Artificial intelligence didn’t write a single word of my latest email campaign. And yet, without it, that campaign would never have existed.

The result? A send to 200,000 subscribers with a 26% open rate and a 4.2% CTR - the best performance in months. No wonder the first comment I heard was: "Seriously, was this done by ChatGPT?”

No, it wasn’t. This wasn’t AI’s work. It was mine, but a few people talked about it.

Instead of creating, AI is monitored. Instead of writing, it was reviewed.

In this article, I’ll walk you through the entire process. I’m a Marketing Manager at MyLead, responsible for email marketing to our database. Over time, I’ve developed a workflow in which AI acts as a control mechanism, not a content creator. And in this article, I’ll show you exactly how I do it.

What you’ll learn:

• How I build context for AI instead of using ready-made promptsHow I created a custom GPT for email validation (trained on YouTube knowledge)
• Why does AI review my emails instead of writing them
• How I use Bouncer to clean my list and Mailgun to monitor campaigns
• My complete workflow from idea to send

Why AI doesn't write my emails

I know it sounds paradoxical in an article about AI in email marketing. 

But these are the most important lessons I’ve learned.

If you let ChatGPT generate an email from scratch, you’ll get a text that sounds like… ChatGPT. Everyone recognizes it now:

“Dive into the world of…”

“Discover the potential…”

“Don’t miss this opportunity…”

These are dead phrases. People ignore them.

That’s why my process looks different:

• I write the email - based on what I know about our users

• AI reviews it - checks for spam red flags, structure, and clarity of the CTA 

• I revise it - based on AI feedback

AI is my editor and reviewer, not my copywriter. And it works.

How I build context for AI

I don’t use ready-made prompts from the internet. Every time, I build context from scratch, tailored to a specific campaign.

I mainly use ChatGPT and Claude, depending on the task.

ChatGPT - for quick validation

When I already have a draft and want to quickly check for obvious issues, I paste it into ChatGPT with proper context.

I don’t ask it to rewrite the email. I ask for feedback. 

I work with a custom GPT, which I’ll explain later in the article.

Claude - for deeper analysis

I use Claude when I need more in-depth analysis, especially in sensitive niches (finance, health). I paste the full email along with additional context about the target audience.

In my opinion, Claude handles longer content better, so I use it when I need to analyze more complex texts.

Prompt Cowboy - when I’m stuck

Sometimes I sit in front of a prompt and don’t know how to phrase it so AI understands exactly what I want. That’s when I use Prompt Cowboy, a tool that helps refine prompts. I enter my rough idea. And Prompt Cowboy suggests a better formulation. It saves a lot of time.

It’s also worth asking AI whether it understands the context and the task. I treat AI like a member of the marketing team; I ask for its perspective, suggestions, and observations. In my experience, this more “friendly” approach works much better. Exchange messages, build context, ask follow-up questions, but also demand quality. Sometimes a simple prompt like “Try again. I know you can do better” is enough to significantly improve the response.

My custom GPT for email validation

This is my secret weapon. I created a custom GPT specifically for validating email campaigns. Its job is to critically analyze my content. The model I’d fed with the right knowledge and context, which allows it to genuinely review emails rather than give generic advice.

How I built it

I had a problem: ChatGPT didn’t know enough about writing newsletters that actually convert. It also lacked context about our company and audience.

So i taught it.

What I did:

1. Found the best YouTube videos about email marketing and copywriting

2. Downloaded their transcripts (these are tools for that)

3. Converted them into clean text

4. Uploaded them into the custom GPT as a knowledge base

Now my custom GPT knows how to write effective newsletters. It draws on dozens of hours of high-quality educational content.

How use it

I paste my email draft and ask for feedback. I get insights from real knowledge in those materials, not vague generalities.

Bonus: You can do exactly the same.

Find good email marketing content, download transcripts, and upload them into a custom GPT.

It takes about 2 hours to work, and it takes months to work. You can apply this framework to any topic, not just email marketing.

Bouncer - cleaning the email list

The best email in the world won’t help if you’re sending it to dead inboxes. That’s why I regularly clean my list using Bouncer.

Why Bouncer

I tested multiple verification tools. Bouncer won because:

• It’s fast - verifying 100k emails takes minutes

• It’s accurate - detects spam traps, temporary emails, inactive inboxes

• It includes a spam test - I can check my email content before sending

How I use Bouncer

My process:

• Every 3 months, I clean the entire MyLead database with Bouncer

• I remove all addresses marked as invalid or risky

• Before every major campaign, I test the email in Spam Checker

• Spam score must be below 3 - otherwise, I review the content

Result:

Bounce rate at MyLead is usually below 1.5%, which is an excellent result for a 200k list.

