When Your AI Won't Stop Copying You (And How I Fixed It)

👟 Topic: When Your AI Won't Stop Copying You (And How I Fixed It)

👾 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

Ready to transform artificial intelligence from a buzzword into your personal revenue generator

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

Hey folks, two things today.

Remember that email I showed you last week?

The one I said was finally good enough to send?

Well, about 10 minutes after hitting publish on that article, I ran into a problem.

I kept generating new messages for different contacts, and something felt off. Every single one sounded... identical. Actually, not just identical. They were basically copying my example word-for-word.

No matter how many times I hit generate, SayMail kept producing the same email with the same phrasing. It was like the AI found my example and thought, "This is perfect. I'll just use this every time."

That's not what I wanted at all.

Here's the thing I learned this week: AI tends to mimic exactly what it's given, even if you explicitly tell it not to.

Last week, I fed SayMail a bunch of my own successful cold emails as reference. Real emails I had sent. Emails that got replies. The idea was to give it a "style guide" to follow.

And it worked. Kind of.

The AI absorbed my style alright. It absorbed it so hard that it just started copying my examples verbatim. Different person, different company, same exact email structure and wording.

That's the opposite of helpful.

I did what I always do when I hit a wall with AI. I hopped out of Cursor and opened Claude to talk through the problem.

The solution it gave me was surprisingly simple: replace my real email examples with fake ones.

Instead of:

Use something like:

So basically full-on fictional scenarios with made-up names, fake companies, and imaginary situations.

I asked Claude to generate three fictional cold email examples that captured the patterns of what made my real emails successful. Then I swapped those into the prompt.

The result? Night and day difference.

The new messages were creative. Each one felt distinct. They followed the pattern of good emails without copying any specific phrasing.

When you give an AI a real example, it latches onto the specifics. The actual names. The actual companies. The actual sentences.

But when you give it a fictional example, the AI can't copy the specifics because they don't apply to your real scenario.

Instead, it's forced to extract the underlying pattern, the structure, the energy, the approach, and apply that to your actual situation.

Once I fixed the generation issue, I finally felt confident enough to get this in front of real users.

BTW — For those of you who signed up for the waitlist, check your email. You should have gotten a message from me around 8 AM this morning with your account link and a quick walkthrough video.

Take your time with it. Poke around. Break things. I want to hear all of it.

Getting SayMail ready for users wasn't just about fixing bugs and sending a link.

The big thing I'm trying to figure out is this:

Is the workflow of recording a voice note and using that to generate an email convenient... or is it an obstacle?

If users find the voice-to-email flow helpful, then I'm on the right path. If they find it frustrating or awkward, I need to seriously reassess what SayMail is supposed to be.

I've already started thinking about what version 2 and version 3 could look like.

I don't want to overpromise, But here's the direction I'm headed:

Version 2: Email integration. Connect your Gmail or whatever provider you use, and push drafts directly to your inbox. No more copy-pasting. Also, a rotating subject line generator so you can pick the option that fits your style.

Version 3: This is where it gets interesting. Some form of LinkedIn integration. And user learning patterns, where you upload your LinkedIn profile or resume, and as you send messages, SayMail starts giving you insights on what's working and what's not.

But all of that depends on what I learn from this first round of testing.

If you haven't signed up yet, head to saymail.app.

I’ll be doing multiple v1 testing rounds.

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

Your time is better spent creating content, not managing ad campaigns. Google AdSense's automatic ad placement and optimization handles the heavy lifting for you, ensuring the highest-paying, most relevant ads appear on your site.

Earn with Google AdSense

{{rp_personalized_text}}

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