Most ChatGPT cold email prompts floating around are one line long: “Write me a cold email to a marketing director.” You know what comes back — “I hope this email finds you well,” a paragraph about how you “came across their company,” a close asking for 15 minutes. Grammatically perfect, dead on arrival.
The problem isn’t the model. A one-line prompt gives it nothing specific to work with, so it falls back on the average of every sales email ever written. The prompts below force you to hand over the ICP, the offer, the proof and a hard word cap — and they ban the phrases that make an email read like a template.
Fifteen prompts, grouped by job. Fill in the brackets and stop editing slop.
The prompt-engineering rules that matter for cold email
Five inputs separate a usable draft from a throwaway. Every prompt below includes them; understanding why lets you adapt them to your own market.
- Give it the ICP, not the job title. “VP of Marketing” produces generic output. “VP of Marketing at a 40–150 person B2B SaaS company who owns pipeline number but has no dedicated demand-gen hire” produces something a human recognizes.
- Give it the offer in plain language. Not your website tagline — the actual mechanism. What changes for them, and how.
- Give it proof it’s allowed to use. Named customers, a public case study, a specific integration. If you don’t supply proof, the model invents it — and inventing proof in a cold email is how you get reported.
- Cap the length in words. “Under 80 words” is enforceable. “Concise” is not.
- Ban the clichés explicitly. Add this to any prompt: Do not use: “I hope this finds you well,” “I came across,” “quick question,” “circling back,” “synergy,” “revolutionize,” “game-changer,” “just following up.” No exclamation marks.
One more: ask for three variants, not one. The first draft is the model’s most predictable answer.
First-touch prompts (1–4)
1. The problem-first opener. You are writing a cold email to [ICP: role, company size, industry, and the one thing they’re accountable for]. Their problem is [problem, in their words not yours]. We solve it by [mechanism]. Proof I can cite: [customer / case study / number]. Write three variants, each under 90 words. Open with the problem, not with me. One clear ask at the end: [ask]. Apply the banned-phrase list. Plain sentences.
2. The observation opener. Public information about the prospect: [paste job posting, funding announcement, product page or LinkedIn post]. What we do: [one sentence]. Write a cold email under 75 words opening with one specific, verifiable observation from that information, connected to the problem we solve. Do not flatter. If the information does not clearly connect to our offer, say so instead of forcing it.
3. The “one metric” email. Write a cold email to [ICP] whose primary metric is [metric]. In under 70 words: name the metric, name one reason it stalls for companies like theirs, and offer [asset or idea] that addresses it. No product pitch, no calendar link. End with a question they can answer in one word.
4. The switch email (competitive). The prospect currently uses [competitor or manual process]. Write a cold email under 85 words that treats that choice as reasonable, names one limitation that shows up at their stage [stage], and states what we do differently in one sentence. Do not disparage the competitor and do not claim we’re better overall — claim we’re different in one specific way. Three variants, three different limitations.
Subject line prompts (5–6)
5. The subject-line batch. Here is the email body: [paste]. Generate 15 subject lines under 45 characters, in five groups of three: (a) plain and literal, (b) a specific noun from their world, (c) a question, (d) three words or fewer, (e) the most specific detail in the body. No clickbait, no “Re:” fakery, no ALL CAPS, no emoji. Then name the three you’d send, one line each.
6. The subject-line critic. Here are 10 subject lines I wrote: [paste]. For each, say in one sentence what a busy [role] would assume the email is about before opening. Flag any that sound like a newsletter, a sales sequence or an automated tool. Rewrite the worst three.
Personalization from public information (7–8)
7. The research-to-line converter. Here is public information about a prospect: [paste bio, recent post, podcast summary, company about-page, or job description]. Extract the three most useful facts for a cold email — facts that suggest a problem, a priority or a recent change. For each, write one opening sentence I could use. Ignore school, hobbies and tenure. If nothing useful is there, tell me the file is thin instead of inventing an angle.
8. The pattern-across-accounts prompt. Ten target companies with a short note on each: [paste]. Group them into at most three segments by the problem they most likely share, not their industry. For each, write one first-touch email under 80 words that works for every company in it with only the name changed, and list the merge fields it needs.
Follow-up sequence prompts (9–11)
9. The four-step sequence. Here is my first email: [paste]. Write follow-ups 2, 3 and 4 for a sequence sent over [X] days. Each must add something new — a different angle, a resource, a shorter reframe. Never repeat the first email’s argument. Never say “just following up,” “bumping this” or “in case you missed it.” Email 2 under 60 words, 3 under 50, 4 under 35. Email 4 is a clean close that makes it easy to say no.
