cold-email-copywriting

Cold Email Copywriting

Overview

This skill writes the emails for one campaign cell: a first touch, two or three follow-ups and a breakup, each with its own subject line and one A/B alternative, a plain text body under 100 words and one ask, the only question mark in the email. Touch 1 opens with a statement (the proof, or a moment in the reader's week), says who is writing and names every segment marker the user gave, and carries the strongest proof the user has. The sequence puts a concrete offer on the table that the reader keeps either way, answers the buyer's first objection in its own touch, and gives every follow-up a new reason to reply. Persona and signal variants of touch 1 go to the rows their gate selects.

The mistakes it prevents: two questions competing in touch 1, a follow-up that restates touch 1 from another side, a result stretched into a promise the brief never made, and "personalization" that changes nothing. Every personalized line comes from a field in the list and connects to the offer.

When to Use

• "Write a cold email to [segment]", "write the sequence for this campaign", "rewrite this email".
• "Give me subject lines", "write first lines from this CSV", "write follow-ups", "write a version for directors".
• "Why am I getting no replies to this email?" (audit mode).
• After cold-email-strategy hands over an offer sentence, cell cards and a touch map.

Prerequisites

1. Offer sentence with a measured result (from cold-email-strategy or the user). With no measured result, lead with the reply offer and a question, never an invented result.
2. The cell: segment, buyer titles, trigger and the ScraperCity source.
3. A sample of rows from the list, to see which fields are filled and how often.
4. Sender first name, full name, company and the opt-out wording the platform will append. Unknown: use {{sender_first_name}}, {{sender_name}} and {{sender_company}}, never "[Sender name]".
5. The sending platform, because merge and fallback syntax differ (Instantly uses {{firstName | there}}).

Ask for anything missing in one message. If the user cannot answer, write with what you have and list the gap once in Confirm Before Launch.

Authoritative Foundations

Digests: references/alex-berman-cold-email.md, references/cold-email-frameworks.md. Routing: references/framework-notes.md.

Workflow

1. Profile the fields

From the row sample, list each field you might use, its fill rate and two example values, in the worksheet's Fill Rates table (the worksheet is your working file, <sequence>-worksheet.md; the client never sees it). Keep a field that is often empty out of the first line, or give it a fallback that reads naturally (references/personalization-and-merge-fields.md).

2. Speak their language and name their slice

List the words this segment uses about its own work, the words they'd say to a colleague, in the "Their words" row (lemlist: write in the prospect's own language). Use them across the sequence. Pull them from the brief, the list's category and title fields and the user's notes; ask the user when unsure. Starter lists by segment: references/line-library.md section 15.

Then list the segment markers: every term the user used for the slice (city, building class, category, platform, size band). Touch 1 and every variant name all of them in the body; dropping "Class B" from "Class B office buildings in Denver" loses the user's own targeting. "Class B buildings in Denver" tells a property manager the email was written for them; "properties like yours" tells them nothing. A marker must be true for every row: gate on the list field, or use a merge tag with a fallback. Say it about the slice you serve or the proof ("we clean Class B office buildings in Denver"), never as a label on the reader's own property: "your Class B building" reads as a rank.

Write the rest in plain operator voice: contractions, short sentences, no marketing adjectives. A timely touch ("before the holiday orders start") is welcome when it's true for the send date.

3. Pick the relevance angle

Place the cell on Schwartz's awareness scale from list evidence (references/cold-email-frameworks.md section 15). No evidence means unaware: lead with a fact or the problem, never the product. Set proof and ask by seniority (references/line-library.md section 12).

Choose one fact per email that connects to the offer. Delete test: if you remove the personalized line and the offer still makes the same sense to this reader, the line was decoration. Fact-to-angle table: references/line-library.md sections 1 to 4.

