Making sure SayMail isn't just ChatGPT for emails

👟 Topic: Making sure SayMail isn't just ChatGPT for emails

👾 Get my AI Coding file: I’m documenting everything I learn about AI coding so you don’t have too. Get my personal notes below

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Download the Free Report

I've been wrestling with this question lately:

Why would anyone use SayMail when they could just open ChatGPT and type "write me a cold email"?

It's a fair question. And honestly, if I can't answer it clearly, then I have no business building this thing.

So far, here’s how I’m trying to figure this out.

Remember a few weeks ago when I mentioned my original idea for SayMail? It was simple: paste someone's LinkedIn URL, hit a button, and get a cold email.

Apparently someone else had already tried this.

A tool called IntroFlow did exactly what I was describing. And when I went to check it out, I noticed something interesting.

They had quietly stopped building it.

I tried it myself to see why, and came to one conclusion. The output just wasn’t good enough.

Not "needs a little editing.” More like "I would rewrite most of this and start over".

The emails screamed "a robot wrote this."

And that's when it clicked for me.

The problem isn't the AI. The problem is what you give it.

If you give an AI zero context about who you are, how you talk, and what you're actually trying to say, you're going to get a robotic email that sounds like everyone else's robotic email.

This is a theory until I prove it, but the answer should lie in the quality of the scaffold (that’s the message drafted from the user’s voice message).

In other words, what examples can I feed SayMail, or what instructions can I prompt SayMail to follow in order to produce messages so good, chatGPT wouldn’t be able to replicate them on it’s own?

That's where prompt engineering comes in.

I've spent the past week tweaking the prompts behind SayMail's email generation, and this is what it looks like.

Round 1: The baseline

I started simple. I recorded a voice memo, let the AI process it, and looked at the result.

It was... not great.

The email had those signature AI tells: the long dashes, overly formal language, and sentences that felt like they came from a corporate template.

Round 2: Direct feedback

I went back into Claude and said, "Hey, here's a screenshot of what you produced. Here's what's wrong with it. Make it more casual, more human."

The second version was better, kinda. The corporate tone was gone. But it still didn't feel like something I would write, in fact, it felt too casual this time.

Round 3: An actual improvement

This is where things got interesting.

I created a Markdown file and copy-pasted a bunch of my own successful cold emails into it. Real emails I had sent. Emails that got replies.

Then I told Claude to reference those examples as a style guide.

The difference was immediate. The AI finally had concrete examples of what "good" looked like, not in theory, but in my actual voice.

Here's the thing about sitting alone and evaluating your own work: you're biased.

I can tweak prompts all day and convince myself the output is great. But that doesn't mean anyone else would agree.

That's why I'm moving to actual user testing.

If you signed up for the SayMail waitlist, check your email in the next 7 days.

I'm reaching out to a small group of 5-10 random people to test the first version of SayMail. You'll get a link to create an account, a short video walkthrough on how to get started, and a chance to try drafting your own messages.

Fair warning: this is a very early version. We're talking core V1 features only. But your feedback is going to shape what comes next.

If you haven't signed up yet, head to saymail.app and fill out the survey. There might still be time to make the first round.

That’s all for today. If you enjoyed this post, share it with a friend!

If they subscribe, I’ll send you my personal AI coding doc!

To refer, use your unique link.

See you next Tuesday 🤝

-Michael Ly

Building a retirement plan can be tricky— with so many considerations it’s hard to know where to start. That’s why we’ve put together The 15-Minute Retirement Plan to help investors with $1 million+ create a path forward and navigate important financial decisions in retirement.

Get your free guide today.

{{rp_personalized_text}}

Or copy and paste your unique link to them: {{rp_refer_url}}