10. The angle-rotation follow-up. My offer is [offer]. Write four follow-ups, each on a different angle: (1) a specific customer outcome, (2) an objection pre-empted, (3) useful content with no ask, (4) a one-line breakup. Each under 60 words, each standing alone if it’s the only one they read.
11. The breakup that isn’t rude. Write three versions of a final follow-up to [ICP] after [N] unanswered emails, under 35 words each. No guilt, no “I’ll assume you’re not interested,” no fake deadline. Leave a door open without asking a question they have to answer.
Reply-handling prompts (12–13)
12. The objection responder. Here is the prospect’s reply: [paste]. Here is context on our offer and pricing: [paste]. Classify it as: real objection / brush-off / timing / wrong person / genuine interest. Then write a response under 70 words matching that classification. If it’s a brush-off, don’t argue — ask one question that surfaces the real reason. If it’s the wrong person, ask for a referral in one sentence.
13. The interested-but-vague reply. The prospect replied: [paste]. They’re warm but non-committal. Write a response under 60 words proposing one concrete next step with a specific date and time, one alternative, and the single most relevant proof point from: [paste proof points]. Do not send a calendar link and a question in the same email — pick one.
List-level cleanup prompts (14–15)
14. The bulk rewrite. Ten emails from my sequence library: [paste]. Rewrite each to under 80 words without losing the core argument. Then produce a table: original word count, new word count, what you cut, one risk of the shorter version. Flag any email whose argument doesn’t survive the cut — those need rewriting from scratch, not trimming.
15. The cliché audit. Here is my entire sequence: [paste]. Table every phrase repeated across emails, every sentence that could appear in any B2B cold email regardless of product, and every claim made without proof. Rewrite only the flagged sentences; leave everything else untouched.
Beyond prompts: ChatGPT as an agent, not a copywriter
Everything above treats ChatGPT as a writing tool — you prompt, it drafts, you paste into a sending platform, upload a list, build the sequence, launch. The writing was maybe 20% of the work.
That gap is what changed. Through MCP — the open standard that lets AI assistants connect to outside tools — ChatGPT can act on systems rather than describe what to type into them. Instead of handing you fifteen drafts, it can build the campaign, load the prospects and hand it to a scheduler, in the same conversation where you wrote the copy.
Which raises the obvious question: what stops an agent doing something reckless with your sending accounts?
Where to actually launch what you wrote
This is where we’d point you at WarmySender — a self-service outreach platform that connects over MCP to ChatGPT, and to Claude, Cursor, Codex, OpenClaw, Hermes Agent, or any agent that speaks the protocol.
The agent can create, launch, pause and resume campaigns across cold email, LinkedIn and Instagram; create and enroll prospects; verify email addresses; configure warmup and read the stats back to you. You can go from “write me a four-step sequence for agency owners” to a built and scheduled campaign without leaving the chat.
What it deliberately cannot do matters just as much:
- The agent never sends anything directly. Creating or launching a campaign writes it and hands it to the scheduler — the agent doesn’t push messages out itself.
- The agent can never raise a limit. Not daily caps, not the warmup ramp, nothing.
- The scheduler paces every action — every email, every LinkedIn action, every Instagram action — inside safe caps and the gradual ramp, whether a human or an agent triggered it. Email verification runs on the same allowance.
- Connecting and disconnecting accounts stays in the app. That’s the one thing that isn’t an agent tool.
Your agent can run the outreach — build it, launch it, manage it, tune warmup — while the scheduler keeps every account inside safe limits no matter who’s driving. Which is what you want when the thing writing your emails is also the thing pressing go.
FAQ
Will prospects be able to tell my cold email was written by ChatGPT?
They’ll tell if you send a first draft off a one-line prompt — the rhythm and stock phrases are recognizable now. They won’t tell if the email has a specific observation about their company, a real proof point and no filler. The tell isn’t AI authorship, it’s genericness. Prompt 15 exists purely to catch what slips through.
Should I personalize every email individually?
Only where the research is genuinely there. Prompt 7 is built to tell you when a prospect’s public information is thin rather than manufacture an angle — a fake-personal opener reads worse than an honest segment-level one. For most lists, prompt 8 is better economics: two or three tight segments, each with copy that’s true for everyone in it.
What’s the right length for a cold email?
Short enough to read on a phone without scrolling. Every prompt here caps at 90 words or fewer, and follow-ups shorten as the sequence goes on. If your argument needs 200 words, that usually means the offer isn’t sharp yet — not that the email should be longer.
Start with the prompts, then close the gap
Take three prompts — one first-touch, one subject-line, one follow-up — and run them against a real segment today. Fill in every bracket properly; the output is only as specific as what you hand over. Keep the variant that sounds like a person, delete the other two.
Then decide whether you want to keep copy-pasting drafts into a sending tool, or connect your assistant to something it can actually operate.