4. Set the reply offer and the objection

Reply offer: one concrete thing the reader gets for replying and keeps whether or not they buy. Nowoslawski's bar: good enough that they would pay for it. It passes when it is specific to the segment (a walkthrough of one building after the night crew leaves, with a written punch list), takes the reader little time, carries no catch, and the user confirms a delivery date. The row says when they get it ("punch list within N business days of the walkthrough", N from the user). If the brief has no offer or no delivery time, write a specific one in the row starting "Proposed:" and list it once under Confirm Before Launch. Never send an offer the user has not confirmed. Every ask in the sequence and in the Test Plan offers only what this row or the facts list: "want to hear what changed?" needs a brief that says what changed.

Likely objection: the first reason this buyer won't reply, in their words ("tenants always complain", "we already have a cleaning company", "Recharge already has skip"). Nowoslawski: the offer has to answer why someone would not respond. Write it in the "Likely objection" row with the fact from the brief that answers it. If the brief has no answering fact, ask for one. Objections by persona: references/line-library.md section 13.

5. Write touch 1

Short lines, in this order, with one question mark per email (Venetz: one low-effort ask; a second question leaves the reader unsure which to answer):

1. Greeting with a {{first_name | there}} style fallback.
2. A statement opener: the proof, the list fact when the angle is a trigger, or a moment in the reader's week stated as a condition ("If the first you hear about a missed restroom is a tenant email, the night crew has already had a full shift to fix it."). Never a claim about the segment the brief does not make ("Most Class B managers...").
3. Who is writing and what they do, naming every segment marker ("We do nightly janitorial for Class B office buildings in Denver."), then the strongest proof: result, comparable client, timeframe. A reader who stops after touch 1 never sees touch 2.
4. One ask ending in "?": the reply offer, or a one-line question ask ("Want the email we used?").
5. Sign-off: the sender's first name, or {{sender_first_name}}. Every touch and variant ends this way; the footer adds the full name and company.

Hook test. A question ask about the reader's week (in a follow-up, or as touch 1's ask when there is no reply offer) passes all three:

More by segment: references/line-library.md section 15. Body under 100 words (Jason Bay, Gong blog, 2025); aim under 75 (Berman's templates page). Hunter's 2024 data found 20 to 39 word emails replied best, by a small margin. Plain text, no images, HTML styling, tracking pixel or links in touch 1.

6. Write the subject lines

Every touch gets Subject: and Subject B:. A few words (the Lean B2B interview says three to four), lowercase or sentence case, under 50 characters (Berman), no pitch, emoji or fake "Re:" or "Fwd:", matching the first line. Pick from references/subject-line-library.md. Subject: is the default: it carries no merge tag whose fill rate is assumed or low on the sample (a "your building" fallback in most inboxes says nothing). Test the tagged version as Subject B:. Each touch also gets a Thread: line. Berman's follow-up guide keeps the first three or four touches in one thread: touch 1 new; touches 2 to 4 reply to touch 1 and keep "Re:" plus touch 1's A and B subjects. A touch that opens a new thread gets its own subjects. "Re:" appears only on a touch that replies in an existing thread.

7. Write the follow-ups

Every touch gets a row in the worksheet's Touch Map (Type, Adds) naming the new reason to reply: a new fact, the objection answered, the reply offer, a route to the right person. If you cannot fill it with something touch 1 did not say, cut the touch. Days follow Berman's follow-up guide (3 to 5 days apart; his example schedule starts day 0, 3, 7, 11), or the strategy plan's cadence. The Sequence opens with one sending line: weekdays, business hours in the recipient's time zone (Mountain for Denver), under 50 sends per inbox per day (Berman's beginner setup, alexberman.com TAM breakdown) unless the strategy plan sets the cap, the send days and the threading:

Berman's follow-up guide lists a case study with numbers as one thing a follow-up can add: restate touch 1's proof in one line in touch 3 or the breakup, so a reader who skimmed touch 1 still sees the number. A reframe (cost instead of time, tenant instead of owner) counts only when it brings a fact the reader has not seen. Other types and lines: references/line-library.md section 14. Mark a no-ask touch with Type "value-add (no ask)" in the Touch Map: no question mark, never touch 1, one per sequence at most. Never "just checking in" or "bumping this". After the breakup, this cell stops mailing the contact; any re-mail belongs to the strategy plan.

8. Write the variants

Variants replace touch 1 for the rows their gate selects; follow-ups stay shared.

