The Cold Email Audit

SUPER_ADMIN_API_KEY=your_api_key_here

Read the .env file in this folder. Confirm you can see SUPER_ADMIN_API_KEY (don't print the value, just confirm it exists). Then use it to hit the Email Bison /workspaces endpoint and list how many client workspaces are accessible. Do nothing else.

Prompt 1

this is the prompt that turned 1.5 million cold emails into three named winning frameworks in a single run.

goal: identify what makes top-performing email copy actually work at the PATTERN level. you want the underlying recipe, not a surface-level list of "good emails."

THE PROMPT (copy-paste ready):

Analyze the last [N] cold emails sent across all client workspaces in [SEQUENCER NAME] using the super admin API key, which is in the .env file as SUPER_ADMIN_API_KEY.

Goal: identify what makes top-performing email copy actually work at the pattern level, NOT just surface a list of good emails.

Do this in 4 phases:

PHASE 1. DATA EXTRACTION
Pull every email body and subject line, plus send date, sequencer step (first email vs follow-up), reply count, and positive reply count.

PHASE 2. DEFINE BEST-PERFORMING
Define winners as: top decile by positive reply rate (positive replies divided by total sent for that variant). Define losers as: bottom decile by the same metric. Exclude any variant with fewer than [MIN_SENDS] sends.
If I want to use different benchmarks, ask me before defaulting.

PHASE 3. PATTERN ANALYSIS
Compare winners vs losers across these dimensions:
→ linguistic patterns: sentence length, word count, casing, whether there's a greeting or preamble, overall tone
→ offer patterns: free deliverable vs paid pitch vs observation, and what the offer actually consists of
→ structure patterns: subject line shape, opener type, body length, CTA type, PS presence
→ vertical breakdowns: which patterns win in which industry segments
→ position-in-sequence: winners that are first emails vs follow-ups

PHASE 4. OUTPUT
Write three files into this folder:
1) findings.md, which holds the named frameworks (Framework A, B, C and so on) with explicit recipes (subject line rule, word count, body skeleton, CTA shape, PS rule) and which ones won
2) winners.md, which holds the top 30 emails verbatim, grouped by framework
3) full_analysis.md, which holds every comparison metric and chart-ready table

Do NOT skip phase 2 or phase 3. Do NOT just dump example emails. The recipes matter more than the examples.

fill in [N], [SEQUENCER NAME], and [MIN_SENDS] before you run.

YOUR TURN (15 minutes):

paste the prompt into Claude Code with your filled variables. wait for it to finish (yes it takes a while at volume, mine took an hour for 1.5M). open findings.md first.

if the findings doc reads like generic advice ("keep emails short, use clear CTAs"), your dataset is too small or your prompt didn't enforce phase 3. re-run with stricter pattern instructions.

KEY TAKEAWAY: the prompt is the entire workflow. one prompt plus your data plus Claude Code equals framework extraction at scale.

Prompt 2

finding the winning script is a great start.

but alone, it won't be enough.

because if all you do is copy the email without understanding WHY it worked, you're going to be back at square one the moment it stops converting. you'll have no idea how to fix it. which is what happens to most cold email writers. they grab a script working for someone on youtube, slot it into their campaigns, see decent results for a couple weeks, and then watch it die. since they never knew why it worked, they go find the next template. repeat forever.

that's what this prompt prevents.

THE PROMPT (copy-paste ready):

Load winners.md from this folder. Pick the framework with the highest positive reply rate.

Do a full structural teardown of that framework. Output a single MD file called framework_teardown.md with these sections:

1) SUBJECT LINE RULES
→ word count range
→ casing pattern (all lowercase, sentence case, etc.)
→ variable usage (first name only, company name only, both)
→ tone

2) BODY STRUCTURE
→ word count range
→ sentence count
→ opener type (greeting, no greeting, signal, pain, offer-first)
→ exact skeleton with [VARIABLES] in brackets

3) MICRO-PATTERNS DOING HEAVY LIFTING
Look for patterns inside the framework that aren't obvious, such as parenthetical scope disclosure, soft-permission CTAs, a list of deliverables in 3 specific items, or an optional PS with a single proof claim. Name each one and quote a winning example.

4) WHY IT WORKS
For each micro-pattern, explain in one sentence WHY it works psychologically. Be specific. Don't write "it builds trust" without explaining the actual mechanism.

5) FAILURE MODES
List 3 specific scenarios where this framework will NOT work. Pull these from the losers data if any losers attempted this framework.