Mailgun - sending and monitoring

I use Mailgun for sending emails.

Not because it’s the cheapest (it’s not), but because it gives me full control over what’s happening with my emails.

Inbox placement test

Before every major campaign, I run an Inbox Placement Test in Mailgun.  It sends a test email to various providers (Gmail, Outlook, Yahoo) and shows where it lands.

My goal: at least 80% inbox placement. If it’s, I revise the email and test again.

Real-time monitoring

Once the campaign is live, I stay in Mailgun watching stats in real time:

• Open rate - how people react to the subject line

• Click-through rate - whether the CTA works

• Bounce rate - if it suddenly spikes, I stop the campaign

This is the most stressful moment, watching the number go up… or not.

But without this, you don’t know what works.

My complete workflow

Now you have all the tools. Here’s how it looks in practice, from idea to send.

          1. I write the email (no AI)

I sit down and write. I base it on what I know about our users: What do they need? What language do they understand?

The draft doesn’t have to be perfect. It’s a working version, and that’s fine.

2. I review it in custom GPT

I paste the draft into my custom GPT. I ask about structure, flow, and CTA clarity.

I get concrete feedback. I revised the email accordingly.

3. I test the spam score in Bouncer

I copied the final version into Bouncer’s spam checker.

If the spam score is above 3, I fix the problematic parts and test again.

4. Inbox placement test in Mailgun

I sent a test email via Mailgun and waited about 5 minutes for the report.

If inbox placement is below 80%, I go back to step 2.

5. I clean the segment in Bouncer

If it’s a new segment or hasn’t been cleaned recently, I run it through Bouncer and remove bad addresses.

6. I send

I configure the campaign in Mailgun.

I set throttling (around 5,000 emails/hour).

I click “send.

7. I Monitor

For the first 30 minutes, I stay in Mailgun watching the numbers. If something goes wrong, I can stop the campaign immediately. 

The entire process takes 30-45 minutes. At the beginning, it used to take 2-3 hours.

What to do - and what to avoid

Do this:

• write emails yourself; use AI only for review

• Build AI context per campaign

• Test spam score before every send 

• Clean your list regularly (at least every 3 months)

• Monitor campaigns in real time

Don’t do this:

• Don’t let AI generate the entire email from scratch

• Don’t use ready-made prompts without customization

• Don’t send without checking spam score

• Don’t ignore bounce rate, if it exceeds 2%, stop the campaign

• Don’t send to an unverified list

Frequently asked questions

1. How much do these tools cost?

• ChatGPT Plus: $20/month

• Claude Pro:  $20/month

• Bouncer: ~$50-100/month depending on volume

• Mailgun: ~$200/month at our scale (400k users)

Unfortunately, no. Custom GPTs require ChatGPT Plus.

You can paste transcripts manually into free ChatGPT each tome, but it’s much more cumbersome.

3. What are your stats at MyLead?

On average:

• Open rate: 22-28%

• CTR: 3-6%

• Bounce rate: <1.5%

This comes from consistency and testing. I started at an open rate of 12%. In affiliate marketing, a 20-25% open rate is already a solid result. 

4. Can I use only ChatGPT without Claude?

Yes. I use both, but if your budget allows only one tool, ChatGPT is enough.

Claude is useful in sensitive niches, but not essential. From my experience, Claude handles long-form analysis better. Since both tools cost roughly the same, the final choice is yours. I recommend testing ChatGPT for a month, then trying Claude.

5. What if I’m not good at writing emails?

Then this process is even more important. Custom GPT will teach you what works. After a dozen iterations, you’ll start understanding the principles. But still - write the emails yourself. Let AI review, not replace you.

Summary

AI in email marketing is not a content-generation tool. It’s a quality control tool.

I write emails because I know our users. AI checks whether I made mistakes. Custom GPT gives feedback based on best practices. Bouncer protects my list. Mailgun shows me what works.

It took me two years to develop this process. You can implement it in a week.

Start with one email. Write it yourself. Check it in ChatGPT. Test it in Bouncer. Send a small batch. See what happens. Improve. Repeat.

And if you need offers to promote, at MyLead, we have over 5,300 affiliate programs.

Join, choose a campaign, and start testing.

Create a free MyLead account.

https://www.mylead.global/register

Good luck!

Nikodem, MyLead