Growth Newsletter #170

AI still doesn’t write usable copy.

It’s verbose and either overly formal or casual. People’s internal AI-detection algorithms are already well-tuned to sniff it out.

This is particularly true with cold emails. People are already wary of you.

But AI is good at writing tightly constrained variables. You can use these variables inside templates written by humans to make them feel fully personalized to get responses like this:

Here’s how to use AI to make cold emails that work

When I talked with Matt from Aurora, their process really impressed me:

#1. Pick a specific dream customer.

Draft a specific email for them. This is a forcing function to write a great email you’ll use as a template.

#2. Break down the dream email.

Identify the static parts (your name, company, etc) versus the parts that should be customized for each person.

Don’t just constrain yourself to boring variables like company name. Instead, think:

• Common pain points your customers experience
• “Jobs to be done” of your product or their product
• Summary of LinkedIn posts or company/personal bios
• Their current goals as an organization (fundraising, scaling a channel, etc)

For example:

#3. Use the 3 C's to build prompts to fill in the blanks.

• Context: Give ChatGPT the context it needs to know what to do.
• Creative Constraints: Tell it exactly what you want the output to look like (x words or less, casual, lowercase, no quotes, etc.)

The data you feed into the prompts can come in two forms:

1. Explicit data:
   1. Data from LinkedIn, like a company's description or a post.
   2. Scraped data from company pages, job boards, fundraising databases, etc.
2. Inferred data: Input for a prompt can be the output from another earlier prompt. For example, for the first email in Aurora’s outbound campaign (example above)
   1. Use the LinkedIn company bio + scraped homepage as inputs to build a “dream ICP” output from ChatGPT.
   2. The ICP output (and the company bio + scraped homepage context) can all be used as inputs to another prompt to output a "pain moment" output.
   3. Both are used as inputs to determine what companies should look for in their prospects' job posts and whether they need their service.

What the above template looks like when you get AI to fill in the blanks (read Matt’s post for example prompts used):

^ Notice this doesn’t have a CTA. He’s just sharing an idea for free. He’s not trying to close a sale immediately. It warrants a response since it doesn't feel like a pitch.

If Mutiny or LexCheck want to run with the idea themselves, cool. If they want to learn more and maybe work with Matt’s agency, even better.

Setting this up is a bunch of work

It’s far less work than manually writing thousands of emails, but it still takes time and expertise to get it all set up, running, and converting. And way more effective than "spraying and praying."

If this is interesting, we recommend working with the folks at Aurora who supplied this tactic. They’re experts in all of this.

Yes, they sponsored this post, but I love this tactic because most cold emails are terrible, and AI is making it worse.

Learn more about Aurora →

Resources to help implement this yourself

• Read Matt’s blog post that dives deeper into all of this. It includes ChatGPT prompts and examples.
• He also has a few rough notes with more example prompts.
• Read our past tactics on Sales, Copywriting, and Email.
• Learn how to write really good hooks for your emails.
• Follow Rob Lennon. He's got some pretty good posts and courses on using AI.