• Persona variant: whenever the cell holds more than one buyer type (portfolio director vs site property manager, founder vs head of ecommerce, owner vs office manager). Same proof, different stake: a portfolio director answers for consistency across buildings and owner reporting; a site manager answers for tomorrow's walkthrough and today's tenant emails. Gate on the title field.
• Signal variant: only when a list field on the contact's own row verifies the signal (an installed app, a technology, a listing date, a rating inside the cell's low band) and measures the problem the offer fixes. A rating fits a reviews offer; for a cleaning offer, an office building's Google reviews are about parking, leasing and tenants, and quoting a low score to its manager reads as a jab. A field from a second pull needs its match step (which column joins it to the contact) in Confirm Before Launch, or the variant goes. Name the field in backticks in Gate: and say what the other rows get. With no such field, write one persona variant and no signal variant. Gates by field: references/personalization-and-merge-fields.md section 9.

9. Use only facts from the brief

Every claim in a body (number, client, result, how the client got it) appears in the worksheet's "Facts from the brief" row or comes from a list field. Three ways an inference creeps in, all banned:

10. Add the footer once

Two lines: sender full name and company, and an opt-out line ("Not the right fit? Reply 'no' and I won't follow up."). Nothing else in the deliverable's footer. As platform settings (not printed): one-click unsubscribe header at bulk volume (Google, Yahoo), and every unsubscribe on one suppression list the same day, across every tool. The footer does not count toward the 100 words.

11. Spintax and AI-written lines

Spin only the greeting or sign-off, never the proof or the ask; preview renders before launch (Instantly: {{RANDOM | option1 | option2}}). For a line that needs more than a merge tag, hand-write about 10 rows (Nowoslawski's count, via alexberman.com), build the prompt from them plus capability and off-limits lists, generate a dry run, read every line, then run the cell. The model only restates values in the row. Details: references/personalization-and-merge-fields.md sections 5 and 8.

12. QA and render check

Render sample rows including one with every optional field empty (one line in the worksheet), read each email on a phone screen, run the hook test on every question ask, then run the checker on the sequence; it reads the worksheet beside it.

ScraperCity Implementation

The copy uses fields ScraperCity returns. Pull or confirm them before writing.

Draft samples:

query_local_business_database  category=["Property management company"] city="Denver" state="CO" hasContact=true limit=20
query_lead_database  title="VP Sales" minEmployees=50 maxEmployees=200 country=["United States"] hasEmail=true limit=20
scrape_store_leads  platform="shopify" apps="Recharge" countryCode="US" emails=true totalLeads=20 offset=0   ($0.08)

Worked cost estimate (copy research for one cell)

Show the user this table and get approval before the full pull. Call check_wallet first.

Field rules

• Ratings to one decimal: "4.9 stars", never "4.9/5.0 rating". Review counts as numbers; below the cell's minReviews filter, skip the count.
• Revenue as a band only when the field is filled; never a guessed figure.
• Leave home addresses, mobile numbers and personal social profiles out of the copy.

Output Format

Two files. The sequence (from templates/output-template.md) is what the client reads; the worksheet (templates/output-template-worksheet.md, saved as <sequence>-worksheet.md) holds Campaign Inputs, Fill Rates and the Touch Map for the checker and stays with you.

The sequence, answer first: title, one line (date, platform), then these ## sections in order: Copy Audit (audit mode only), Sequence, Variants, Footer, Merge Fields and Fallbacks, Test Plan, Confirm Before Launch.

• Sequence: one sending line (step 7), then each touch as a ### Touch N block with Thread:, Subject:, Subject B:, Send day: and the body in a fenced text block (merge tags, not filled values) ending with the ask and the sign-off. No Type, Adds, framework or word-count lines.
• Variants: each ### Variant N: name block has Kind: (persona or signal), For:, Gate:, Subject: and a fenced body.
• Footer: the two footer lines in a fenced block.
• Merge Fields: one line naming the tag notation (the user's platform, or {{tag | fallback}} with "set each fallback in your tool"), then Tag, List column, Fallback. Fill rates and the empty-row render stay in the worksheet.
• Test Plan: one row per test in run order (split, at least 100 delivered per arm, from Berman's 100 to 200+ per variation on his templates page, metric), then a Decide: line: act when one arm's positive replies beat the other's by at least 2 x the square root of both counts added together (a two-proportion check at about 95% confidence, from the arms' own counts); otherwise keep sending. Every alternative ask offers only what the brief backs.
• Confirm Before Launch: each open question once (the proposed offer and its delivery time, a missing fact, a match step, sender details), or "Nothing open." Nothing here repeats the header.

