https://calendly.com/kierangilmurray/results-not-excuses
✉️ kieran@gilmurray.co.uk
🌍 www.KieranGilmurray.com
📘 Kieran Gilmurray | LinkedIn
🦉 X / Twitter: https://twitter.com/KieranGilmurray
📽 YouTube: https://www.youtube.com/@KieranGilmurray

Speaker 1:

Welcome to the Deep Dive. Today we're diving into something really transforming the legal world generative AI and specifically how lawyers can get good at well, prompt engineering. We know you want to get up to speed on this quickly and maybe prompt. Engineering sounds a bit technical, maybe a bit in the weeds. It can sound that way, but think of this deep dive as your shortcut, your express lane to really grasping what it's all about.

Speaker 2:

Yeah, absolutely. We're looking closely at this guide Prompt Engineering for Lawyers. It's designed specifically to help legal pros like you really leverage these AI tools effectively.

Speaker 1:

So our mission here is what To pull out? The key stuff.

Speaker 2:

That's it. We're extracting the most important insights, showing you how getting those prompts right is well fundamental to making this technology work for you.

Speaker 1:

Okay, let's jump in. The guide kicks off by talking about how AI in law has, you know, evolved.

Speaker 2:

Right.

Speaker 1:

It mentions the early days data analytics, e-discovery, due diligence, support useful stuff sure. Standard applications yeah, but then it says generative AI. Is this unprecedented opportunity, a real shift?

Speaker 2:

And that shift is key. It's about moving lawyers away from some of the, let's say, more routine work.

Speaker 1:

So you can focus on higher value things.

Speaker 2:

Precisely. The guide highlights that potential for you to spend more energy on strategic thinking, client relationships the things that really need your specific expertise, your human touch.

Speaker 1:

It mentioned some stats too right, Something about Microsoft's lawyers.

Speaker 2:

Oh yeah, quite striking. Their legal team, using these tools, apparently got tasks done 32% faster.

Speaker 1:

Wow, 32%.

Speaker 2:

And get this 20% better accuracy, so faster and more accurate.

Speaker 1:

That's significant. It really shows how getting the prompts right can unlock serious efficiency.

Speaker 2:

Huge efficiency gains yeah.

Speaker 1:

Yeah.

Speaker 2:

It really underlines the potential here. And the guy notes this tech isn't just theoretical, it's being baked into legal tech tools more and more.

Speaker 1:

Like for research and drafting.

Speaker 2:

Exactly Research contract drafting. They even give examples like the Singapore courts using it to help litigants, or the Singapore Academy of Law beefing up its research databases.

Speaker 1:

So it's becoming quite embedded.

Speaker 2:

And it's important to know. As the guide points out, there's difference between AI tools built specifically for law.

Speaker 1:

And the general ones everyone's hearing about, like ChatGPT and others.

Speaker 2:

Right. The legal specific ones might have deeper training on case law statutes. That sort of thing can give them an edge.

Speaker 1:

But the general ones are still useful.

Speaker 2:

Oh, definitely Very useful. You just need to understand the strengths of each, and the guide lists a ton of ways lawyers can use this stuff.

Speaker 1:

OK, Like what Well?

Speaker 2:

the obvious ones drafting documents, research, analysis, summarization, but also things like negotiation support.

Speaker 1:

Negotiation support.

Speaker 2:

Maybe analyzing opposing arguments, suggesting counterpoints based on provided case law, that kind of thing. Okay, Also, knowledge management within a firm client communications handling, meeting logistics.

Speaker 1:

Wow Okay, it goes beyond just document work.

Speaker 2:

Way beyond Practice management billing even marketing. It's spreading his wings.

Speaker 1:

you could say Right, but the guide stresses something crucial, doesn't it? About the input.

Speaker 2:

Absolutely critical point, All this fancy AI. Its effectiveness really, really hinges on what you feed it.

Speaker 1:

The prompts so bad prompt, bad result.

Speaker 2:

Essentially yes, or at least not the result you wanted. The guide makes it clear Even the best AI needs clear, specific, well thought out instructions.

Speaker 1:

Which brings us back to prompt engineering. So can you break that down? What is it?

Speaker 2:

simply, In essence, it's the skill, maybe the craft, of writing those instructions, the prompts to get the AI to do exactly what you need it to do.

