A reusable cold-email prompt structure for ChatGPT, Claude, Gemini and Copilot, plus the four pre-send checks that keep your sequence out of spam.
Last revised May 11, 2026
Cold email is in a worse place than it was a year ago. Deliverability has tightened (Google and Yahoo's 2024 bulk-sender requirements added SPF/DKIM/DMARC enforcement for any sender doing more than 5,000 emails a day), reply rates are falling, and AI-generated boilerplate has trained buyers to spot template emails inside the opening line.
But the channel still works, if you do three things most senders don't bother with.
1. Warm the domain like an adult
You cannot send 200 cold emails a day from a brand-new domain. Inbox providers see it, mark you as spam, and you spend three months trying to claw your way back. The right pattern:
- Buy a secondary domain (not your main one) for cold outreach. If you burn it, your real domain is fine.
- Run it through a warmup tool (Mailreach or Lemwarm are the two I have seen used most) for at least 14 days before sending anything cold.
- Verify SPF, DKIM, and DMARC are passing, MXToolbox gives you a free check.
This is unsexy work. It is also the difference between landing in primary and landing in spam, which is the difference between a viable channel and a wasted quarter.
2. Write like a human or do not write at all
The single biggest reason cold emails get ignored: they read like they were written by a model. Long opening line, generic flattery, vague CTA, no specifics.
The fix is uncomfortable: write fewer emails, but make each one obviously human. The structure I keep returning to:
[Specific, current observation about the prospect's business. One sentence]. [One sentence connecting it to what you do, without selling]. [One question that's easy to answer.]
That's it. Three sentences. No "I hope this finds you well." No "I came across your profile and was impressed." Buyers can spot those phrases instantly and they trigger an immediate skim-and-delete.
Lavender's 2025 cold email benchmarks show emails under 75 words consistently outperform longer emails on reply rate.
3. Make AI work with you, not for you
AI is bad at writing finished cold emails. It is great at: researching prospects, identifying signals, drafting variants of a single line so you can pick the strongest, and removing template language from drafts you wrote tired.
A workflow that produces emails worth sending:
- Research (model-assisted): pull 3 specific things about the prospect from public sources, a recent post, a hire, a product change. Rank by which is most likely to be a current concern.
- Draft the opening line yourself based on the strongest signal. This is the one sentence that has to feel human.
- Hand the draft to the model and say: "Rewrite this in 60 words or less. Remove any line that could appear in a template email. Keep the opening sentence exactly as-is." The model is good at compression; it is bad at originality.
- Read it out loud. If you would not say it that way on a phone, edit it.
What about volume?
Most teams chase volume because it feels like progress. A team sending 500 personalized emails a week often books more meetings than a team sending 5,000 templated ones. Gong's outbound data has documented this consistently for years.
Personalization isn't a luxury. With current deliverability and inbox-fatigue dynamics, it is the only path that scales.
The follow-up problem
Most cold emails get a reply on follow-up 2 or 3, not the first email. Most teams give up at follow-up 1. The follow-ups that work add new information each time, a relevant data point, a comparison, a question better than the one in the first email. Never send "bumping this to the top of your inbox." It tells the prospect you have nothing new to say.
A note on lists
If your list is bad, none of this matters. Bought lists from data brokers have email-deliverability rates 30-50% lower than lists you build through research. Single-sourced lists from one provider tend to be older and noisier than blended lists across two or three sources. The first hour of every cold-email project should be list quality, not copywriting.
For prompt scaffolds I use across this whole workflow, research, drafting, follow-up sequencing. Our Sales AI Prompt Pack collects the ones I have iterated on most.