Worksheet rows: Mode, Offer sentence, Segment and buyer, Segment markers, Their words, Facts from the brief, Reply offer (with when they get it), Likely objection, Timing, List source (filters in plain words; a database named only if the user named it).

Mode write leaves out Copy Audit. Mode audit (the user pasted an email) requires ### Original (fenced, with its subject), a line-level table (Line, Issue, Rule broken, Fix) with at least 3 rows, a count line for words and pronouns, and ### Rewrite, which becomes touch 1. examples/worked-example.md and examples/worked-example-worksheet.md show one.

Economy: the client reads emails and the decisions they must confirm, nothing else. Leave out the brief echoed back, QA columns, framework and author names (state a cadence or rule as a choice, or link the page), line-by-line rationales, list-building tool plans, tool ids, cost tables (unless asked), sender-rule tables, and this skill's file paths. No bracketed blanks in copy or footer: a merge tag, or a Confirm Before Launch item.

Quality Check

python3 scripts/check-output.py sequence.md (reads sequence-worksheet.md beside it) enforces:

• Every body under 100 words with exactly one question mark, the ask, followed by a sign-off line; no link in touch 1 or any variant, one at most elsewhere.
• Every number and word ratio ("half", "twice", "a third") in a body appears in the worksheet's Campaign Inputs. Call lengths are exempt.
• Touch 1 and every variant name every segment marker, show the proof or a marker in the first two sentences, and never write "your" + a marker.
• Every touch has a unique, non-bump Adds row in the Touch Map; one Type contains "objection"; touch 1's proof number reappears in a later touch.
• At least one variant; each has Kind, For, Gate and Subject; a signal variant's Gate names a list field in backticks.
• Subjects under 50 characters; Subject B: on every touch; reply touches keep "Re:" plus the parent's subjects; default subjects carry no tag with an assumed or low fill rate.
• Every merge tag has a fallback and a Fill Rates row; no body opens with I, we, our or my (Venetz flags seller-first openings).
• Sequence (or Copy Audit) comes first under a one-line header; the sending line names a time zone and a per-inbox cap; Test Plan has 100+ per arm and a Decide line; Reply offer states a delivery time and a proposed offer appears in Confirm Before Launch.
• The client's document has no Campaign Inputs, Fill Rates, Touch Map, Type or Adds lines, fill-rate or render notes, footer table, bracketed blanks, unlinked author or guide names, legal content, file paths or tool ids.

Earn the Right gate before anything sends (Venetz, 1-10-100 post):

Read for what the checker cannot see:

• Every question ask about the reader's week passes the hook test; the gap answer is the one the proof fixes.
• The reply offer is concrete, kept either way, and confirmed by the user (or marked proposed).
• The objection touch names the objection in the buyer's words and answers it with a fact, not reassurance.
• Each touch's Adds row is true of its body: a reader of touch 2 learns something touch 1 did not say.
• Results are stated as the client's results; nothing is projected onto the reader.
• Persona variants differ in stake, not only in length; signal variants say nothing the gated field does not show.
• No phrase from the banned list in references/line-library.md section 10.

Common Pitfalls

Execution Artifacts

Related Skills

• cold-email-strategy supplies the offer, the cells and the touch map.
• reply-handling writes the responses once someone replies.
• multi-channel-outreach writes the call, LinkedIn and mail touches between these emails.
• email-deliverability and sending-platforms take the finished sequence to the inbox.
• signal-triggered-outbound supplies dated trigger fields for signal variants.
• local-business-list, ecommerce-by-techstack and icp-list-builder build the lists these merge fields come from.