You're using ChatGPT or Claude to write your cold emails, but they sound generic. Like every other AI-generated email hitting inboxes right now. So you're manually tweaking them, spending 30 minutes per email, and still not confident they'll convert.

The problem isn't that AI can't write good cold email. It's that you're giving it bad prompts.

Most people throw a vague request at an LLM - "Write me a cold email about my consulting service" - and get vague garbage back. Then they blame the tool instead of understanding that prompt engineering for cold email is a specific skill. It has patterns, structures, and tricks that actually work.

Here's how to get AI to write cold emails that sound like a real human who knows the prospect's business.

The Core Framework: Context > Angle > Output

Every effective cold email prompt needs three layers. Most people skip the first two and wonder why the output sucks.

Layer 1: Context - You're telling the AI what business you're in, who you help, and what outcome you deliver. Not vague mission-statement stuff. Specific.

Layer 2: Angle - This is the hook. The specific reason you're emailing this person. What problem are you solving for them right now? What do they probably care about?

Layer 3: Output - The format and style you want. Length. Tone. Structure.

Most people write prompts that are only Layer 3. That's why they get generic AI slop.

The Actual Prompt Structure That Works

Here's a template you can use right now. Fill in your specifics and paste it into Claude or ChatGPT.

You are a B2B cold email copywriter. I help [specific business type] solve [specific problem] by [specific mechanism]. Our typical results are [outcome metrics]. Write a cold email to a [prospect role] at a [prospect business type] company. The email should: - Open with a specific observation about their business (mention something real about them or their industry) - Address one core pain point they face - Show how we've solved this for similar businesses (one brief example) - End with a specific, low-friction ask (not a meeting request - something smaller) - Be 75-100 words - Sound conversational, not corporate Do not use generic openers like "I noticed you're in [industry]" - be specific about something real about their company or recent news.

That's the structure. Now let's make it specific with an actual example.

Real Example: Web Design Agency Cold Email

Let's say you're a web design agency that specializes in converting freelance designers to agency owners. Here's what a filled-in prompt looks like:

You are a B2B cold email copywriter. I help freelance web designers transition to running an agency by building the operational systems they need so they can hire their first designer and stay profitable. Our typical result is designers going from $8-12k/month solo to $30k+ monthly revenue with a team. Write a cold email to a freelance designer who has been in business for 3+ years. The email should: - Open by referencing something specific about their design style or a recent project they posted - Address the core pain point: they're capped at what one person can do, but hiring feels risky - Share a brief example: "One designer we worked with was doing $10k/month solo, hired one person using our system, and hit $28k in month 6" - End asking if they'd be open to a 15-minute call to discuss how we structure this - Be 85 words exactly - Sound like a peer who understands the freelance-to-agency transition, not a corporate sales rep

See the difference? You're not asking for a generic email. You're giving the AI a role, a specific business context, a real target, and actual constraints.

The Angle Matters More Than You Think

The angle is what separates an email that gets opened from one that gets trashed. And this is where most prompts fail - they don't specify the angle clearly enough.

Don't just say "write an email to a SaaS founder." That's too broad. Instead, identify the specific angle:

Your prompt needs to tell the AI which angle you're using. Otherwise it defaults to generic.

Here's how to build this into a prompt:

I'm emailing [prospect role] at [company]. They recently [specific event - fundraise, job change, published content, hired new team member]. This matters because [why it's relevant to my service]. Write a cold email that opens by acknowledging this specific event, connects it to a problem I solve, and asks for a brief response (not a meeting). 80 words. Tone: peer-to-peer, not salesy.

That angle makes the AI write something targeted instead of scattered.

Style Constraints Beat Vague Tone Requests

Don't tell the AI "write in a friendly tone." Tell it what not to do.

Instead of: "Sound conversational"
Use: "Do not use: value proposition language, exclamation points, the word 'just', testimonial quotes, or calls to action with urgency. Avoid any mention of 'synergy' or 'solutions.'"

Negative constraints are more specific than positive ones. They tell the AI what to actively avoid, which makes the output tighter.

Also, add a style anchor. Reference a specific writing style the AI understands:

Write in the style of someone sending an email to a colleague they want to grab coffee with - direct, no fluff, mentions something specific about their work that you noticed.

That one sentence does more than "sound casual" ever will.

The Iteration Loop: How to Use AI Feedback

You're not going to nail it on the first prompt. But the way you iterate matters.

Don't just ask "make it better." Tell the AI what specifically is wrong:

Specific feedback generates specific changes. Generic feedback generates generic tweaks.

When to Use AI, When to Hand-Write

AI is great for the template. The opening structure. The logical flow. What it's bad at is micro-specificity about an individual prospect.

Here's the real workflow:

  1. Use AI to generate the framework and body copy (60% of the work)
  2. Hand-write or research a specific detail for the opening (10% of the work, 80% of the impact)
  3. Review and tighten the ask at the end (30% of the work)

This is faster than writing from scratch, but better than pure AI copy. You're using the tool for what it's actually good at.

The hand-written part is what makes it land. That's why personalization matters - it's the thing AI can't do without your input.

Testing Your Prompts

You should A/B test your prompts the same way you'd test email copy. If you're generating 100 emails a week with a prompt, run that prompt against 2-3 variations and track which angle / opening style gets more opens and replies.

Track what's working by angle:

Then build more prompts around the angles that are winning. Your prompt engineering improves when you have data telling you what actually works with your audience.

The Gap Between Knowing This and Running It at Scale

You now have a framework for writing better cold email prompts. And honestly, it'll work. You can take this and generate solid emails today.

But there's a gap between "I can write a good prompt" and "I have 200 personalized emails with research-backed angles going out this week." The gap is: research, list management, sending infrastructure, reply handling, and optimization. Prompt engineering is 20% of cold email. The other 80% is execution - and that scales or breaks based on your systems, not your copy skill.

If you want to hire someone to own the whole thing - the research, the angles, the prompts, the sends, the replies - that's what we do. We build the infrastructure and operations so you're signing 5-20+ clients per month using cold email only.

Related Guides