Speaker 1:

And it's important because the AI is sensitive to how you ask.

Speaker 2:

Extremely sensitive. Tiny changes in wording can sometimes lead to very different outputs. That's why mastering prompts is so vital.

Speaker 1:

The guide calls it more of an art than a science. I found that interesting.

Speaker 2:

Me too. It suggests that while there are principles, techniques you can learn, there's also an element of sort of intuition experimentation, getting a feel for what works with a particular AI for a specific task.

Speaker 1:

Okay, so let's make this practical. How do you actually build a good prompt? The guide suggests four key things, right framed as questions yes, a really helpful structure.

Speaker 2:

basically, what do you want, why and who's involved? How should it answer and what? What should it read first?

Speaker 1:

Okay.

Speaker 2:

Which leads directly to the four pillars Goal context, expectations and source.

Speaker 1:

Right, let's take goal first. What's the key here? Clarity.

Speaker 2:

Absolute clarity. Start with clear instructions. Put the main task right at the beginning. Use strong verbs. Draft this. Summarize that. Analy, analyze this E-direct Exactly, and if it's a big, complicated goal, break it down. Smaller steps are usually better.

Speaker 1:

Makes sense. Keep it manageable. What about context? Giving the AI the background?

Speaker 2:

Crucial Context is well, everything around the task Background info, any limitations or constraints, who the parties are.

Speaker 1:

And you mentioned persona.

Speaker 2:

Yeah, Defining who you are in this scenario, the lawyer asking, and who the audience for the AI's output is. Is it for another lawyer, a client, a judge? That affects the tone and detail. Okay, the guy also mentioned something called few shot learning, few shot. What's that?

Speaker 1:

It's like giving the AI a couple of examples of what you want the final output to look like. Ah, like showing it a sample?

Speaker 2:

Exactly. It helps the AI grasp the style, the format you're after, without needing tons of explicit instructions, Sort of like learning by example. For legal work, context means things like the legal issue, the jurisdiction, super important relevant laws, case details, stakeholders, the audience.

Speaker 1:

So paint the full picture, got it. Then expectations. This is about the output format.

Speaker 2:

Format, yes, but also depth, level of detail. The tone should it be formal, friendly, firm, persuasive.

Speaker 1:

Like instructing it how to sound.

Speaker 2:

Precisely, and the format bullet points, a table, a formal memo, a contract clause. Specify the jurisdiction like. Analyze this based on Singapore law.

Speaker 1:

And there was that phrase think step by step.

Speaker 2:

Right chain of thought prompting. It's a technique especially for complex tasks, where you ask the AI to sort of narrate its reasoning process.

Speaker 1:

Like showing its work in math class.

Speaker 2:

Kind of it often leads to more accurate, more reasoned outputs, particularly for tricky legal analysis. You're guiding its thinking process.

Speaker 1:

Fascinating.

Speaker 2:

Utterly critical. The guide really pushes this. Always try to provide reference material, especially with general AI tools, internal documents, specific web pages, case files.

Speaker 1:

And be specific about what in the source.

Speaker 2:

Yes, Point to specific paragraphs, specific clauses. If you can, don't just dump a whole hundred page document if you only care about section five.

Speaker 1:

More precision, better results.

Speaker 2:

Exactly. Many tools let you upload docs or paste URLs now, which helps, and if you're asking it to synthesize from multiple sources, ask it to cite its sources in the output.

Speaker 1:

So you can check its work essential.

Speaker 2:

Absolutely. Verification is key.

Speaker 1:

And there's a huge caveat here about privacy Massive.

Speaker 2:

Cannot stress this enough Always, always, follow your firm's or organization's data privacy and security policies.

Speaker 1:

Because you're sending information to the AI.

Speaker 2:

Right Understand the terms of the AI service you're using. Free public tools might use your prompts, including any sensitive info in them, for their own training.

Speaker 1:

Oh wow, Didn't realize that.

Speaker 2:

Enterprise solutions usually have much stronger safeguards but you need to know the terms and anonymize your prompts whenever you possibly can protect that client confidentiality.

Speaker 1:

Okay, super important reminder. Now the guide gets really practical with examples, which is great. It tries to show this isn't just theory.

