Prompt Engineering for DTC Ecommerce: The 7-Part "Prompt Sandwich"

You ask ChatGPT to write your email copy. This is the future. Copywriting has never been so easy. But as the AI returns your text, all you get is lifeless, robotic slop in return. It takes longer to edit the AI slop into something usable than it would have to write the draft yourself in the first place!

You're not alone. With one simple strategy—the "7-part Prompt Sandwich"—you can transform your copywriting from repetitive disappointment into consistently compelling, brand-perfect emails. Let's go!

Why Prompts Fail

Most prompts fall short because they're vague or lack context. A generic prompt like "write a welcome series" will always yield flat, uninspired results. Compare this to:

"As an expert email marketer, craft a 3-email welcome series for our new gut-health supplement brand, [Brand Name, Brand URL]. Target health-conscious women aged 30-45 who struggle with bloating and fatigue. Include an engaging subject line, a short educational hook about gut health, and a soft-sell CTA. Max 200 words per email and use headings for easy skimmability."

Now you're setting the LLM up to succeed.

Anatomy of the 7 Layer Prompt Sandwich 

The "Prompt Sandwich" includes:

Order Matters 

Research shows large language models weigh initial tokens most heavily. That means early details in a prompt significantly impact LLM attention.

Leading your prompt with a clear Role and Task primes the AI effectively, resulting in higher-quality output. If the LLM knows where you want it to go early, it will do a much better job synthesizing the rest of the context you give it into the objective and output that you want.

In our research at Raleon, we've also found significant improvements in output quality by restating key details and requirements at the end of the prompt as well. If you find the LLM is consistently making a mistake, include an “Important Note” at the end of the prompt directly addressing that mistake. This should help quite a bit in output consistency and model obedience. 

Tweaking the Layers 

In crunch times (like when you're on a deadline for a weekend send and its Friday at 4:30!) or for lower priority tasks, you can condense the sandwich to essentials (Role, Task, Context). The 80/20 rule applies here; the core 20% of your prompt will drive 80% of the results. 

For subtle tone shifts, stack multiple examples. Modern LLMs have quite large context windows. Don't be afraid to include 3-5 examples (or more) to show the LLM the full range of outputs that you like. 

No examples? No problem. Examples are a powerful tactic for consistency in output, but they're not a hard requirement for driving high quality LLM collaboration.

PRO TIP: Speed things up by dictating prompts into transcription software (Siri, WISPR, or any other app will do). LLMs excel with natural speech, no edits required. Speak and go!

Common Pitfalls & Fixes

• Overloaded Context → Be concise, direct, and relevant. LLMs can handle big context windows, but there is a limit. Keep the LLM focused with context that matters and nothing distracting.

• Shifting voice mid-prompt → Consistently reiterate brand voice.

• Forgetting Format → Always explicitly request your desired output structure. Be as specific as you need!

Hands-On Examples (Copy-Paste Ready)

Use these examples to level up your prompts today. Copy and paste them into your LLM of choice, add in your brand details and specific context, and see how they do for you. Use an AI transcription app to update them even faster (and definitely copy/paste context you've already assembled!).

1. Win-Back Email Copy

Role: You're a top-tier email copywriter.
Task: Craft a compelling win-back email targeting customers inactive for 90 days.
Context: This win-back campaign is for [Brand Name, Brand URL], premium skincare for busy professionals. This campaign is being written for existing customers who have engaged with a previous email or visited the website in the last 90 days, but haven't made a repeat purchase. Follow a problem/solution format to maximize the conversion rate.
Constraints: 100 words, output in markdown in a code block, include 3 engaging subject lines I can choose from, and write in a conversational tone.
Quality bar: Hemingway Grade 6 readability, include exactly one emoji

2. Subject-Line A/B Test Generator

Role: Act as an email marketing specialist.
Task: Write 10 engaging subject lines (max 45 characters) and preview texts for our Spring Sale campaign.
Context: These subjects will be used for a campaign for [Brand Name, Brand URL], premium skincare for busy professionals. The campaign is targeting our “Ready to Replenish” segment, which are customers who have made a purchase and our predictive model indicates a high likelihood of repurchasing in the next 2 weeks. 
Constraints: Avoid spam triggers, highly clickable. No emojis.
Format: Markdown numbered list.
Important Note: After generating the 10 subject lines, select your top 3 choices with brief reasoning

3. Post-Purchase Nurture Flow Brief

Role: You are an experienced retention marketer.
Task: Outline a 5-email post-purchase nurture sequence for first-time buyers of [Hero Product] from [Brand Name, Brand URL]. Context: Our current post purchase sequence is underperforming industry standards, and we need to get our numbers up by implementing gold standard tactics for the Post-Purchase sequence. We use Klaviyo as our ESP; include relevant configuration details such as send delays, conditional logic (if any), etc. Also include the goals and short hooks for each of the emails in the sequence and why they work together to improve our post-purchase flow..
Format: Output as a table with appropriate columns, each email is in its own row

PRO TIP: Use an LLM to write the prompt for you! Open up ChatGPT or another LLM, then using voice mode or a dictation app, simply tell the LLM what you need. Encourage it to ask you questions to get all the information it needs, and then the LLM will output a high quality prompt that's ready for use!

Instantly Upgrade Your Prompts

Writing great prompts is critical to using AI to its full potential. It's an investment that will pay dividends for you time and again. But it can still be a challenge managing multiple prompts, changing context, and evolving goals over time. 

Raleon was built specifically for DTC marketers who want the benefits of state-of-the-art AI without the hassle of managing, updating, and copy/pasting into and out of ChatGPT. Raleon dynamically and automatically gives the model the best possible objectives, context, and constraints, integrating real-time brand, Klaviyo, and Shopify data—no manual copy-paste required. 

• Reduce copywriting time by 50% or more.

• Generate precise, brand-aligned emails from your first prompt.

• Build reusable templates for various campaigns and clients.

To see how Raleon can help your agency or brand send more emails with better strategy, segmentation, and less cost, schedule a demo today.