How Neptik’s AI is Redefining Email Content

Until mid-2024, at Neptik, we relied solely upon our own internal content writing capabilities.

With over two and a half decades of sales and marketing experience, and a strong niche for B2B business development, we were pretty good at crafting them. We'd never used an email content generator, but we had developed a manual framework for creating content that cut through the noise of competing outreach emails and delivered great results. Here’s a quick breakdown of the philosophy behind our approach.

Email content do’s:

• Remember, we haven’t been invited to pitch—so don’t write email content that reads like a deck, brochure, or landing page.
• Focus on tapping into mental availability to spark curiosity and interest in learning more about what you’re offering.
• Highlight what’s in it for them—prospects want to know how you can make their lives easier, better, save them money, reduce hassle, or help them achieve more.
• Tailor your message to your audience—targeting specific roles in vertical industries allows you to focus on their challenges and how you can help. While some campaigns work for broader audiences, it’s fair to say, a lot of the best results come from crafting content that truly resonates with specific sectors and job roles. Simply merging in targeted keywords to achieve the same effect doesn’t cut it anymore—people spot those tricks a mile away.
• Craft your emails like a tabloid newspaper—not with oversized fonts or shocking headlines, but by designing each paragraph to grab attention and naturally lead into the next. By the time prospects reach the call-to-action (CTA), they’re already engaged and invested.
• Sprinkle in touches of credibility and assurance—mention recognised accreditations, certifications, or big-name clients. It builds trust and makes prospects think, “If it’s good enough for them, it’s good enough for us,” or “We want to be where they are, so let’s talk to these guys.”
• Keep it conversational—remember, it’s meant to feel like an email from you, so write like yourself. Don’t get too formal, avoid coming across as arrogant, but don’t sound desperate either. We like to call this tone: “Humble Confidence.”

Email content Don’ts:

• Don’t write a novel—remember, they didn’t search for you or request your brochure. You’ve appeared out of the blue, so make it easy to digest. Stick to 140-200 words max for the first email, with smaller follow-ups afterward. Use those follow-ups to explore different angles that show how you or your offering can help, confirm you’ve got the right person, and end with something memorable—at the very least, to keep the door open.
• Don’t get caught up in all the little details about what your company offers that you’re passionate about—what we call “the curse of knowledge”. At this stage, the prospect typically doesn’t care about those specifics.
• Don’t use AI tools to scrape the web for things like blogs prospects have written, awards they’ve won, posts they’ve shared, or customers they have on their site. Lately, we’ve seen this a lot, and it’s obvious when it’s been done. Think of it like this—imagine going on a first date and reeling off everything you know about them. It’s creepy. In email prospecting, it comes across as fake. People know it’s not genuine, and I’m pretty sure we’ll look back on this early use of AI and laugh (or cringe).
• Don’t make promises—it often comes across as insincere, and guarantees are a surefire way to trigger spam filters.
• Don’t rely solely on tools like ChatGPT to craft your emails—it’s blatantly obvious when you have (my pet hate opener: "I hope this email finds you well").

For years, the responsibility of producing our email content fell to me.

During our client onboarding process (which we still do), we’d host a workshop to gather insights, then create the initial copy for the client. This typically included 5-6 stages: an initial email followed by the follow-ups. After a client’s reviewed and approved any tweaks, we’d sign it off. From there, we’d create daily variations of the initial email and follow-ups, ensuring that each campaign launched from Monday to Friday (as we launch new campaigns every day) featured:

• Different openers
• Different paragraph structures 
• Different sentence structures with varied synonyms
• Different closers, tailored to propose specific days for a call (based on when the email is scheduled to land in the prospect’s inbox)

However, despite all the above changes, we’d deliberately retain:

• The same tone of voice
• A similar content word count
• The same information
• The same CTA

Why did we do all of this?

• To personalise the email based on when it would be sent—for example, opening with “Hope your week’s started nicely” on a Monday morning.
• To propose specific days for a call, such as “I was wondering how you might be fixed tomorrow, or sometime Mon/Tue” in a Thursday email.

But also:

To avoid spam folders. Our findings show that simply changing merge fields or using Spintax isn’t enough to achieve consistent high levels of inbox deliverability. These tactics might work temporarily, but it doesn’t take long before email service providers' alarm bells start ringing. As you can imagine, done manually, this stuff takes time to put together. But, we did it for one simple reason—it works. And because it works so well, we knew the final outputs we were producing were far too important to compromise.

