If your firm has rolled out ChatGPT Enterprise with Outlook integration, you're sitting on an underutilized capability: the ability to train GPT to draft emails in your voice and style - emails that actually sound like you.

I've been experimenting with this workflow, and while it takes some upfront effort to set up, the payoff is significant. Here's how to do it and why it's worth your time.

Why This Matters

We've all seen AI-generated emails that read like they were written by a particularly formal robot. They're technically correct but somehow sound off, too formal, or too generic. In other words, the emails are missing the small touches that make your communication recognizably yours.

The problem is that when you ask GPT to "draft a professional email" or "draft an email written by a biglaw associate addressed to his/her colleagues," it defaults to a one-size-fits-all approach. It doesn't know how you tend to open client emails or whether you prefer direct language or more casual speech.

But if GPT can analyze a corpus of your actual sent emails, it can learn those patterns: your greeting style, your sign-off preferences, whether you use bullet points or prose, how you hedge requests versus making direct asks, and more. The result is draft emails that need minimal editing because they already sound like you.

Setting Up Your Email Style Project

Using your firm's ChatGPT Enterprise account (you might achieve something similar with other AI tools, but this post specifically addresses ChatGPT integration), here's how to create a dedicated project for email drafting.

Step 1: Create the Project

In ChatGPT, create a new project and name it something like "Email Drafts" or "My Email Style." This project will serve as the container for your style training and all future email drafting requests.

Step 2: Export or Grant Access to Your Sent Emails

Depending on your firm's setup, you'll either export a selection of your sent emails or grant the project direct access to your Outlook sent folder. Aim for somewhere between 200 and 500 sent emails. This gives GPT enough data to identify meaningful patterns without overwhelming the analysis.

Step 3: Generate Your Style Profile

Once your emails are accessible, run this initial prompt:

Analyze my sent emails and produce a concise, actionable writing style guide. Identify patterns in:

- Tone (formal vs. conversational, warm vs. direct)
- Greeting and sign-off styles
- Sentence length and paragraph structure
- Use of bullet points vs. prose
- How I phrase requests (hedging language vs. direct asks)
- Any recurring phrases or expressions I tend to use
- How my style varies by recipient type (if discernible)

Produce this as a reusable style guide that can inform future email drafts.

GPT will analyze the emails and produce a style profile. Save this within your project.

Step 4: Set the Project Instructions

Once you have your style profile, add it to your project's custom instructions along with a directive like:

When drafting emails, always follow the writing style guide saved in this project unless I explicitly request a different approach.

Now, every email you draft within this project will automatically apply your personal style.

Taking It Further: Specialized Projects for Different Email Types

Here's where this gets really useful. You don't write the same way to everyone, and your AI assistant shouldn't either.

Consider creating separate projects for different communication contexts:

Internal Emails: Train this project on emails to colleagues and staff. These are probably more casual, use more shorthand, and get to the point faster.

Client Emails: Train this project specifically on client correspondence. These likely have a different tone, perhaps warmer openings, more thorough explanations, and different sign-off conventions.

Court Emails: If you correspond with court clerks and chambers, train a project on those emails. The tone here is typically more formal and precise.

Opposing Counsel Emails: Train this on your correspondence with opposing counsel. These might be more measured, more carefully worded, with different conventions around confirmations and follow-ups.

For each specialized project, select only the relevant emails for training. When you create your "Client Emails" project, export only emails sent to clients. This ensures the AI learns the specific voice you use in that context rather than blending all your communication styles together.

The Iterative Reality

This doesn't work perfectly on the first try.

After generating my initial style profile, I noticed the AI was using phrasing I don't actually use. It might have picked up patterns that were technically present in my sent emails but weren't representative of how I actually want to sound.

This is where iteration matters. When you see something off in a draft, don't just fix it and move on. Update your project instructions to explicitly address it.

For example, we all know AI loves em-dashes (though, to be fair, so do many lawyers). If you're sensitive to that, add an instruction like:

Use em-dashes sparingly. Default to other forms of punctuation instead of em-dashes when structuring sentences.

Or if you notice GPT is being too formal:

Avoid overly formal phrases like "I hope this message finds you well" or "Please do not hesitate to reach out." Use warmer, more natural openings.

Don't be discouraged by the iteration process. It takes time, but it goes a long way towards developing your long-term skill in employing AI.

What You'll Notice

After working with this setup for a few weeks, here's what you'll find:

First, your email drafts will require less editing. Instead of rewriting AI-generated emails from scratch, you'll only need to make minor tweaks because the AI is building the bones already.

Second, consistency improves. When you're tired or rushed, your writing quality can vary. The AI maintains a consistent baseline that reflects your best work, not your 6 PM Friday work.

Third, you will catch things that you might otherwise had missed. For example, sometimes the AI's draft remind me to include something I would have forgotten, like a relevant attachment reference or a follow-up date.

A Practical Starting Prompt

If you want to try this today, here's a prompt to test in your next email drafting session (even without the full project setup):

I need to draft an email to [recipient type: client/opposing counsel/colleague]. 

Context: [Brief description of situation]

The email should: [What you need to communicate]

Before drafting, please ask me any clarifying questions about tone, context, or specific points to address.

Why This Is Worth Your Time

If your firm offers ChatGPT Enterprise with Outlook integration, building a personalized email drafting project is a practical and relatively low-stakes application of AI available right now. It simply saves you time on a task you do dozens of times per day while maintaining the personal touch that matters.

Give it a try and let me know how it goes. I'd love to hear what works (and what doesn't) as you build out your own email drafting projects.