Ispent three years sending cold emails that went absolutely nowhere. I’m talking about carefully crafted, spell-checked, totally ignored messages to potential clients, editors, investors — you name it. I’d hit send, watch the silence roll in, and wonder what I was doing wrong.
Then one evening, mostly out of desperation, I pasted one of my drafts into Claude and just asked: “Why wouldn’t someone reply to this?”
The feedback was brutal in the most useful way. My email started with “I hope this email finds you well” (already dead), my subject line said “Collaboration Opportunity” (absolutely meaningless), and I had buried the actual ask in paragraph four like I was ashamed of it. The AI wasn’t cruel about it — but it was honest in a way that no colleague had ever been.
That was the turning point. Not because AI magically writes better emails than me, but because I learned how to work with it to stop writing emails that deserved to be ignored.
The Real Problem With Most Emails People Write
Before we even get into the AI stuff, it’s worth being honest about why most emails don’t get replies. It’s rarely bad grammar or typos. The real culprits are simpler and more embarrassing.
The email is clearly about you, not them. It opens with your background, your needs, your timeline. The recipient reads the first line and thinks, “okay, what does this person want from me?” before they’ve felt any reason to care.
The subject line lies about what’s inside. “Quick question” when it’s a 400-word request. “Following up” as if that means anything. “Touching base” — don’t get me started.
There’s no single, clear ask. You’ve included three possible options, two questions, and a vague offer to “chat sometime.” The reader doesn’t know what to do, so they do nothing.
Most emails don’t fail because of what’s written. They fail because of what the writer was thinking about while writing — themselves.
AI doesn’t fix these problems automatically. But when you use it right, it forces you to confront them.
How I Actually Use AI to Write Emails — Step by Step
My process has evolved over the past year and a half using a mix of tools. Mostly Claude for drafting and revising, occasionally ChatGPT for quick rewrites. Here’s what actually works.
- 1Start with a brain dump, not a draftOpen the AI and just explain the situation in plain language — who you’re emailing, why, what you want them to do, and any relevant context. Don’t try to make it sound good yet. Something like: “I want to email this editor at a tech magazine. She doesn’t know me. I wrote a piece on AI in healthcare and want to pitch it. I need her to reply yes or no within a week.”
- 2Ask for a first draft with a specific toneDon’t just say “write me an email.” Tell the AI the relationship (stranger, colleague, boss), the tone you need (direct, warm, casual, formal), and how long it should be. I usually say: “Write a short, direct cold email. No fluff. Under 150 words. Make the ask clear in the last sentence.”
- 3Interrogate the draft before you touch itAsk the AI to critique its own output. “What’s the weakest part of this email?” or “What would make this person not reply?” This sounds weird, but it works. The AI will often flag things like a generic opening or a buried ask — sometimes things you wouldn’t catch yourself because you’re too close to it.
- 4Rewrite the subject line four or five timesAsk for five subject line options with different angles — curiosity, specificity, direct ask, shared interest. Then pick the one that sounds least like a template. My response rates improved noticeably once I stopped writing my own subject lines.
- 5Add your own voice before sendingThis is critical. Read the draft out loud. If it doesn’t sound like you, change it. Find one specific, genuine detail to add — a real reference to their work, a specific reason you’re reaching out now. This is what makes it actually land.
What a Bad AI Email Looks Like vs. a Good One
Let me show you a real before-and-after from an email I sent last year, pitching a freelance article to a newsletter editor:❌ Before — typical AI output, sent as-is
SubjectCollaboration Opportunity — Freelance Article Pitch
Tosarah@techdigest.com
I hope this email finds you well. My name is Reza Malik and I’m a freelance writer with experience covering technology and innovation. I recently came across your newsletter and was very impressed by the quality of your content.
I am reaching out because I believe I have an article that would be a good fit for your readership. I have been writing for several years and have covered topics across AI, sustainability, and more.
Please let me know if you’d be interested in discussing this further. I look forward to hearing from you.
Classic failure. Zero specificity, zero value for the editor, buried ask — and that opening line is the email equivalent of clearing your throat for two minutes before saying anything.✓ After — refined with AI + personal edit
Subject1,200-word piece on why NHS trusts are quietly shelving their AI pilots
Tosarah@techdigest.com
Your piece last month on enterprise AI adoption gaps was exactly right — and I’ve been reporting on a related story your readers would probably want to hear.
I’ve spoken with four NHS trust IT leads over the past six weeks. They’ve all quietly paused or cancelled AI rollouts — not because the tech failed, but because procurement, liability, and staff training timelines made it untenable. Nobody’s writing about this yet.