Speaker 2:

Yeah, they frame it as inspiring you to experiment, which I like. It shows how applying those four elements goal, context, expectations, source works in real legal scenarios.

Speaker 1:

This is the shortcut part. Seeing it in action.

Speaker 2:

Exactly, and they have that quick tip just paste the relevant text or attach the file, if the tool allows. Make it easy for the AI.

Speaker 1:

So let's look at some areas. Contract review and drafting bread and butter for many lawyers, right.

Speaker 2:

Examples like detail the ESOP equity benefits for senior managers over five years, citing specific sections from the attached ESOP plan.

Speaker 1:

Very specific goal and source.

Speaker 2:

Or graft an IP indemnity clause favoring the licensor. Use a formal but concise tone. Base it on these two sample clauses See Goal context favoring licensor expectations tone source samples.

Speaker 1:

You can see how it built.

Speaker 2:

They even have one for creating a playbook for reviewing commercial agreements. Telling the AI to spot legal versus commercial issues. Suggests positions based on past agreements and comments.

Speaker 1:

That's quite sophisticated. It's asking for reasoning, not just retrieval.

Speaker 2:

That's the key insight. Yes, you guide its reasoning within defined boundaries.

Speaker 1:

Okay, what about dispute resolution? Identifying inconsistencies.

Speaker 2:

Yeah, critical task. Yeah, prompts like compare the plaintiff's affidavit attached with their statement of claim attached. Identify inconsistencies in a table referencing the specific sections.

Speaker 1:

Saves hours of manual comparison, potentially.

Speaker 2:

Huge time saver. Or analyzing a witness's affidavit versus their trial transcript. Again, put discrepancies in a table. Assess the impact on credibility. Imagine getting a first pass on that automatically.

Speaker 1:

Very powerful, then litigation support.

Speaker 2:

Things like generating discovery lists. Generate a list of documents we need to provide. We represent the plaintiff. Include relevant client emails from Jan March from this archive Thank you. Attached emails involving this former employee Table format date recipient summary max 20 words. Note large attachments link Streamlining investigation workflows.

Speaker 1:

What about M&A due diligence?

Speaker 2:

Absolutely Analyze this target company contract for red flags. For us, the acquirer Focus on liabilities, indemnities, ip Use a table issue clause, summary reference implications, remediation Use only the attached contract.

Speaker 1:

Focus analysis.

Speaker 2:

Or analyzing third-party IP licenses in a merger, identify potential impacts if these licenses attached terminate due to the merger. Analyze key terms like scope, fees, revocability in a table.

Speaker 1:

Okay, regulatory compliance.

Speaker 2:

Vital area. Examples like list the applicable Singapore rules of court for personal service in the high court. Use formal legal language. Cite official court statute websites.

Speaker 1:

Getting precise legal info.

Speaker 2:

Or analyze these Singapore data protection references provided? Can our spa clients stop collecting NRIC numbers? Give an objective assessment with justification, citing specific PDPA sections and guidelines. Think step by step.

Speaker 1:

Using that chain of thought prompting again.

Speaker 2:

Exactly or generating a checklist for retrenchment obligations in Singapore. Create a checklist for HR finance legal covering employer duties during retrenchment limit to Singapore Employment Act and official guidelines.

Speaker 1:

Very practical. And finally, client communication and billing.

Speaker 2:

Essential stuff. Yeah, summarize these meeting minutes attached Draft a plain language follow-up client email covering attendees, topics, decisions and a detailed action plan with deadlines. Base it only on the minutes you can shift all of it. Or drafting billing narratives. Compose work descriptions for client billing based on these emails. Specify date range. Client attachments Use a table. Exclude internal chatter. Follow firm standard language See attached file.

Speaker 1:

Helps maintain consistency and accuracy in billing.

Speaker 2:

There's even one for creating a consolidated bill narrative Comparing against estimates, attributing work, really detailed instructions to get a polished output.

Speaker 1:

And the guide doesn't stop there, right, it has specific co-pilot examples.

Speaker 2:

Yes, which is super helpful because it ties it to tools Many people use Daily Word, Teams, Outlook, et cetera.

Speaker 1:

Like using Word to pull IP clauses from files.

