Extract a brand voice guide from your five best-performing emails
Reverse-engineers a usable voice guide from the five emails that already earned replies — fingerprint, vocabulary map, evidence-cited tone rules, a 'never' list, and a paste-ready system prompt block for AI drafting. Every rule cites your actual copy, so the guide describes the voice that works instead of the voice a workshop wished you had.
You are a brand voice analyst who reverse-engineers style from evidence, and you hold an unpopular opinion: most brand voice guides are useless because they're aspirational ('bold, human, innovative') instead of observed. The real voice guide is hiding in the copy that already worked. Your job is forensic — extract the patterns from winning emails so any writer, human or AI, can reproduce the voice without the original author.
I'll give you my five best-performing emails. Analyze them and produce a working voice guide:
1. Voice fingerprint — sentence length distribution, paragraph rhythm, contraction usage, how emails open and close, punctuation habits, formatting patterns.
2. Vocabulary map — words and phrases that recur, words conspicuously avoided, and how the copy handles numbers, names, and claims.
3. Tone rules — five to eight observed rules stated as instructions with an example from my emails for each ('Opens with the prospect's situation, never with the sender: see email 2's first line').
4. The 'never' list — patterns absent from all five emails that weaker copy typically has (hedging phrases, exclamation points, whatever the evidence shows).
5. A rewrite demonstration — take one generic sample sentence and show it rewritten in my voice, with a note on which rules did the work.
6. A paste-ready system prompt block, under 200 words, that I can drop into Claude, ChatGPT, or an n8n workflow so future AI-drafted emails match this voice.
Rules: every rule must cite evidence from the emails I gave you — no inventing a voice you'd prefer I had. If the five emails contradict each other stylistically, say so and ask me which direction wins rather than averaging them into mush.
Before you write anything, interview me. Ask me these questions ONE AT A TIME, waiting for my answer each time:
1. Paste your five best emails, with a line on why each earned its spot (reply rate, revenue, the meeting it booked).
2. Who wrote them, and should this guide capture that person's voice or the company's?
3. Where will this guide be used — human writers, AI drafting in a sequencer, a reply agent, all of it?
4. Is there anything in these emails you'd rather NOT canonize — a habit that worked but you're moving away from?
Once you have my answers, run the analysis and build the guide.How to use it
- 1
Copy the prompt into Claude, ChatGPT, or any LLM.
- 2
Pick the five emails by results — reply rate from Smartlead or Instantly, meetings booked, deals sourced — not by which ones you're proud of.
- 3
Answer question 4 thoughtfully; this is your one chance to keep a bad habit out of the canon.
- 4
Drop the system prompt block into your AI drafting flow — n8n reply agents, sequencer AI steps, or your team's LLM workspace.
- 5
Test it: have the AI draft three emails with the block, and check them against the tone rules yourself.
Best practices
Five emails from the same author beat ten from mixed authors — decide whose voice wins before extracting, or the guide averages into mush.
Rerun the extraction quarterly as new winners accumulate; voice drifts, and the guide should follow evidence, not nostalgia.
Give the guide to human writers too, not just the AI — the evidence-cited rules end style debates faster than any opinion.
Pair the system prompt block with a real example email in your AI workflow; rule plus specimen outperforms either alone.
Example: what this looks like in practice
A founder who writes all his own outbound wants his SDR and an n8n reply agent to sound like him. He pastes his five best emails — including one with a 12% reply rate — noting each one's stats. The model's fingerprint finds: emails open with an observation about the prospect, average nine words per sentence, exactly one question per email, no exclamation points ever, sign-offs are just his first name. It flags that email 4 contradicts the others with a formal tone; he confirms it's the outlier and it's excluded. The 190-word system prompt block goes into his n8n reply workflow, and a week later a prospect replies to the AI-drafted follow-up with 'appreciate you writing these personally' — which settles the question.
Best fit
This prompt is one gear in a bigger machine. We orchestrate 20+ tools into outbound systems our clients own — and guarantee the results.
Apply for a Pilot Spot → →Frequently asked questions
Extract rules from copy that already worked, not adjectives from your head. Feed your best-performing emails to the model, have it identify the observable patterns — sentence length, openers, vocabulary, punctuation — and compress them into a system prompt block with examples. That block plus one specimen email in your drafting workflow reproduces the voice far better than 'write casually but professionally'.
More sales copywriting & assets prompts
Generate an outcome-first value proposition buyers can repeat
Builds a complete value proposition system — primary line, three angle variations, a disqualifier, and channel-ready versions — instead of one throwaway tagline. The interview forces buyer, outcome, mechanism, and proof out of your head first, so the output is positioning you can defend, not word salad you'll rewrite next quarter.
Write a case study from a raw customer interview transcript
Turns a raw, rambling customer interview into a publish-ready case study — result-led headline, snapshot metrics box, four-part story, and verbatim pull quotes — while refusing to invent quotes or numbers. If the transcript is missing a strong metric, it tells you exactly what to ask the customer instead of papering over the gap.
Align your cold email and landing page so the click doesn't break the story
Audits your cold email and landing page side by side, finds where the story breaks between click and page, and fixes the cheaper side with minimum edits. The 'message spine' it outputs becomes a five-line contract both assets obey, so the next person who edits either one doesn't silently reintroduce the mismatch.