I have a 1,200-word piece ready to share. Would you like to read it?
The second one got a reply within four hours. Not because AI wrote it better — but because the AI helped me figure out what actually mattered, and then I added the specific details that made it real.
💡 The Key Insight
AI is good at structure and clarity. You are good at specificity and authentic detail. The magic is in combining both — not replacing yourself with the AI output.
The Tools Worth Knowing About
There are a few different tools in this space and they’re genuinely different in how useful they are for email writing:
- Claude (Anthropic)
- ChatGPT (OpenAI)
- Gemini (Google)
- LinkedIn AI Drafts
- Gmail Smart Compose
I use Claude for anything that needs real nuance — a difficult professional email, something sensitive, a pitch where tone really matters. It’s notably good at understanding context and giving you options rather than one definitive answer.
ChatGPT is faster for quick rewrites and I use it when I just need a different angle on something I’ve already drafted. Less back-and-forth needed for straightforward tasks.
Gmail’s Smart Compose is fine for short replies but I wouldn’t rely on it for anything where the stakes matter. It autocompletes sentences, which can subtly shift your meaning in ways you don’t notice.
The integrated LinkedIn AI drafting tools are honestly pretty weak. They tend to produce extremely corporate, hedge-everything language that sounds like nobody in particular wrote it. Fine as a starting point, but treat it like rough scaffolding.
Mistakes I Made (So You Don’t Have To)
- ✕Sending AI drafts without reading them aloud first. I once sent an email with the phrase “I am excited to explore synergies” to a journalist. She replied with a screenshot of that sentence and just wrote “really?” I deserved that.
- ✕Using AI to over-explain. My first instinct was to use AI to make emails longer and more thorough. The opposite is almost always right. Shorter, clearer, one-ask emails perform better every single time. If AI gives you 200 words, ask it to cut it to 120.
- ✕Asking AI to “make this more professional.” This turns every email into beige corporate mush. Try “make this more direct and confident” or “remove anything that sounds like hedging.” Way better results.
- ✕Using it for emotional emails. I once tried to use AI to write a difficult email to a client about a missed deadline. The output was technically fine but had no warmth. Some emails need human discomfort in them — that’s what shows the person on the other end that you actually care.
- ✕Skipping context in the prompt. If you just say “write a follow-up email,” you’ll get garbage. The more context you give — the history, the relationship, what happened last time, what you need now — the better the output. Treat the AI like a smart assistant who wasn’t in the room when any of this happened.
Prompts That Actually Work
I’ve tested a lot of prompts over time. These are the ones I come back to regularly:
❌ Vague Prompts
“Write me a professional email to my client about the project.”
✓ Specific Prompts
“Write a 100-word email to a client who hasn’t responded in 10 days. Friendly but creates a clear deadline. One question at the end.”
❌ Style-Free Asks
“Make this email better.”
✓ Directed Rewrites
“Rewrite this to sound like someone who doesn’t need this deal. Confident, not desperate. Remove all apologies.”
❌ Open-Ended Asks
“Help me write a cold email.”
✓ Context-Rich Asks
“I’m a developer emailing a startup CTO I’ve never met. Offer a 2-hour free audit of their checkout flow. Under 120 words. Focus on what they get, not who I am.”
⚠ Watch Out For This
When you give AI very specific instructions, it sometimes sticks so rigidly to the constraint that it loses the natural feel. Always do a final pass to loosen anything that feels mechanical. The goal is to sound like a real person who happens to be clear and direct — not like a brief was filed.
When AI Email Help Genuinely Shines
There are specific situations where AI-assisted drafting has saved me real time and improved real outcomes:
Cold outreach — this is where the biggest gains are. When there’s no relationship to fall back on, structure and clarity do most of the work. AI is excellent at making sure your value proposition is stated clearly in the first two sentences.
Tricky conversations after a gap — re-engaging someone six months after the last message is awkward. AI helps you strike the right balance of acknowledging the gap without making it weird.
Non-native English speakers — I’ve watched colleagues who learned English as their second or third language go from anxious about professional email tone to genuinely confident. The AI handles the idioms and formality norms while they handle the substance.
Reducing emotional charge — sometimes I write a first draft of a difficult email when I’m frustrated. I’ll ask Claude to “rewrite this but remove any language that sounds defensive.” It’s like having a level-headed editor review your words before you do something you’ll regret.
📬 Practical Tip
For recurring email types — weekly client updates, follow-ups, pitch emails — build a template prompt that includes your usual context. Paste and adjust. You’ll cut your drafting time to almost nothing for those.