Forem: ZeeeTH

I Built an AI Email Tool That Checks for Spam (Because Most AI Emails Are Terrible)

ZeeeTH — Fri, 11 Jul 2025 13:00:00 +0000

I was using ChatGPT to write cold emails for my consulting work. They looked great but got zero responses.

Turns out they were scoring 35/100 on spam filters. Basically guaranteed to hit spam folders.

So I built ColdCopy - an AI email generator that checks deliverability before you send.

The Problem

I tested popular AI email tools:

ChatGPT emails: 30-40/100 spam score
Other AI tools: 25-45/100 spam score

They optimize for what sounds good to humans, not what passes spam filters.

The Solution

ColdCopy generates emails then runs them through SpamAssassin to catch issues:

const checkDeliverability = async (subject, body) => {
 const response = await fetch('https://spamcheck.postmarkapp.com/filter', {
   method: 'POST',
   headers: { 'Content-Type': 'application/json' },
   body: JSON.stringify({
     email: `Subject: ${subject}\n\n${body}`,
     options: "long"
   })
 });

 const result = await response.json();
 return Math.max(0, 100 - (result.score * 6));
};

If issues are found, AI automatically fixes them while preserving the core message.
Results

ColdCopy emails: 80-95/100 spam score
Other AI tools: 25-45/100 spam score

Users report 23% average improvement in deliverability scores.
Tech Stack

Next.js + Tailwind CSS
OpenAI GPT-4 for generation
Postmark's SpamAssassin API
Vercel hosting

What I Learned
Deliverability is a huge overlooked differentiator. Most AI tools compete on writing quality, but if emails don't reach inboxes, quality doesn't matter.

Try it: coldcopy.xyz
What boring-but-critical problems are you solving?

48 Hours After Launch: What I Learned Building ColdCopy

ZeeeTH — Thu, 10 Jul 2025 22:14:29 +0000

Two days ago I soft launched ColdCopy, my AI-powered cold email generator with built-in A/B testing. Here's what happened and what I learned.

The Numbers

78 visitors, 91 page views in 48 hours
90% bounce rate (ouch, but expected for a single-purpose tool)
~15 people actually tried the generator (estimated from API calls)
1 email signup from someone wanting premium features
Multiple platforms tested: HN, Dev.to, Reddit, Discord

What Worked

Hacker News was the clear winner. Both my posts there drove significant traffic, especially when I focused on the A/B testing angle rather than just "I built this."

The A/B testing feature resonated. Comments like "That's exactly how I use ChatGPT - asking for variations" validated this wasn't just feature bloat.

Dev.to community engagement. You all are genuinely supportive! Got helpful feedback and hearts on my posts here.

What Didn't Work

Reddit was a disaster. Posted in relevant communities and got my account suspended for "spam." Their automated systems are broken for indie makers right now.

Conversion rate reality check. Only ~19% of visitors actually tried the tool. The other 81% bounced immediately, which taught me my value prop needs to be clearer upfront.

Unexpected Lessons

Time zones matter. Posting from Australia during US night hours killed initial momentum. My biggest traffic spike happened when America woke up.

Comments > Upvotes. One thoughtful comment validating the concept was worth more than dozens of passive likes.

The problem is real. Multiple people mentioned they use ChatGPT the same way I do - asking for variations. This confirmed I'm solving a genuine pain point.

What's Next

Adding payment processing this week to test if people will actually pay for unlimited access. The one email signup suggests there's conversion potential.

Also planning a proper Product Hunt launch once I have a few more features polished.

Key Takeaway

Ship fast, get feedback faster. Two days of real user interaction taught me more than weeks of planning. The tool works, people use it, and the A/B testing feature is genuinely useful - not just clever.

Building in public is both terrifying and invaluable. Thanks to everyone here who's been supportive! 🙏

Try it: coldcopy.xyz

What's your biggest challenge with cold outreach? Always curious to hear from fellow makers.

How I Added A/B Testing to My Cold Email Generator (and Why)

ZeeeTH — Wed, 09 Jul 2025 22:28:37 +0000

Yesterday I soft launched launched ColdCopy, a cold email generator. But I kept getting stuck on one question: "Should this email be more direct or more friendly?"
That's when I realized - why not generate both and let the user decide?
The Problem
Cold emails are tricky. The same message can work completely differently depending on tone:

Too formal = sounds robotic
Too casual = unprofessional
Too direct = aggressive

Instead of guessing, what if users could see all approaches side-by-side?
The Implementation
Before: One API call, one result
const response = await openai.chat.completions.create({ model: "gpt-4", messages: [{ role: "user", content: prompt }] });

After: Three API calls, three variations
const tones = ["Professional", "Friendly", "Direct"]; const promises = tones.map(tone => openai.chat.completions.create({ model: "gpt-4", messages: [{ role: "user", content: getPromptForTone(tone, formData) }] }) ); const results = await Promise.all(promises);

The UI Challenge
Displaying three emails clearly without overwhelming users:

Responsive grid layout (3 columns on desktop, stacked on mobile)
Clear labels for each tone
Individual copy buttons for each version
Same glassmorphism styling as the main form

Early Results
Users are actually using it! People seem to like comparing options rather than wondering "what if I tried a different approach?"
The conversion rate stayed strong (27% of visitors actually try the tool), which suggests the extra complexity isn't scaring people away.

What I Learned
Prompt engineering for tone is harder than expected
Users want options, not just "one size fits all"
Sometimes the obvious feature is the right feature

Check it out: coldcopy.xyz
What do you think - is A/B testing for AI-generated content useful, or just feature bloat?

I Built ColdCopy in 3 Days to Stop Writing Terrible Cold Emails

ZeeeTH — Wed, 09 Jul 2025 01:42:41 +0000

Hey Dev.to! 👋

Just finished building ColdCopy because I was tired of spending 20+ minutes writing cold emails that still sounded terrible.

What it does: Fill out a form with your details + recipient info, pick your tone, and get a personalized cold email in 30 seconds.

Tech stack: Next.js 14, TypeScript, Tailwind CSS, OpenAI API

Why I built it: Cold emails are necessary but painful. You either spend forever crafting each one or send generic templates that get ignored.

Built it in 3 days focusing on:

Clean, minimal UI
Actually personalized output (not just mail merge)
Multiple tones and message types
Rate limiting for MVP testing

Currently free to try: https://coldcopy.xyz

The hardest part was prompt engineering - getting AI to generate emails that sound human rather than robotic corporate speak.

Looking for feedback from fellow developers! Does this solve a real problem or am I overthinking cold emails?

What's your biggest struggle with outreach/cold emails?