Paste the prompt below into ChatGPT or Claude, fill in three blanks, and you'll have a usable draft in under a minute. What decides whether the draft is any good is what you put in those blanks. Give the model a real fact about the person and a real number for your offer and it writes something you can send. Give it "small business owners" and "great service" and you get the email everyone already deletes.
I'll show the prompt, the three inputs, what a typical bad output looks like, and the edits that fix it. Then the point at which this whole approach starts costing more time than it saves.
What the model can and can't do for you
A chat model is good at wording. It can tighten a rambling paragraph, offer five ways to say the same ask, and keep a tone steady. It is bad at the three things that make a cold email work: knowing who to write to, knowing something true about that person, and knowing what you can actually deliver. It has none of that unless you hand it over. Ask for "a cold email to a restaurant owner" and it will invent a reason to write, usually a compliment on "your impressive reputation."
The prompt to paste
You are writing one cold email from [my name], who [what I do, in one sentence], to [their first name], [their job title] at [their business].
What I know about them (all of this is true): [one specific fact].
What I'm offering: [the offer in one sentence, with a price or a range].
Write the email in under 80 words. Rules: the first sentence is about them, not me. End with one question that can be answered yes or no. No compliments. Do not use "I hope this finds you well," "I wanted to reach out," "I came across," or words like leading, innovative or solution. Use only the facts I gave you. Do not invent anything about them or about me. Then give me three subject lines of five words or fewer, in lowercase.
When to use it: the first email to one person you've looked up. Why it works: the rules remove the habits that make AI email obvious, and "use only the facts I gave you" is the line that stops invented details. Variation: add "Write it so a reply of 'not now' is easy to send" if your list is mostly people who don't have an urgent need.
The three facts you have to supply
Here are the blanks for Owen, who runs a one-truck pest control route in Boise and sells quarterly contracts to restaurants.
- Who. First name, job title, business. "Elena, owner, Cedar and Rye." The title matters because it shapes the ask. An owner decides on the spot. A general manager has to check with someone.
- One true thing about them. Something you can point to in two minutes: their site says they opened a second location this spring, they posted a hiring notice for a line cook, their last health inspection is public. "Cedar and Rye opened a second location on State Street this spring."
- The offer with a number. "Quarterly pest control, $145 a visit, with a written log for the health inspector." A price or range tells the reader what kind of conversation this is.
Two minutes of looking per person is the cost. It is also the thing no prompt can replace. If you can't find a true fact, don't ask the model to make one up. Send that person a plainer email or skip them.
Fixing the generic opener it writes
Even with good inputs, models drift back to a stock opener. A typical first draft for Owen:
Hi Elena, I hope this finds you well. I came across Cedar and Rye and was impressed by your growth. At Owen's Pest Services we help restaurants like yours stay compliant and pest-free with our reliable quarterly service. I'd love to set up a quick call to discuss how we can support your new location. Would you be open to that?
It has the stock greeting, a compliment, "help restaurants like yours," "I'd love to," and an ask that is a meeting, not a question. Run these five fixes in order:
- Delete the first sentence. It's always filler.
- Replace the compliment with the fact: "Cedar and Rye opened on State Street this spring."
- Swap "we help restaurants like yours" for what you do and what it costs.
- Change the ask to a small yes or no.
- Remove any claim you can't back up. "Reliable" and "compliant" mean nothing until you show the log.
After the edits:
Hi Elena, Cedar and Rye opened on State Street this spring, which means a second kitchen to keep inspection-ready. I run quarterly pest control for Boise restaurants: $145 a visit, and you get a written log you can hand the inspector. Would it be useful if I sent a sample of the log?
Fifty-three words, one question, no flattery. If your outputs still read stiff, our guide on writing a cold email that gets replies goes through the structure a draft should follow.
Three more prompts worth keeping
Rewrite my draft. "Here is my draft. Cut it to under 80 words, keep every fact, remove anything that sounds like a sales pitch, and tell me which sentence you removed and why." Use it when you've written something yourself and it's too long. Variation: ask for two versions, one blunter.
The follow-up. "Here is my first email and they didn't reply. Write a 3-sentence follow-up that adds one new piece of information (here it is: [new fact]) and does not say 'just following up.'" Use it on day four. It works because you supply the new fact. Variation: ask for a version that ends with 'should I close the loop on this?' for the last email.
Subject lines. "Give me 10 subject lines for this email, each under 6 words, lowercase, none that sound like an ad." Pick one, then change a word so it matches how you'd actually write.
Why one-by-one pasting stops working around twenty emails
Per email, the work is: find the person, find the fact, fill the prompt, edit the output, paste into Gmail, send. At five minutes each, 20 emails is an hour and forty minutes. A three-email sequence to those 20 people is 60 emails and the follow-ups must be written, sent on the right days, and stopped for anyone who replied. Nobody does that by hand for long. The prompt keeps working. The process around it is what fails: missed follow-ups, someone emailed twice, a reply sitting unread for two days.
Letting a worker find the buyer, write and send
The step the chat model can't do is the first one, finding who to write to, and the last, sending and following up. Rocketship does those. You describe the buyer, it finds them by name, writes the emails from the facts it has on each person, sends from your own Gmail, follows up, and books the meeting. If you already work in Claude, the Rocketship connector lets you do this in the chat: ask it to find 25 restaurant owners in Boise, write a three-email sequence, and show you the emails before anything goes out. You can also let the worker run without the preview once you trust the output. The outreach tool for people who do not do sales explains what that setup looks like.
Read it once more before anything goes out
Whatever writes it, check four things. Every fact about the recipient is true, because a wrong detail is worse than none. The email names a price or a concrete next step. There is exactly one question. And you'd be comfortable if the recipient forwarded it to a friend. For a longer look at spotting mistakes in a draft, see seven cold email mistakes with rewrites, and for the day-four message, better openers than "just following up" gives you the wording.
A chat model makes the writing easier, but the writing was never the slow part. Rocketship finds the buyers, emails them from your own Gmail, follows up, and books the meeting on its own. Plans that get you customers start at $24.99 a month (Launch), and every monthly plan starts with a 7-day free trial, card required.