How I Write Emails With GPT-3
Final tally was somewhere north of $500,000 when everything was said and done.
Last year, I optimized and wrote a 7-day email onboarding sequence for a client (selling software).
It ended up generating over $500,000 over the course of a few months.
And I did so with the help of GPT-3.
(My client knew and wanted me to do so. Because as I was doing so, I was also training their marketing team on how to use AI for marketing).
The onboarding sequence for SaaS was meant to turn free trial users into paying users.
Which means these emails need to hit on several things over about a week:
Get the user to actually use the features of the software.
Get the user to the “A-ha!” moment, where they experience the true value of the software and become invested in using it.
Counter objections along the way.
Frame the software as beneficial in all sorts of ways (future-pacing, dimensionalize benefits, implication chains, etc.)
Trigger emotional investment and use logic to justify paying for it.
Every email had a purpose, outcome, and structure.
This makes it fairly easy to get good text output from prompts you give GPT-3 (or whichever tool you use).
Let’s look at one email example.
The purpose of Email #6 (in the sequence) was to provide benefit bullets, covering logical and emotional benefits, that communicated the value of the software.
I gave GPT-3 some existing content already written (so it knew the topic and some details).
Then I ended with a simple prompt:
“Give me a list of both logical and emotional reasons as to why and how this fleet management software can help fleet managers with their job.”
Click ‘Submit’. Watch the circle whirl. And I got this:
There’s some repetition and almost duplicate content. But there’s nuance.
From there, I asked GPT-3 to give me a paragraph with an argument for using fleet management software, along with telling me how it works and what to do next:
The black boxes are redacting the client’s name.
Why?
Because it’s still a bit controversial to have your marketing done by robots. And, if people know, you’ll get novelty clicks and attention that makes it hard to get real, accurate results and data.
(Sidenote: Everything AI-related I do with clients is under NDAs and other agreements. I only share names and details if I get approval from a client. This is to both protect their IP, my IP, and again: get as accurate results from testing as possible. It’s not possible if you get novelty views, clicks, and people gaming your ads or emails).
Anyway.
All of that text is pure GPT-3 output with zero edits.
With some editing, this can be a great email.
(And it took me maybe 4-5 minutes in total to get the bullets and other email copy written for me).
The email, after some editing:
Can you see the possibilities here?
It’s not about writing faster (though that can happen).
It’s not about expecting text that’s flawless and ready to be used (though that can happen, too).
It’s about getting text, in this case email copy, that you can mold and edit.
Several of the emotional and logical bullets were reused in ads and in headlines. And they performed really well (not just in click volume but in CPC, CAC, and other metrics).
The key to getting useful output from any AI tool is to use it with a process and structure.
Don’t prompt it with:
“Write me an email about how to lose weight”
You won’t get a lot of useful text back.
Instead, follow a process of:
Gather any materials and research that would help you write copy.
Pick a template or structure you would use.
Prompt GPT-3 (or in whichever tool you use) a few times with different information and variations of your prompt.
Take what’s useful.
Edit into final copy.
I’m trying to help you see this.
So you can start using AI and GPT-3 in your own process.
I don’t recommend a specific tool other than GPT-3 directly.
But tools like Jasper.ai, Copy.ai, Copylime.com, and so on are all great for different things.
Don’t ask “which tool is best??” (it’s GPT-3).
Instead, ask yourself how you can implement and start using it in your workflow.
If you do, you’ll soon have emails done that generate serious cash.
Discovery and magic comes from working with the tools.
Stop procrastinating. Stop wondering which tool. Just get started.
If it wasn’t obvious by now, let me say it again:
AI, and in particular, GPT-3 is capable of writing anything.
It may not be award-winning prose or poetry that makes you weep right away.
It may not be Facebook ads that produce millions instantly. Or sales pages doing the same.
But it’s perfect for ideation, drafting, surprising insights, interesting twists—and text you can mold into gold.
And soon, GPT-4 will be available.
In the meantime:
Did you see? Here’s how “bionic ads” can generate dollars
Talk soon,
Samuel Woods
The Bionic Writer