Speaker 2:

Exactly, or using Teams to recap a meeting with pros and cons, or PowerPoint to draft presentation slides based on a document.

Speaker 1:

Right.

Speaker 2:

Or Edge to summarize a web page.

Speaker 1:

So integrating it right into the workflow.

Speaker 2:

That's the direction things are moving, making AI assistance seamless.

Speaker 1:

Okay, this all sounds amazing, potentially game changing, but the guide rightly brings up ethics and good practices.

Speaker 2:

Crucial conversation, non-negotiable for lawyers.

Speaker 1:

What's the main message there? Responsibility.

Speaker 2:

Ultimate responsibility. The lawyer is always responsible for the final work product. Ai is a tool, not a replacement.

Speaker 1:

It's not perfect. The AI.

Speaker 2:

Far from it. The guide hammers this home. Verify everything. That old principle don't rely on an authority you haven't read applies doubly here. Review, check, validate.

Speaker 1:

They use that phrase co-pilot, not autopilot.

Speaker 2:

Great analogy. It assists, it augments, but you are flying the plane. It doesn't replace your legal judgment, your advocacy skills, your drafting expertise. It helps with fluency, exploring options maybe, but not core subject matter knowledge.

Speaker 1:

And disclosure. Do you need to tell clients or courts you used AI?

Speaker 2:

It depends Firm policy, professional codes, court rules. Sometimes disclosure might be required. You need to know the rules that apply to you.

Speaker 1:

And confidentiality. Again we touched on this.

Speaker 2:

Can't say it enough Know the terms of service of your AI tool? Are your prompts being used for training? Could client data be exposed?

Speaker 1:

Free tools might be riskier.

Speaker 2:

Potentially. Yes, Enterprise solutions generally offer better privacy, but you must understand the specific terms and anonymize prompts whenever possible.

Speaker 1:

To help remember this they have a DO and don't list.

Speaker 2:

Yeah, very practical. Do use it for comparisons, summaries, identifying issues, brainstorming using trusted sources. Do no start a new chat for each task. For clarity, do rerun related prompts sequentially, maybe even rerun to check consistency. Do experiment and iterate.

Speaker 1:

And the DON'Ts.

Speaker 2:

Don't ask it to do too much at once in one prompt. Don't expect perfection first time. Don't assume the output is accurate much at once in one prompt. Don't expect perfection first time, don't assume the output is accurate or fit for purpose without checking and definitely don't just copy-paste the output as your final work without careful review and your own judgment.

Speaker 1:

Sound advice. The guide also includes a glossary.

Speaker 2:

Yes, defining key terms like AI context, generative AI model output, prompt training data token.

Speaker 1:

Tokens. What are those exactly?

Speaker 2:

Think of them as pieces of words like engineering might be split into engine, aero, engine. It's how the AI processes text.

Speaker 1:

And why does that matter?

Speaker 2:

Because models have limits on how many tokens they can handle in a prompt and its response. It affects how much text you can input or how long an answer you can get. Good to be aware of.

Speaker 1:

Useful glossary then and they acknowledge contributors yes, nice touch Shows it was a collaborative effort to build this resource. So wrapping this up, the big takeaway feels like prompt engineering isn't just a nice to have. It's becoming a core skill for lawyers using AI.

Speaker 2:

Absolutely essential. It's not just asking questions, it's strategically guiding the AI to get useful, accurate results. That's how you unlock the efficiency gains and free yourself up for that higher level strategic work.

Speaker 1:

So hopefully after this deep dive you feel you've got a solid handle on the basics, the principles and how this applies practically in law.

Speaker 2:

You should have a strong foundation now, yeah.

Speaker 1:

And it does make you think where does this go next? As the AI gets smarter, as prompt engineering becomes more sophisticated, how will that continue to change legal practice?

Speaker 2:

What new possibilities open up.

Speaker 1:

Exactly what haven't we even thought of yet?

Speaker 2:

It's definitely something for everyone in the legal field to keep mulling over. The best advice now is probably start experimenting. Explore these techniques, see how you could apply them to your own work, your own documents.

Speaker 1:

Great final thought. Thanks for joining us for this deep dive. We hope it's given you a quick and valuable understanding of this really important area.

Speaker 2:

Hope it was helpful.

Speaker 1:

Until next time.

People on this episode