AI Prompt Test: How Structure Impacts Email Copy

I set out to test tone. What I discovered? The prompt itself can make or break your copy.

This all started with a casual test. I wanted to see how tone impacted AI-written emails. But the results? They revealed something much deeper.

Welcome to the first of many behind-the-scenes posts from my AI email experiments. The goal isn’t to replace marketers, but to train a really useful assistant.

Experiment Setup

The Use Case

I asked ChatGPT-4o to write a welcome email for a professional development platform that helps corporate learners upskill in leadership and communication. I kept things simple for this first round:

• • Model: ChatGPT-4o
  • Test Focus: Prompt structure
  • Email Type: Welcome email for a professional development platform
  • Tone: Friendly
  • Parameters: Both versions asked for a 150-200 word welcome email, ending with a clear CTA

Before I show you what I tested, let’s look at how I structured each prompt. A well-structured prompt doesn’t just help the model, it sets the foundation for a better result.

The Prompt Structure

To make the test repeatable, I broke each prompt into a consistent set of elements.

These are by no means the only elements we can use to structure our prompts. For example, you might also specify a writing format (like a bulleted list or a narrative story). Or you can instruct the model to emphasize or exclude certain topics.

The elements you’ll use depend on what output you’re looking to generate.

As you can see in the chart below, Prompt v2 has additional elements to guide ChatGPT in generating the best output possible.

Here’s how the two prompts compare in terms of structure:

Prompt Elements Used

How I Evaluated the Outputs

To evaluate each output, I used a custom Copywriting Rubric based on key marketing criteria:

• • Tone Accuracy (0-2): Did it match the prompt’s requested tone?
  • Clarity & Structure (0-2): Was it easy to read and logically organized?
  • CTA Strength (0-2): Was the call-to-action clear and compelling?
  • Alignment to Prompt Goal (0-2): Did it follow the assignment?
  • Originality & Value (0-2): Did it add any unique insight or brand value?

Each score is subjective, but the rubric helped keep comparisons consistent. And, it will evolve in future tests as I refine my evaluation methods.

Now that we have the structure set up, including how to evaluate our outputs, let’s take a look at the actual prompts. 

The Two Prompt Versions

Version 1: Basic Prompt

🤖 Prompt v1

Write a welcome email for a professional development platform that helps corporate learners upskill in leadership and communication. Use a tone that is friendly and approachable. Make it 150–200 words. End with a clear CTA to explore available courses.

📤 AI Output v1

Subject: Welcome to Your Next Step in Leadership & Communication 🚀

Hi [First Name],

Welcome to [Platform Name] — we’re thrilled to have you on board!

You’ve just taken a big step toward sharpening your leadership and communication skills. Whether you’re looking to lead with confidence, manage teams more effectively, or communicate with impact, we’re here to support your growth every step of the way.

Our platform offers expert-led courses designed with busy professionals in mind. You’ll find flexible learning paths, real-world insights, and practical tools you can apply right away—no matter where you are in your career journey.

Ready to dive in?

👉 **[Explore Courses Now]**

We’re excited to be part of your professional development journey. If you have any questions or need a recommendation, we’re just a message away.

Here’s to learning, growing, and leading with purpose.

Warmly,
The [Platform Name] Team

Evaluation for Version 1:

• • Tone Accuracy: 1/2
  • Clarity & Structure: 1/2
  • CTA Strength: 1/2
  • Alignment to Prompt: 1/2
  • Originality & Value: 1/2

What Worked:

• • Comprehensive overview of the platform offerings
  • Logical structure with introduction, features, and CTA
  • Addressed practical concerns like learning at your own pace

What Didn’t Work:

• • Generic language that could apply to any learning platform
  • Overloaded with features rather than benefits
  • Weak CTA that didn’t create urgency or excitement
  • Missing empathy for the learner’s specific challenges

Version 2: Role-framed + Audience + Flow

🤖 Prompt v2

You are a senior lifecycle marketer. Write a welcome email for a professional development platform that helps corporate learners upskill in leadership and communication. The recipient is a time-strapped mid-level manager. Use a tone that is friendly and approachable. Start with a personal greeting, share one benefit of the platform, and end with a clear CTA to explore available courses. Make it 150–200 words.

