How to Set Up ChatGPT, Claude & Gemini for Legal Work

There's a 5-minute setup for all of the major AI platforms that separates lawyers who are getting mediocre AI output from lawyers who are consistently getting partner-level work product. And

based on my consulting experience working with legal teams, most lawyers don't even know that this exists. And

right now, your [snorts] AI is writing like a first-year associate who's terrified of giving a direct answer. And

the reason is because that's exactly what AI thinks you want. So today, I'm going to show you how you can quickly configure those settings regardless of which platform you're using so that you can improve the quality of output. I've

configured custom instructions across ChatGPT Claude Gemini Perplexity Copilot for dozens of lawyers at this point in time. Same AI, same prompts, but completely different results. And

the only variable, just a few minutes of setup. Today, in this video, I'm going

setup. Today, in this video, I'm going to show you the exact setup that I use and that I advise. Not theory, the actual instructions. And then later in

actual instructions. And then later in the video, I'll show you how to get a free template that generates custom instructions for your specific practice area litigation corporate immigration, whatever you do, so you can

just copy and paste when you're done.

Here's the thing, every time you open ChatGPT or Claude or Gemini, it doesn't matter. It has no idea who you are. It

matter. It has no idea who you are. It

doesn't know that you're a lawyer. It

doesn't know your jurisdiction. It

doesn't know that your managing partner hates emails longer than one paragraph.

So what does it do? It gives you the safest, most generic response possible.

It hedges everything. It says, "Consult a qualified professional." You are the qualified professional. And the worst

qualified professional. And the worst part for lawyers, well, certainly the most frustrating, is that you're probably re-explaining the same context every single session. "I'm a corporate attorney in New York. I need plain language. Don't give me disclaimers."

language. Don't give me disclaimers."

Over and over. So that stops today. I'm

going to show you how you can stop that once and for all.

Every major AI platform has a feature called custom instructions or some version of it. It's a place where you tell the AI who you are, how you work, and what you expect. And it remembers that across every single conversation.

I guess think of it like onboarding a new associate. You do it once, and from

new associate. You do it once, and from that point on, hopefully they know how you like things done. With AI, they definitely know how you like things done. And every platform calls it

done. And every platform calls it something slightly different, but the concept is identical.

There are at least four things that every lawyer should include in their global instructions. But when we're

global instructions. But when we're talking about global instructions, we don't want to flood it with lots of information, just the most important information that we want the AI to know about us, our company, what we do,

across all conversations.

And number one is your role. And I'd

suggest keep it to just two or three lines at most. So include your practice area, your seniority level, your jurisdiction. For example, I'm a

jurisdiction. For example, I'm a litigation partner at a mid-size firm in Texas. I handle commercial disputes and

Texas. I handle commercial disputes and employment matters.

That's it. The AI now frames every answer through that lens.

Number two, here's where you input your preferred communication style. So do you want concise bullets or detailed memos?

Formal or conversational? Be direct.

Lead with conclusion. Use plain

language. No legalese unless I ask for it. I would say that this alone

it. I would say that this alone eliminates probably half the amount of generic output that you're getting.

Number three, this is what I would call behavioral rules. And this is a really

behavioral rules. And this is a really powerful and useful addition. So for

example, never tell me to consult a lawyer, I am the lawyer. Cite specific

statutes and rule numbers, not just statute names. Flag

statute names. Flag jurisdiction-specific differences when relevant. These rules shape how the AI

relevant. These rules shape how the AI thinks, not just how it writes.

Number four, this is where we tell the AI what not to do. Now, if you've ever used AI platforms before, you know that it can be incredibly agreeable. And

after about the third or fourth time of being told that every single one of your ideas is great, it gets a little bit tiresome. And you actually want direct

tiresome. And you actually want direct and honest answers. And so one way of doing that is telling it what not to do.

And so things such as don't open with great question. Don't hedge it with depends

question. Don't hedge it with depends unless you explain the variables. Don't

give disclaimers about this being legal advice. You'd be surprised how much this

advice. You'd be surprised how much this cleans up the output. Okay, let me show you live in a few of the platforms exactly where you go to enter the custom instructions. So right now we're here in

instructions. So right now we're here in Claude. Just come down to your name. The

Claude. Just come down to your name. The

it's the same if you're in the desktop app. Come up to settings, then general,

app. Come up to settings, then general, and then it's right here. What personal

preferences your Claude consider in its responses. Now, I'm going to paste in my

responses. Now, I'm going to paste in my template which has role, communication style, behavioral rules, what not to do.

And I've got that pre-prepared. So I'm

going to put that in there. Pause it,

have a read if you like. I'm going to save that. Now, what I'm going to do is

save that. Now, what I'm going to do is I'm going to ask it the same question that I asked at the start of the video in ChatGPT.

