The 30-Minute Claude Cold Email System

The old way vs the new way

The old way: you export 200 names, write three good emails by hand, get distracted, and the rest rot in a spreadsheet. Or you dump the list into a generic AI tool, it writes "I came across your profile and was impressed," and you quietly burn your domain.

The new way: you set Claude up once with your product and rules, then run the list through three prompts. Out comes a scored, researched, drafted queue.

Most founders do not have a lead problem. They have a lead-to-email problem. You already have names. What you lack is a repeatable way to turn each name into a message that feels timely.

This will not make you a cold email expert in 30 minutes. It gives you a working system to turn a lead list into a reviewable outbound queue.

Why most AI cold emails fail

They fail for one reason: the workflow is incomplete. Claude is not the problem, it is being asked to write before the real work is done.

The email is only the final 10 percent. The other 90 percent is the part people skip:

• • identifying a relevant buyer signal
• • understanding why that signal matters
• • connecting the signal to a business pain
• • deciding if the lead is even worth contacting
• • writing a specific first-touch email
• • preparing the follow-up logic
• • reviewing the draft before it goes out

Skip that work and you get "I came across your profile and was impressed." Do that work and you get an email the prospect actually answers.

What counts as a signal

A signal is a recent, observable reason to reach out now. It is the single thing that makes a cold email feel timely instead of random. No signal, no good first line, which is exactly why this system qualifies and researches before it drafts.

• Company signals: something changed at the business (funding, hiring, a launch, expansion).
• Person signals: something the individual said or did (a post, a job change, a comment).

The strongest emails lead with one specific signal in the first line. Here are the ones worth looking for and where to find them.

Plug whatever signal sources you have into Prompt 2. The richer the signal, the better the draft, and a lead with no signal is a lead you hold, not force.

Setup: your Claude Project

Do this once. It is what makes every output consistent.

Create a Claude Project named Cold Email Operator - [Your Company]. Paste the block below into the Project instructions, then add your product one-pager, your ICP, and 2-3 of your best past emails. Do not add hype decks or claims you cannot back up.

You are my cold email operator. You help me turn a lead list into a
researched, reviewable outbound queue. I review and send everything
myself. You never send anything.

Rules for every cold email you write:
- Keep emails under 100 words.
- Avoid fake compliments and generic personalization.
- Cite the specific signal you used for each angle.
- Never invent facts. If you do not know, say so.
- Separate known facts from hypotheses.
- Write in a direct founder/operator tone. No corporate fluff.
- Include a confidence score (0-10) on each lead and each draft.
- Flag weak leads instead of forcing copy.
- Keep CTAs low-friction and permission-based.
- Never claim I have done research I did not actually do.

My product:
[Paste product one-liner, who it helps, and the core outcome.]

My ICP:
[Paste who you sell to and who you do not.]

My tone:
[Paste 1-2 example emails or describe your voice.]

When I paste a lead list, follow the workflow I give you. If context
is missing, ask before guessing.

The 30-minute workflow

Realistic first batch: 20 leads if doing it manually, up to 100 if the list is clean and your context is strong. Review 10 yourself before sending anything.

The 3 core prompts

Run them in order. Paste your data into the bracketed fields, and keep the human review at the end.

Here is my lead list as a table. Classify each lead as A (strong fit,
write first), B (possible fit, needs context), C (weak fit), or Skip
(not relevant or risky).

For each lead return: priority, fit_reason, likely_pain,
missing_context, confidence (0-10), recommended_next_action.

Use only the data I gave you. Do not invent facts. If a lead is
unclear, mark it B or C and tell me exactly what context is missing.

[PASTE LEAD LIST]

For each A and B lead, build a 60-second brief using ONLY the data
and source material I paste (website copy, LinkedIn text, job posts,
notes). Cite which source each point comes from. Do not invent
research.

Return: what_we_know, what_we_infer, possible_trigger, likely_pain,
personalization_hook (the reason-to-contact first line), source_used,
confidence (0-10).

If you cannot infer a real hook, say "insufficient context" and tell
me what to paste.

[PASTE LEAD ROWS + SOURCE MATERIAL]

For each lead with a brief, write one cold email under 100 words.
One idea, cite the signal in the first line, no fake flattery, no
links, single permission-based CTA.

Then QA it (1 point each, 10 total): real-signal first line,
prospect-specific, plausible pain, under 100 words, single CTA,
low-friction CTA, no fake personalization, does not sound like AI,
no unsupported claims, forward-to-a-friend test.

Return the email, the score, and a one-line fix for any failed item.
Decision: 8-10 send, 6-7 revise, 0-5 kill.

[PASTE BRIEFS]

Claude cold email prompt example

Subject: Quick question

Hi Sarah, I hope this email finds you well. I came across AcmeOps and
was really impressed with what you are building. I would love to hop
on a quick 30-minute call to show you our revolutionary AI platform
that can 10x your outbound. Are you free this week?

Subject: outbound before the SDR ramp

Hey Sarah, saw AcmeOps is hiring SDRs while you are still posting
about outbound experiments. Usually that means the founder is carrying
the first version of outbound before the hire ramps. We turn a lead
list into researched drafts your team reviews and sends. Worth testing
on 20 leads?

What the queue looks like after one pass

Example is illustrative and anonymized.

Cold email QA checklist

Run every draft through this before it leaves:

• First line based on a real signal
• Only makes sense for this prospect
• Pain hypothesis is plausible
• Under 100 words
• One CTA only
• CTA is low-friction
• No fake personalization
• Does not sound like AI
• No unsupported claims
• You would forward it to a friend

Why this takes longer than people think

One good personalized cold email is not one task. It is a stack of small decisions: check the company, check the person, find a recent or relevant signal, decide whether the signal is strong enough to use, translate it into a business-relevant angle, write the email, check that it does not sound fake, and prepare the next touch.

For one lead, that is manageable. For 100 to 300 leads, it becomes the same loop over and over: research, prompt, rewrite, QA.

The full manual loop

Templates to run the system

Lead list template

Use this to give Claude clean inputs. The cleaner the list, the less time you waste fixing bad outputs. Track these columns per lead:

Signal strength: weak / medium / strong. Draft status: not started / drafted / approved / sent. Follow-up status: not started / scheduled / done.

Nurture layer: the 3-touch follow-up path

The first email starts the conversation. The follow-up path keeps the account warm without sounding desperate. Good outbound is a controlled sequence of timely, signal-based touches.

Keep each follow-up short. No fake urgency. No guilt. No "just checking in."

Reply handling: what to do when the signal works

When signal-based emails work, replies are not all clean yes or no answers. Use a simple way to route each reply type.

Claude cold email prompts FAQ