Over the years, we’ve tried bringing people in to take on this role, but it’s not been easy. Here’s why:

1. Good prospecting content isn’t like writing typical marketing material—it’s not the kind of work content specialists or copywriting experts typically aspire to. It’s about crafting sales emails that don’t feel salesy. They need to look like they were written off the cuff, while still steering clear of a restrictive library of spam words.
2. Even if you nail the initial content that the client signs off on, creating a labyrinth of variations that don’t conflict with each other—yet say the same thing and still read as if written off the cuff—is no easy feat. It’s the krypton factor of content writing, and sadly I’ve seen really solid creative writers completely stumped on this part.

Hopefully, the picture’s becoming clear…

We’d developed a framework that worked brilliantly, but it was tedious, and no one wanted to take it on. As a result, it kept falling to me, consuming so much of my time.

So, what did we do?

We set out to see if we could get AI to create an email content generator that writes  the way we do—with the expertise and writing philosophy we pride ourselves on. The big challenge was getting it to generate believable (ie: I wrote it) variations, not just churn out random differences for the sake of variety, which was pretty much the case when we’d tried tools like ChatGPT on its own. Whilst we were marinated in ideas, and had a blueprint of the logic we wanted to follow, we decided to bring in an AI expert to collaborate with us, rather than attempting to build everything in-house. The next steps weren’t without their challenges. We initially thought we’d focus on getting the AI to produce the initial content—the content our clients sign-off on before we begin creating daily variations. Our first approach was to:

• Train the AI with thousands of successful email campaign examples we’d written.
• Develop a questionnaire template based on the workshop questions we use to guide the AI’s process.

The concept was that if we provided the trained AI with the answers to our workshop questions, it would generate email content in line with our style and outreach philosophy. However, it didn’t quite work out that way, and what we got was a bit of a dog’s dinner. Back to the drawing board, we shifted gears and asked the AI what questions it needed to ask for the email content generator to produce results similar to the content we had trained it on. Surprisingly, the questions it generated were vastly different from those we typically use in our workshops. We ran tests, and this reverse-engineering approach proved effective—the new email content was much closer to the standard and style of content we’d produce independently. That said, it still wasn’t perfect. So, we took it a step further, combining advanced prompting techniques with AI.

Using two language models (ChatGPT-3.5 Turbo and Claude 3.5 Sonnet), we:

• Used the answers updated into the new questionnaire template to create the initial email content structures.
• Fine-tuned prompts to optimise email openers and propose suitable call days based on email delivery timing.

We then introduced a feature to adapt content tones using a temperature gauge or a “language enricher” function. This allowed us to also convert email tones into “Neptik Styles,” which were generated based on the thousands of successful hand crafted emails we’d trained the models on. With this system, we could then manually make final edits to the content before submitting for client approval. Following that it would be time to work on generating the variations. As mentioned earlier in the blog, before we decided to build our own email content generator, we experimented with using ChatGPT to create content variations. However, we found that it often produced variations for the sake of variety, without adhering to the parameters and styles we’d asked for. During the development of our AI email content generator, we took a more structured approach by segregating the campaigns we’d trained the language models on into two categories: initial copy and variation copy. This allowed us to train the AI on how to:

• Maintain a consistent tone
• Relay information differently while staying accurate
• Keep a similar word count.
• Position calls-to-action in unique ways

Initially, the AI still produced variations that we wouldn’t recommend using. To address this, we developed a word and phrase exclusion library to work along-side the content we'd trained the AI on, and the system parameters we'd introduced—maintaining language/tone and structural consistent, and preventing the AI from going off the rails.

Email Content Generator - End result.

After the functionality was nailed, we spent time turning a complex tool into a pretty straightforward user interface. Being a stickler for user experience and attention to detail, I wouldn’t say it’s perfect, but for our internal requirements it certainly does the trick for now.  In addition to speeding up the process of writing both the initial email content and the variations, our team's now able to produce content with greater confidence, test it effectively, and adapt it quickly when needed. Our email content generator isn’t available as a SaaS product—for now, we’ve chosen to keep it as a little gem exclusive to our Neptik clients. That being said, to further strengthen our deliverability strategy, we’re now working on integrating a spam word analysis feature (rather than running our content via a separate spam assessment tool). We're also integrating a Spintax feature on top of our variation model (offering literally endless unique content variants), and refining our prompt logic with optimised structure-awareness. Plus, a few tasty tweaks on our user interface!