AI Learning Journey  /  6 min read

I told one of our clients we’d write their next email sequence using AI only. No me at the keyboard for the actual copy. I’d write the prompts and review the output. The words on the page would be ChatGPT’s.

Five emails. One real campaign. 4,200 subscribers. I tracked it for a full month.

Here’s what actually happened.

The setup

The client runs a small wellness brand selling supplements through Shopify. They had a list of 4,200 people who’d bought once and never come back. We needed a re-engagement sequence. Five emails over three weeks.

I had two reasons for letting AI write the whole thing.

First, I wanted to know if it could actually do the work. We hear a lot about AI replacing copywriters. Most of what I’d seen so far was rough. I wanted to test the harder cases.

Second, my client had a tiny budget. If AI could do most of the lift and I just sharpened the output, we could deliver something she could afford.

The 4 prompts

I used four prompts across the five emails. The same shape each time.

Prompt 1. Voice training.

Read these 12 product page descriptions and the brand’s About page. Capture the tone in 5 bullet points. Then write three sample sentences that match it.

This was the most important prompt. Without it, every email sounded like generic marketing. With it, the emails sounded like the brand.

Prompt 2. Strategy.

Here’s the audience profile and the offer. Write the subject lines, hooks, and CTAs for a 5-email sequence. Email 1 reactivates. Emails 2 to 4 nurture with story and proof. Email 5 creates urgency. Don’t write the body yet. Just give me the architecture.

Prompt 3. Drafting. (Run for each email.)

Write Email 2 of the sequence using the architecture from before. Match the brand voice from the first prompt. 180 words. One clear CTA. No exclamation marks. No words like “amazing” or “transform”. Read like a friend, not a brochure.

Prompt 4. Editing pass.

Read this email. Rewrite any sentence that sounds like AI wrote it. Keep the meaning. Use shorter sentences.

That fourth prompt did more work than I expected. ChatGPT noticed its own tells.

The results

We sent the sequence over three weeks in March. Here’s what came back.

Metric	Result
Open rate (avg across 5 emails)	26.2%
Open rate, Email 1	31.0%
Click-through rate	3.4%
Reply rate	0.8%
Unsubscribe rate	0.3%
Sales attributed (30-day window)	47
Revenue from sequence	£2,914

The benchmark for a re-engagement sequence in this client’s industry is roughly 18 to 22% open rate and 1 to 2% click rate. We hit 31% open on Email 1 and stayed above 24% throughout.

The number that mattered most: 47 sales attributed in the 30-day window. Average order value £62. That’s £2,914 in revenue from a list that was about to be deleted.

Pure AI cost? About £3 of API credits. My time? Two and a half hours, mostly on the first prompt and the editing pass.

What worked

The voice-training prompt.

Without it, every email reads like a robot. With twelve pages of brand content fed in first, the output sounded right almost every time.

Limiting the body length.

180 words felt mean at first. It made every email tighter. People are skimming. Tighter is better.

The “rewrite anything that sounds like AI” pass.

ChatGPT noticed its own tells. Too many adjectives. Sentences that all start the same way. It edited better than I would have.

Asking for one clear CTA per email.

AI’s instinct is to give you three. Specifying one fixed it.

What didn’t work

Hooks.

ChatGPT writes safe, generic openings. I rewrote every hook by hand. The opening line is the only thing that decides if the email gets read. AI couldn’t get there.

Subject lines.

Same problem. Every option felt like 100 other emails the subscriber was getting that morning. I wrote those manually using a 30-year-old habit of testing three angles per email.

The middle emails sagged.

Email 2 and Email 3 had less narrative pull than the rest. AI defaults to listing benefits. People don’t open emails for benefit lists. They open them because something feels personal. I had to add real stories from the brand’s customer reviews. AI couldn’t invent these.

The lesson

AI can do the heavy lifting. The parts that actually move the needle are the parts AI can’t fake. The hook. The subject line. The personal story. The voice.

If I’d let AI write 100% of this with no editing pass, I’d estimate the open rate would have come in around 16%. Below benchmark. Possibly enough that the client would have pulled the plug.

With prompts that trained the voice, structure that constrained the output, and human edits where they matter, AI did about 80% of the work in 20% of the time.

That’s the real number. Not “AI is replacing copywriters.” It’s that AI plus an experienced editor is faster, cheaper, and roughly as effective.

I’m using this approach on every sequence now.

If your email list is gathering dust and you’d like to set up something like this for your business, message us on WhatsApp. Happy to talk through how it would work for you specifically.