How to write a cold email with AI (that a person would answer)
Ask a chatbot for "a cold email to a marketing director" and you get the email everyone deletes: a warm-up line, three features, and a request for 30 minutes. The model is not the problem. The input is. A good cold email needs facts only you have — who the person is, why you picked them, and one piece of proof.
This guide walks through writing one with AI, using free skills that put that structure into the model for you. Nothing here promises a reply rate. It gives you a cleaner email and a process you can repeat.
Step 1: Collect the four facts before you open the chat
The cold-email skill (MIT licence, by Corey Haines) asks these before it writes, and they are worth having ready:
- Who you are writing to — their role, their company, and why them in particular.
- What you want — a reply, a meeting, an intro, a demo. Pick one.
- What problem you solve for people like them — in their words, not yours.
- Your proof — one result, one customer name, or one credible fact.
If you cannot fill in point 4, fix that first. An email with no proof is a request for a favour.
Step 2: Find a real reason to write now
The signal-cold-email skill builds the email around a dated signal: a funding round, a new hire, a post the person wrote. Its rule is strict — the signal must be real, recent, and come with a source link. Mentioning the company name is not personalisation; referencing something they did last month is.
Its structure is short enough to remember:
- Hook tied to the signal.
- One sentence of value for someone in their exact situation.
- One proof point. One, not five.
- One low-friction ask — a single question, or a short call. Never two asks.
The file caps the email at under 150 words, plain text.
Step 3: Ask for the draft, then cut
Install one of the two skills (see how to install a skill), paste your four facts and the signal, and ask for a cold email. Then read it aloud. The cold-email skill's own quality check is a good filter: would you reply to this if you received it? Does every sentence serve the reader, not the sender?
Cut anything on its "avoid" list:
- "I hope this email finds you well" or "My name is X and I work at Y"
- Buzzwords like "synergy" and "circle back"
- A list of features instead of one proof point
- HTML, images, or more than one link
- Fake "Re:" or "Fwd:" subject lines
- The same template with only the first name swapped
If the draft still reads like a machine wrote it, run it through content-humanizer, which flags filler words and hedging and rewrites for rhythm.
Step 4: Plan the follow-ups before you send
Most replies are not to the first email, so plan the next ones now. The cold-email skill says each follow-up should add something new — a different angle, fresh proof, a useful resource — and that "just checking in" gives the reader no reason to answer. It suggests 3–5 emails in total with growing gaps, each one readable on its own.
For a longer automated sequence (onboarding, nurture, re-engagement), email-sequence-2 designs the whole flow.
If you only need one quick email
Not every email is a campaign. The write-an-email prompt is one line with blanks for tone, recipient and topic. Paste it into any chat app, fill the blanks, and add the four facts from step 1 underneath.
If you are sending at volume
Sending hundreds of cold emails is a different job: domains, inbox warm-up, list building, a sending tool. cold-email-starter-kit walks a beginner through that whole setup. Read it before you buy anything — it involves paid services (domains and a sending platform), and the choices are yours to make.
The short version
Give the AI facts, not a vibe. Ask for one hook, one value line, one proof, one ask. Cut the cliches. Plan the follow-ups. Every skill above is free to read in full on its page before you use it.