Write Cold Outreach Sequences with Personalization Tokens

Tier: Core · Hands-on time: ~35 minutes · Tools: ChatGPT, Claude, Google Sheets

Cold outreach fails in two opposite ways: fully generic blasts that scream template, and artisanal one-off emails that cap you at eight a day. The token method splits the difference honestly — the sequence's logic and voice are crafted once, and the parts that must be true per-prospect are structured fields you research, not adjectives a model invents.

When to use — and when not to

• Outbound prospecting at modest scaletwenty to two hundred genuinely qualified prospects who fit a defined profile and plausibly need what you sell.
• Event and trigger campaignsoutreach after a funding round, a hiring spree, a conference — where the trigger itself is the personalization.
• Partner and recruiting outreachthe same mechanics apply to BD and sourcing candidates, with the offer swapped.

And when not to:

• You can't say why this list, why now. If the answer is "we bought a list", no sequence quality survives it. The token method amplifies targeting; it cannot replace it.
• Regulated or consent-required channels. Cold email law differs by country (consent regimes, opt-out rules, B2B vs B2C). Confirm your list and channel are compliant before writing a word — that is a legal question, not a copy question.
• Ten prospects that really matter. Whale accounts deserve genuinely bespoke emails. Use the research prompts here, skip the tokens.

The recipe

1. 1.

   Build the prospect sheet with token columns (10 minutes). In Google Sheets, one row per prospect, and — beyond name/company/title — three researched token columns that will carry the personalization:

   • •
     {{trigger}} — the observable reason for outreach now: "raised Series B in June", "hiring 3 ops roles", "posted about invoice chaos".
   • •
     {{pain_hypothesis}} — the specific problem you believe they have, in their language: "manual reconciliation across 4 entities".
   • •
     {{proof_point}} — your most relevant evidence for someone like them: "cut close time 40% for a 200-person logistics firm".

   Fill these from real research (site, LinkedIn, job posts, news). Rows you can't fill honestly get cut or downgraded to a generic nurture list — an empty token is the system telling you this prospect isn't researched yet.

2. 2.

   Generate the sequence skeleton. One prompt, written once:

   Copy

   Write a 4-touch cold email sequence for [MY OFFER] to [ROLE] at
   [COMPANY TYPE]. Use the literal placeholders {{first_name}},
   {{company}}, {{trigger}}, {{pain_hypothesis}}, {{proof_point}}
   where personalized content belongs — do NOT invent examples.
   
   Touch 1 (day 1): open with {{trigger}}, connect it to
   {{pain_hypothesis}}, one-line offer, soft CTA (a question, not a
   meeting demand). Max 90 words.
   Touch 2 (day 4): new angle — lead with {{proof_point}}. Max 70
   words. Do not say "just following up".
   Touch 3 (day 9): give something free and useful related to
   {{pain_hypothesis}} (insight, benchmark, teardown offer). Max 80
   words.
   Touch 4 (day 14): polite close with an easy out ("if this isn't
   a priority, tell me and I'll stop"). Max 50 words.
   
   Rules: subject lines under 6 words, no clickbait. No flattery, no
   fake familiarity, no "I hope this finds you well". Write like a
   competent peer, not a marketer. Every sentence must survive the
   test: "would I say this to their face?"
   

   A good Touch 1 skeleton looks like this:

   Copy

   Subject: {{trigger}} — a question
   
   Hi {{first_name}} — saw {{company}} {{trigger}}. Usually when that
   happens, {{pain_hypothesis}} gets worse before anyone owns it.
   
   We help teams like yours fix exactly that — {{proof_point}}.
   
   Worth a look, or is this handled already at {{company}}?
   

3. 3.

   Stress-test the skeleton with three real rows (8 minutes). Paste three actual prospect rows from your sheet: "Fill the sequence with each of these rows. Flag any sentence that reads generic or false for a specific prospect." Read the output as the recipient would. Where a fill reads awkward, fix the skeleton, not the fill — the skeleton is the asset that must work across every row.

4. 4.

   Mail-merge and send in small waves. Use Sheets with your mail-merge tool of choice to fill tokens at send time. Send 15–25, wait two days, read the replies — including the annoyed ones, which are your best copy feedback — then adjust the skeleton before the next wave. Never fire the full list on an untested skeleton.

5. 5.

   Instrument replies back into the sheet. Add columns: replied (Y/N), sentiment, objection heard. After 50 sends, paste the sheet back to the model: "Which {{trigger}} types and {{pain_hypothesis}} phrasings correlate with replies? What should the next skeleton change?" Now the sequence improves on evidence, and the objections you hear feed pb-26's ICP work directly.

Make it yours

• Claude for skeleton quality: strongest at the "competent peer" register and at ruthlessly cutting marketer-speak when asked. Keep the skeleton, rules, and your best-performing past emails in a Project so every campaign inherits the voice.
• ChatGPT for research assist: paste a prospect's public info (their site's about page, a job posting) and ask for the three token values as suggestions — you approve or rewrite each. This speeds the sheet-filling without surrendering its honesty; never let suggested tokens flow to send unreviewed.
• Google Sheets as the campaign brain: token columns, merge source, reply tracking, and analysis all in one place. Add a "QA" column — a checkbox per row that someone reviewed the filled email — and merge only checked rows.

Quality check

• Every token is researched fact, not model-invented flattery — spot-check five rows against their sources.
• The skeleton survives the face testyou would say every sentence to the prospect in person.
• Word counts held90/70/80/50; cold email quality is mostly subtraction.
• Touch 4 contains a genuine opt-out and your process honours "no" permanently.
• Compliance checked for your geography and channel before the first send, not after the first complaint.