6) ONE-PARAGRAPH CONVERSATIONAL SUMMARY
Write one paragraph (no bullets, no headers) describing the framework as if you were explaining it to a peer over coffee.

FILLED EXAMPLE, what came back for our winning framework:

subject line: 2-4 words, all lowercase, 1-2 variables (company name or first name). example: FYI [First name].

body: 25-35 words, 2-3 sentences. it drops straight into the offer with zero "hey [name]" warm-up. skeleton looks like this:

[First name], if we could [concrete deliverable], completely free, [we'd do X, Y, Z]. You just [overcome objection]. Would that be worth hearing more?

P.S. [single aligned proof claim, optional]

micro-pattern doing heavy lifting #1: parenthetical scope disclosure. after the offer, in parentheses or brackets, you tell them how little they have to do. "we'll do X, Y, Z" plus "you just forward the last invoice." the email is throwing the offer in their face AND overcoming their objection in the same breath.

micro-pattern #2: soft-permission close. "would that be worth hearing more?" is intentionally low-friction. high-friction would be "want to jump on a 30-minute call?" which earns an instant delete. soft-permission lets the prospect either say yes (so you push for a meeting) or surface an objection ("yeah but send more info first"). same direction either way.

micro-pattern #3: optional PS with a single proof claim. test it with and without. sometimes the PS makes the offer feel reasonable. sometimes it reads inorganic. AB it.

WHY IT WORKS:

the offer IS the entire email. just the concrete free thing, no setup or observation or pain hypothetical. when the offer is strong, the email can literally be one sentence. when you talk about pain points or observations, there's a huge chance you missed the mark and emphasized something the prospect isn't experiencing or doesn't care about. once that happens, even the best offer in the world won't save you. you've lost credibility. they're gone.

this framework also has a peer-to-peer texture. lowercase, casual, no greeting, conversational. it reads like a text from a friend offering something incredibly valuable. cold prospects don't know you, they don't trust you, and they don't like you. the only way through that wall is to respect their time.

FAILURE MODES (3, copy verbatim):

1) the offer is weak or unaligned with the prospect's worldview → nothing in the email can save it

2) the offer is too complicated to express in one sentence → you'll need additional explanation and the framework breaks

3) the deliverable isn't actually free, or has a hidden ask → the soft-permission close gets flagged as bait

bookmark these. when this framework stops working for a campaign, the failure is almost always one of these three.

the whole point of this prompt: when the framework stops converting (and it will, eventually), you can pinpoint which mechanism broke instead of throwing out the whole thing.

Prompts 3, 4, 5

once you understand WHY a script works, you're in the perfect position to adapt it.

because the framework does the heavy lifting on its own.

you don't need to create a completely new email from scratch every campaign. all that changes is the offer itself, the pain point you're calling out, and the CTA language. the rest of the recipe stays the same.

this is exactly what prompts 3, 4, and 5 are for.

PROMPT 3, INDUSTRY ADAPTATION (copy-paste ready):

Load framework_teardown.md from this folder.

I'm adapting this framework to a new offer. Here are the inputs:

→ INDUSTRY: [INDUSTRY, e.g. "freight forwarding"]
→ AUDIENCE: [JOB TITLE + COMPANY SIZE, e.g. "importers doing $5M-$50M/yr in goods"]
→ MY OFFER: [WHAT YOU SELL, e.g. "a free rate check across their top 5 shipping lanes"]
→ EFFORT REQUIRED FROM PROSPECT: [WHAT THEY DO, e.g. "forward last invoice"]
→ PROOF POINT FOR PS (optional): [SPECIFIC RESULT, e.g. "average $40k/yr in lane mismatches"]

Write ONE email that:
→ obeys the subject line rule (2-4 words, all lowercase, 1-2 variables)
→ obeys the body rule (25-35 words, 2-3 sentences, no preamble, offer-first)
→ uses the parenthetical scope disclosure micro-pattern
→ ends with the soft-permission close "would that be worth hearing more?"
→ includes the optional PS with the proof point if I provided one

Do NOT add a greeting. Do NOT add an observation about the prospect. Do NOT exceed 35 words in the body. If my inputs are too vague to meet the framework, ask me what's missing before writing the email.

FILLED EXAMPLE, freight forwarding campaign we adapted live:

subject: your shipping rates (alternate to the trusty FYI [First name])

body:

[First name], if we could run a free rate check on your top 5 shipping lanes, we'd pull current carrier pricing, compare it against what you're paying today, and send a side-by-side breakdown. You just forward the last invoice. Would that be worth hearing more?

