Ask ChatGPT for "a cold email to a marketing agency" and you get the same email everyone else gets. The model knows nothing about you, your offer or the reader, so it guesses. These prompts start by giving it the facts.
Step 1: write a fact sheet once
This page does most of the work. Fill it in once, keep it in a document, and paste it at the start of every cold email chat.
What I sell: [one sentence, with the price or price range].
Who it is for: [the type of business, its size, and the problem it has].
Proof: [one or two real results, with numbers, and the client's permission to mention them].
How I write: [three rules, such as under 80 words, no exclamation marks, one question at the end].
Never say: [phrases to avoid, such as "I hope this finds you well", "quick question", "synergy"].
My ask: [the small next step, such as a reply, not a meeting].
In ChatGPT, you can save this in a project. OpenAI's help article on projects explains how to add files and instructions that apply to every chat in that project. For more on what to include, see how to give AI context about your business.
Step 2: collect one real detail per reader
A cold email works when it is clearly about one person. Before you prompt, write down one true detail about each reader. A recent post, a job opening, a page on their site, a product launch. Find it yourself and check it.
The prompts
Paste your fact sheet first. Then use one of these.
Using my fact sheet, write a cold email to [name], [role] at [company]. The one detail I found: [detail, with the source]. Open with that detail in one sentence. Then link it to the problem I solve. End with my ask. Under 80 words. Follow my "never say" list.
Here is a detail about [name]: [detail]. Write three different first lines for a cold email that mention it. Each under 20 words. No compliments, no "I noticed".
Here is my draft: [paste]. Cut it to under 70 words. Keep the detail about the reader, the proof and the ask. Remove everything else. Do not add anything new.
Here is a draft: [paste]. Check it against my "how I write" and "never say" lists. List each rule it breaks and quote the line. Then suggest a fix for each. Do not rewrite the whole email.
Write five subject lines for this email: [paste]. Two to five words each, lower case, plain. Nothing that sounds like an ad.
The email below got no reply after three working days: [paste]. Write a follow-up under 50 words that adds one new detail from my fact sheet and repeats the ask in one line. Do not start with "just checking in".
[Name] replied: [paste reply]. Write a short, friendly answer that thanks them, asks if I can check back in [month], and asks nothing else.
Here are three emails I wrote myself: [paste]. Note how I open, how long my sentences are and words I use. Then rewrite this draft so it sounds like me: [paste draft].
Why these work
Each prompt does three things. It gives the model real facts. It sets a limit, such as words or rules. And it asks for one small job at a time. A prompt that asks for research, writing and a whole sequence at once leaves the model more gaps to fill with guesses.
The word limits come from the same idea as how long a cold email should be. Prompt 4 is the one to keep using. It turns your rules into a check, which is the method in why cold emails sound like a robot.
Check every draft before you send
- Facts about the reader. If ChatGPT added a detail you did not give it, delete it.
- Proof. Make sure every number is real and you are allowed to use it.
- Tone. Read it out loud. If you would not say it, change it.
- Privacy. Do not paste private customer data into a chat tool.
When you want more than one email
For a warm lead, the job is different. See how to write a sales email. For prompts across the rest of your business, see ChatGPT prompts for small business owners. If drafts still sound like everyone else, how to make AI sound like you goes deeper on samples and rules.
Nive is being built for this problem: you write your business facts and your voice down once, and every draft starts from them, with a check against your rules. You still review and send each email. Join the waitlist to get the demo first.