Most cold emails fail because they sound like cold emails. ChatGPT can write a hundred of them in two minutes — and ninety-eight of them will be ignored. The problem isnt AI. Its that we use AI to crank out volume when what cold email needs is precision.
This is the exact prompting workflow I use to get a 30%+ reply rate on cold outreach. It takes about ten minutes per email and doesnt sound generated.
Why most ChatGPT cold emails fail
Three classic mistakes:
- The write me a cold email prompt. Generic in, generic out. The recipient knows what a templated AI email looks like by now.
- Too many values and synergies. AI defaults to corporate-speak unless you explicitly tell it not to.
- No real research about the recipient. A cold email that doesnt reference something specific to that person gets deleted in three seconds.
The fix isnt a better prompt. Its a better workflow.
The four-step prompting workflow
Step 1: Research the recipient (5 minutes)
Before opening ChatGPT, find:
- Their LinkedIn — what they posted in the last 30 days, recent role changes
- Their company website — recent product launches, blog posts, careers page
- One specific pain point you can credibly help with
Copy the most useful 3-5 bullet points into a notes file.
This is the part you cannot skip. Every successful cold email pivots on something specific. AI is bad at this part — it can only work with what you give it.
Step 2: Give ChatGPT context, not commands
Bad prompt: Write a cold email to Sarah, head of marketing at Acme.
Good prompt:
Im writing a cold email to Sarah, head of marketing at Acme.
Context I gathered: – Sarah posted on LinkedIn last week about how their content team cant keep up with their podcast publishing schedule. – Acme just launched a new B2B podcast called The Builder Show. – Their last 3 podcast episodes have an average of 200 LinkedIn impressions, which is well below industry standard for their company size.
I run a small podcast clipping service. I help podcasters turn each episode into 5-10 short-form clips for LinkedIn and X. My typical clients see a 4-6x lift in social impressions.
Write a 70-word cold email that: – Opens with a specific observation about Sarahs LinkedIn post (not flattering — just acknowledging the pattern) – Pivots to a one-line offer – Ends with a low-friction ask (not hop on a call) – Sounds like a real person, not a template – Avoids the words value, leverage, circle back, sync, and thoughts?
Thats a useful prompt because ChatGPT now has something specific to work with.
Step 3: Generate three versions, pick parts of each
Dont accept the first output. Ask for three versions, each with a different angle:
Give me three versions: 1. Direct — leads with the offer 2. Story-led — opens with a quick observation about a similar company 3. Curiosity-led — opens with a question
Then take the best opening from one, the best middle from another, and your favorite close. ChatGPT is great at producing parts; youre great at editing them together.
Step 4: Manually rewrite 30% of it
This is what separates the 30% reply rate from the 3% reply rate. After ChatGPT gives you a draft:
- Cut every sentence that doesnt earn its place. If a sentence could be in any other cold email ever sent, delete it.
- Add one specific personal detail ChatGPT couldnt have known. Reference a specific blog post, a specific job change, a specific tweet.
- Read it out loud. If you wouldnt actually say it to a stranger at a coffee shop, rewrite it.
The final email should feel like ChatGPT helped you write it, not like ChatGPT wrote it.
Templates that actually work
Below are three prompt templates I use weekly. Adapt them to your own service or product.
Template 1: The I noticed opener
I noticed [specific thing the person or company recently did]. I help [type of business] with [specific problem this thing implies]. My typical [client/customer] sees [specific result] within [time frame].
Worth a 5-minute call next week, or would a Loom walkthrough be easier?
Template 2: The I built this for someone like you opener
[Recent observation about their company/role.] I just finished [specific project/case study] for [similar company]. The exact same approach would apply to [their company].
Want me to send the case study? No call required.
Template 3: The wrong reason to reach out opener
Quick honest one — Im reaching out because I think your competitor [Competitor Name] is doing [specific thing] better than you. I help [type of business] close that gap.
Happy to share what theyre doing, no pitch attached. Reply send it and Ill get the breakdown to you tomorrow.
The last one has the highest reply rate of any opener Ive tested. People are curious about their competitors.
The biggest mistake: sending too many
The temptation when you have AI is to send a hundred personalized-but-not-really emails. Dont. Send ten emails that are deeply personalized and watch your reply rate hit 30%+. AI gets you the speed of a hundred emails with the quality of ten — but only if you actually edit each one.
If you find yourself sending the same email to ten people with the names swapped, youve slipped back into mass-email mode. Thats where AI cold email goes to die.
Quick reference: prompt template
Save this as a Notion or text file for reuse:
Im writing a cold email to [NAME] who is [ROLE] at [COMPANY].
Context I gathered: – [SPECIFIC LINKEDIN POST OR COMPANY ACTIVITY] – [SPECIFIC PRODUCT OR INITIATIVE] – [SPECIFIC PAIN POINT EVIDENCE]
I [ONE SENTENCE ABOUT WHAT YOU DO]. My typical [CLIENT] gets [SPECIFIC RESULT].
Write a [70-100 word] cold email that: – Opens with a specific observation (not flattering) – Pivots to a one-line offer – Ends with a low-friction ask – Sounds like a real person – Avoids: value, leverage, circle back, sync, thoughts
Give me three versions: direct, story-led, curiosity-led.
The honest truth about AI and cold email
ChatGPT doesnt write better emails than you. It writes faster drafts. The reply rate comes from research, specificity, and editing — none of which AI can do for you yet.
If youre getting low reply rates with AI cold emails, the answer is almost never use a better AI. Its do more research and edit harder.
That said, if youre already doing the research, AI cuts the writing time per email from 20 minutes down to 5. Multiply that across 50 emails a month and youve reclaimed a workday for the parts of the business AI cant help you with.
If you found this useful, subscribe below for the next post — Im publishing the exact research process I use to find prospects who actually need what Im selling.