P.S. most importers we run this for find $30-50k in annual savings in lane mismatches.

that's 28 words in the body. 3 sentences. parenthetical scope disclosure ("you just forward the last invoice"). soft-permission close. PS with a proof claim aligned to the offer.

same framework adapted clean.

PROMPTS 4 + 5, RUN AGAINST 2 MORE INDUSTRIES SIMULTANEOUSLY:

the same prompt template works for any industry. we ran it live against three.

the first was a SaaS LinkedIn lead magnet offer ("if we could build your next LinkedIn lead magnet funnel completely free, we'd write the magnet, design the landing page, and queue the nurture sequence").

the second was email marketing for e-com ("if we could rebuild your highest revenue email flow completely free, we'd write the copy, design the templates, and stage them inside of Klaviyo").

the third was freight forwarding (the example above).

all three obey the same micro-pattern recipe. the deliverable changes per offer. the objection-overcome changes per audience. the proof point in the PS changes per result. everything else stays the same.

YOUR TURN (10 minutes per industry, 3 industries):

fill in the inputs for prompt 3 with YOUR offer.

run it for your primary ICP first.

then run it again for an adjacent ICP you've been thinking about testing, using the same framework with a different deliverable.

then run it again for a stretch ICP, one you wouldn't normally touch. the goal is to prove to yourself that the framework holds across verticals so you stop rewriting copy from scratch every time you onboard a new client.

FAILURE MODE TO WATCH FOR:

if the email Claude Code spits out has a greeting, an observation, or runs over 40 words, your inputs were too vague. re-run with a sharper offer and a sharper effort-required-from-prospect line. don't accept the first output if it broke the recipe.

KEY TAKEAWAY: lock the recipe and swap the inputs.

The One-Page Cheat Sheet

tape this to your monitor.

THE 5 PROMPTS (in order):

1) extract winning patterns from your sent volume → outputs findings.md, winners.md, and full_analysis.md

2) dissect why the top framework actually won → outputs framework_teardown.md

3) adapt the winning frame to industry #1 → outputs one email

4) adapt to industry #2 → outputs one email

5) adapt to industry #3 → outputs one email

THE WINNING FRAMEWORK RECIPE (Framework A, the front-end offer pattern):

Element	Rule
Subject line	2-4 words, all lowercase, 1-2 variables
Body word count	25-35 words
Sentence count	2-3 sentences
Opener	NO greeting, NO observation. drop straight to the offer
Body skeleton	[First name], if we could [deliverable], completely free, [we'd do X, Y, Z]. You just [low-effort action]. Would that be worth hearing more?
Parenthetical scope disclosure	After the offer, tell them how little they do
CTA	Soft-permission: "would that be worth hearing more?"
PS	Optional single proof claim aligned to the offer

THE 3 FAILURE MODES:

→ the offer is weak or unaligned with the prospect's worldview

→ the offer is too complicated to express in one sentence

→ the deliverable isn't actually free, or has a hidden ask

FOLLOW-UP DECAY DECISION TREE (from the 1.5M-email findings):

follow-ups made up 32% of eligible step rows AND 58% of all losers. they consume sending capacity. they burn sender reputation. they produce bottom-decile results. second-email patterns ALWAYS need surgery.

default: send fewer follow-ups, rewrite the first email instead.

THE BENCHMARKS WORTH MEMORIZING (from our 1.5M-email run):

Framework A came in at 63,000 emails sent, 1.57% reply rate, 0.41% positive on percentage of sends.

the winning angle proof campaign came in at 14,309 sent, 7,239 contacted, 151 replies, and 141 positive (93.38% qualified rate of replies).

the capital-raising list pivot came in at 2,305 sent, 156 replies (6.77%), and 67 interested (42.95% of replies) inside 2 days.

the brand-new offer launch came in at 45,804 sent, 1,001 replies (3.96%), and 276 interested across 30 days.

YOUR ASSETS BUILT IN THIS GUIDE:

[ ] Claude Code installed and .env wired up

[ ] findings.md generated from your own sent volume

[ ] winners.md (your top 30 emails grouped by framework)

[ ] framework_teardown.md (your winning framework and 3 failure modes)

[ ] 3 adapted emails for 3 different ICPs

finish this list and you've done what most cold email writers will never do: you've audited your own data instead of borrowing somebody else's screenshot.