Generate Personalized Outreach at Scale

1. Open https://claude.ai and start a **new chat** (stay in this single thread through step 4). 2. Paste either (a) your real column headers plus **5–10 anonymized sample rows**, or (b) headers plus **5 real rows** if this stays private. 3. Paste this prompt and fill the brackets: ``` I'm running segmented outbound — batch prep only. Do NOT draft subject lines, email bodies, or CTAs in this step. Mirror and plan only. My role / offer (one line): [...] Who this is NOT for (one line): [...] Industry / geography focus: [...] Compliance / ethics boundary (opt-outs, no scraped personal emails, daily send cap): [...] Spreadsheet columns + sample rows (paste below): ... Reply with ONLY: (A) A bullet list of columns you can see and what each is good for. (B) "Thin" columns (too empty or noisy to personalize on) — name them. (C) 2–3 proposed **segments** with plain-English **inclusion rules** (who belongs in each bucket). (D) One paragraph on what we should NOT pretend to know about prospects given these columns — no invented facts. ``` 4. Read the output. If segments feel mushy, reply: "Tighten inclusion rules so a row can only match one primary segment; flag overlaps."

1. In the **same chat**, paste: ``` Using the segments and column assessment from your last message, build a merge-field contract for this campaign. Output a markdown table with columns: - merge_token (literal {{like_this}}) - source_column - example_non_empty_value - when_empty_rule (omit sentence / use neutral fallback / skip row) - proof_required (yes/no) Rules: - Every token must map to a column I actually pasted OR to an explicit segment label I control — no {{pain_point}} unless there is a real column. - Add one row for hook_angle: it may ONLY draw from columns tagged proof_required=yes. - Add a final section: BANNED — inferring funding, stack, hiring, or news not present in the sheet. ``` 2. Scan the table. If any token lacks a real column, delete that row and ask Claude to regenerate only the table.

1. Same chat. Prompt: ``` Using ONLY merge tokens from the contract table (same spelling), draft **per segment**: - **Subject variants:** 2 options, each ≤60 characters, meaningfully different angles. - **Opener variants:** 2 options, each ≤90 words, different structure—not just synonym swaps. - **One CTA:** binary or low-friction (reply yes/no, book 15 min, link click). - **Banned phrases** for this campaign (always reject if seen): "I hope this finds you well", "touch base", "synergy", "quick question", "circling back", "I came across your profile". Segments to cover (repeat the block for each): [List segment names from step 1] Constraints: - Tokens stay as literal {{snake_case}} — do not resolve them to example values here. - Openers must differ by **job-to-be-done per segment**, not by adjectives alone. ``` 2. Read aloud one opener per segment. If two segments sound interchangeable, ask Claude to rewrite so the **problem frame** differs.

1. Same chat. Paste your **segment assignment** (rules or a `segment` column) and either: - The **full prospect table** if ≤50 rows, OR - **Rows 1–50 only** with a note that more chunks follow. 2. Paste: ``` Generate outreach rows using ONLY the templates + merge contract from above. Do not invent facts; follow when_empty_rule exactly. Output **TSV** (tab-separated) with header row: segment\tprospect_id_or_email\tsubject\tbody\tpersonalization_notes Rules: - personalization_notes must cite the **exact column name** used for the hook (or say "segment-default" if no proof column fired). - Keep body length within reasonable cold-email bounds for your channel. - If a row lacks data for a proof-required hook, apply the empty rule—never fabricate. After the TSV block, output a **QA tally**: counts per segment + count of rows using fallback vs proof hooks. ``` 3. For **51+ rows**, repeat with "Rows 51–100", etc., keeping the **same header** each time. Append chunks bottom-up in Sheets later. 4. Copy Claude's entire TSV block(s) — you'll paste into Google Sheets next.

1. Open a new Google Sheet → paste the TSV starting at **cell A1** → Data → **Split text to columns** only if tabs didn't auto-columnize. 2. Insert columns: `qa_ok` (Y/N), `send_date`, optional `esp_batch_id`. 3. **Sort** by `segment`, then `prospect_id_or_email`. 4. QA protocol: - Read **5 random rows** aloud from different segments. - Read **every row** where `personalization_notes` shows fallback or empty-hook logic. - Fix copy directly in the sheet OR return to Claude once with: "Rewrite only rows [list] using the segment templates from this thread—do not change other rows." 5. Add a one-line **send-cap note** at the top (Notes cell or row 1 comment): max sends per day + test-send-to-self rule before the real batch. 6. Optional: send yourself row 2 as a real test from your ESP before wider sending.