Sure, it's not the exact same platform, but the point being, the custom instructions will improve the output regardless of the platform that you're on. So let's ask the same question.

on. So let's ask the same question.

Draft a demand letter for breach of a commercial lease agreement where the tenant has failed to pay rent for 3 months. The landlord wants to recover

months. The landlord wants to recover unpaid rent and terminate the lease.

We're going to use Sonnet 4.6.

4.6 is great for this type of thing. 4.

Opus 4.6 is even better, but more expensive. You probably want to preserve

expensive. You probably want to preserve using Opus 4.6, not just because it costs more, but it's also slower. And

Sonnet is perfectly fine for something like this. If you're doing a task where

like this. If you're doing a task where complex reasoning is required, like a detailed contract review and redline, then Opus 4.6 is going to be a better model for that.

Okay, so I just let that run, and we've now got a response. So let's take a look. So what have we got? We've got a

look. So what have we got? We've got a direct answer. Um specific statute is

direct answer. Um specific statute is cited. No hedging, no disclaimers. Let

cited. No hedging, no disclaimers. Let

me check. No, no disclaimers. So same

question, completely different output.

Now, I'll just quickly show you how you can configure the uh custom global instructions in all the other platforms, all the major platforms. So in ChatGPT, go down to settings, go down to personalization, and then you've got

your custom instructions there. In

Gemini, go down to your settings, go to personal context, and then your custom instructions for Gemini, right? So you

can just add them there.

And then finally, I will show you Perplexity. Come down to your account

Perplexity. Come down to your account and settings, go to personalization, and you've got your custom instructions there. Okay, so that's global custom

there. Okay, so that's global custom instructions. Now, let's talk about

instructions. Now, let's talk about project-level instructions. Every major

project-level instructions. Every major AI platform has a second layer, and that's project-level instructions. In

Claude, it's called projects. In

ChatGPT, it's also called projects.

Gemini has Gems. Perplexity has Spaces.

And the way to think about using projects, and what I advise, is this is where you create a dedicated workspace for a specific client or matter. You

upload the relevant documents, set instructions specific to that engagement, and every conversation in that project has that full context. Now,

one thing to keep in mind, when you start a new session inside of a project, it won't always uh scan all of the documents in that specific project. So

it's always safe to prompt it and say, "Explore the documents in this project," and then give it your prompt. It's just

a sure-fire way to make sure that you're getting the AI is getting all the context it needs. It may, based on your prompt, scan the documents that it thinks it needs to scan, but it may also miss

some. And just to really hammer home

some. And just to really hammer home this point, think of it this way. Global

instructions tell the AI who you are.

Project instructions tell the AI what you're working on. Together, I would say that they eliminate 90% of the re-explaining that you might be doing today if you're not leveraging global

instructions or custom instructions.

Now, a question I get at this point from lawyers commonly is, "Is it safe to put my practice details into these tools?"

Now, that's a fair question, but here's the thing. Custom instructions don't

the thing. Custom instructions don't contain client data. You're telling the AI your role, your preferences, and your rules, right? That's no different from,

rules, right? That's no different from, I guess, your LinkedIn profile.

For actual client work, use the paid tiers. Absolutely use the paid tiers

tiers. Absolutely use the paid tiers that are dedicated to business customers. ChatGPT team, Claude Pro on

customers. ChatGPT team, Claude Pro on either business or enterprise. They

don't train on your data. And I always advise, just to be safe, never put real client information or confidential details into any tool that would be

detrimental to your team or your firm if those details were to get out. Just use

anonymized scenarios.

So let's recap. In about 5 minutes, you can go into your preferred AI platform that you use at your business, and you can, one, set up global instructions that make every AI conversation start with the full context on who you are and

how you work. Number two, create project-level workspaces for specific clients or matters. No more re-uploading the same documents. And number three, do this across ChatGPT, Claude, Gemini,

Perplexity, and Copilot using the same four-element framework.

And this is just the foundation. In one

of my next videos, I'm going to show you how to make your AI actually learn from every single session so that it gets smarter the more you use it without you doing anything extra. That's the

compounding effect, and it really does change everything when lawyers learn how to use these platforms in such a way.

And finally, I've put together the lawyers' AI setup template. It's free.

It's linked below. Just pick your practice area, fill in a few details about your firm, and you'll have copy-paste custom instructions for every major AI platform in under just a couple of minutes. It's located at

of minutes. It's located at legalairoi.com, and there's a whole bunch of free tools there that we've created focused on how legal teams can get value from AI tools

and how they can measure the ROI of different AI platforms. If this was useful, I would be grateful if you subscribed. I put out a weekly video on

subscribed. I put out a weekly video on AI for legal teams. Drop a comment if you've got something to add to the conversation, and I will see you in the next one.