A ChatGPT prompt earns a reply to a cold email when it carries the prospect's context into the draft: who they are, what changed at their company, what you actually do for people like them, and the one action you want next. Ask ChatGPT to simply write a cold email and you get the same template every other rep is sending. The difference is not the model. It is the prompt, and specifically the context you load into it before a single word is drafted.
Why most cold email prompts produce junk
The failure pattern is consistent. A rep types a one-line request, gets a five-paragraph email that opens with a compliment about the company's website, and concludes the tool is overhyped. What actually happened is the model was given nothing to work with. A cold email lives or dies on specifics — a trigger event, a relevant number, a sentence that could only have been written to this prospect — and specifics have to arrive in the prompt. The prompts below are structured so that the research goes in first and the drafting comes second.
The context block every cold email prompt needs
Before any drafting instruction, give the model a context block with six lines. Who you are selling to: role, company, industry. The trigger: why now — a funding round, a job post, a product launch, an exec quote. Your one-sentence value case for this persona, stated as an outcome, not a feature. Proof: one customer result you can name. The call to action: what a reply should agree to. Constraints: word count, tone, and what to avoid. Reps who keep this block saved and swap the variables per prospect get first drafts that need trimming, not rewriting. The prompts in the Outreach & Messaging category are built around exactly this structure.
A first-touch prompt you can copy
Here is a complete first-touch prompt, not a truncated teaser. Replace the bracketed variables and run it as-is.
"You are writing a cold email for a B2B sales rep. Prospect: [name], [title] at [company], [industry]. Trigger: [what changed and when]. What I sell, framed as an outcome for this persona: [one sentence]. Proof point: [customer + result]. Desired reply: [specific low-friction ask]. Write a cold email of no more than 90 words. Open with the trigger, not a greeting about their website. One idea only. No buzzwords, no 'I hope this finds you well', no exclamation marks. End with a one-line question the prospect can answer yes or no to. Then write a subject line of five words or fewer that references the trigger."
The constraints do the heavy lifting. Ninety words forces one idea. Banning the clichés removes the tells that mark an email as AI-written. The yes-or-no closing question measurably lowers the effort of replying.
Follow-up prompts that add something new
Most follow-ups say some version of "just bumping this", which gives the prospect a second chance to ignore you. A better follow-up carries new information. Prompt for it directly: give the model the original email plus one new input — a fresh piece of news, a different angle on the same problem, a short customer example — and instruct it to write a follow-up that stands alone, makes the new point in under 70 words, and does not reference being ignored. Run this three times with three different inputs and you have a sequence where every touch earns its place. For sequence design across email, LinkedIn, and phone, the Prospecting & Pipeline Creation category covers cadence structure end to end.
The rewrite prompt for emails you already drafted
Sometimes the draft exists and the problem is length or tone. A rewrite prompt is the fastest win in the whole workflow: paste your draft and instruct the model to cut it to 80 words while keeping the trigger, the proof point, and the ask, then to list what it removed and why. The removal list matters — it teaches you what you habitually over-explain, and after a few weeks your first drafts get shorter on their own.
Research before writing
Every prompt above improves when the trigger is real and current. Rather than drafting from memory, run an account research prompt first: company priorities, recent announcements, likely pain for your persona, one opening angle. The Account Research & Buyer Intelligence category exists for exactly this step, and its output slots directly into the context block. Research prompt, then drafting prompt, then rewrite prompt — that is the whole system, and each piece is copyable from the library. Start with a free prompt, and if the workflow sticks, see pricing.
Frequently asked questions
How do I get ChatGPT to write a cold email that doesn't sound generic?
Load the prompt with specifics before asking for a draft: the prospect's role and company, a trigger event that explains why you are writing now, one outcome-framed value sentence, one proof point, and a specific ask. Then constrain the output to 90 words or fewer with a yes-or-no closing question. Generic output is almost always the result of a prompt that contained no specifics to begin with.
What should I include in a ChatGPT cold email prompt?
Six things: the persona (role, company, industry), the trigger event, your value case stated as an outcome, a named proof point, the reply you want, and constraints on length and tone. Keep this as a saved context block and swap the variables for each prospect rather than writing prompts from scratch.
Can ChatGPT write a whole cold email sequence?
Yes, if you prompt each touch with a new input rather than asking for five follow-ups at once. Give the model the original email plus one fresh angle, news item, or customer example per follow-up, and instruct each message to stand alone in under 70 words. A sequence generated in one shot tends to repeat the first email five ways.
How long should a cold email written with ChatGPT be?
Ninety words or fewer for a first touch, and under 70 for follow-ups. Short constraints force the model to commit to one idea, which is what a cold read on a phone screen rewards. Set the limit in the prompt itself; models run long by default.