📤 AI Output v2

Subject: Welcome! Let’s Level Up Your Leadership

Hi [First Name],

Welcome aboard! We’re thrilled to have you with us.

We know your time is valuable—between meetings, team check-ins, and everything else on your plate. That’s why our platform is designed with busy professionals like you in mind. Whether you want to strengthen your leadership style, sharpen your communication skills, or inspire your team more effectively, you’ll find focused, practical courses that fit your schedule.

One key benefit? You’ll get bite-sized learning you can apply immediately—so you grow without the overwhelm.

Ready to dive in? Your personalized learning path is just a click away.

👉 [Explore Courses Now]

Here’s to growing your impact—one smart step at a time.

Cheers,
The [Platform Name] Team

P.S. Have questions or need help picking a course? We’re just a message away!

Evaluation for Version 2:

• • Tone Accuracy: 2/2
  • Clarity & Structure: 1/2
  • CTA Strength: 1/2
  • Alignment to Prompt: 2/2
  • Originality & Value: 1/2

What Worked:

• • Prompt Alignment: Nailed the audience, tone, and CTA flow
  • Problem-Solution Structure: Acknowledged the challenge and offered a solution
  • Benefit-Focused: Highlighted real user value
  • Strong CTA Setup: Clear and personalized

What Didn’t Work:

• • Subject Line: Too long, title case, not punchy
  • First Paragraph: Friendly but not concise
  • Second Paragraph: Buried empathy and value too deep

What I Learned

When comparing the two outputs side-by-side, a few differences stood out.

To wrap up the side-by-side evaluation, I highlighted the most telling performance factors from each version that shaped performance in each email. Here’s how they stack up:

Side-by-Side Comparison

These insights are already influencing how I structure my next round of tests — I plan to further explore how each prompt element affects emotional resonance and conversion potential.

• • Role framing matters
  • Prompt structure affects output structure
  • Quality over quantity
  • AI needs guidance on priorities

Why This Matters for Marketers

Even if you’re not planning to run formal experiments, the lessons here have real-world implications. Knowing how to guide AI outputs with clearer, more human-centered prompts can elevate the quality of your email marketing across the board.

• • Better prompts = better copy
  • Models respond to role framing
  • You don’t need to be technical
  • Prompt design is creative direction

Prompt Template for Email Marketing

🤖 Prompt

You are a [specific marketing role].

Write a [email type] for [product/service] that helps [target audience] achieve [specific goal/outcome].

The audience is [detailed persona description] who face [specific challenges].

Use a tone that is [tone descriptors].

Structure the email to:
1. [First section goal]
2. [Second section goal]
3. [Third section goal]
4. End with a CTA to [specific action]

Make it [length constraint].

Limitations

Before we wrap, it’s worth noting a few limitations of this experiment. These remind me to stay cautious when drawing conclusions. And, to stay curious about what to test next.

• • Single model test
  • Limited sample size
  • Subjective evaluation

What’s Next

Now that I’ve seen how prompt structure affects tone and clarity, I’m excited to keep testing. The next phase of my experiments will build on these learnings and help illustrate what is possible with thorough and thoughtful prompt design.

• • Test across AI models
  • Explore tone impacts
  • Develop objective criteria
  • Test with users

Final Thoughts

AI can get you 80% of the way there, and fast. But that last 20%? That’s where strategy, empathy, and human instinct come in.

Prompting well is a skill, but so is knowing what still needs rewriting. I’ll keep refining both as I go. And if you’re testing your own prompts, I’d love to hear what’s working for you (and what’s absolutely not).

Let’s build smarter together.

P.S. This experiment is part of a larger project: I’m building a Lifecycle Nurturing App designed to help marketers guide customers through smarter journeys. More soon. 👀