Brooke Wright
Brooke Wright · @wright_mode
FREE SETUP

A month of emails in one sitting

The exact setup — connect Claude to your Gmail, let it read your real sent emails, and it builds a reusable skill that writes like you actually write. Not like a robot with a newsletter.

Gmail-connected — no pasting Sounds like YOU One reusable skill No copywriter needed
Something went wrong. Please try again.

No spam. Unsubscribe anytime.

Wright Mode — Free Setup

A month of emails in one sitting

Connect Claude to your Gmail once. It reads your real sent emails, turns your voice into a reusable skill — personal banned list included — then drafts every email you'll send this month. Your job shrinks to editing a couple of lines and hitting schedule.

📧 Claude reads your real sent mail 🗣️ Sounds like you, not AI 🧰 One reusable skill, built once 📬 Works with any email platform

What's inside

You've tried this before. You open Claude, type "write me a newsletter about time management", and get back something that reads like a motivational poster had a baby with a corporate memo. So you rewrite most of it, which takes longer than writing it yourself, and you conclude AI can't do your emails.

Here's how bad generic AI writing has got: Wikipedia's own editors keep a field guide for catching it — Signs of AI writing — a running catalogue of the exact phrases and habits that give AI away. Puffery like "stands as a testament". Tidy rule-of-three sentences. "Not only... but also." Your readers can't always name the tells, but they can smell them — and they hit delete.

AI isn't the problem. The setup is. The fix: instead of describing your voice to Claude, or pasting emails in one by one, you connect Claude to your Gmail and let it read your actual sent mail. Not how you think you write — how you provably, in-the-wild write. (Yes, we're using AI to make sure your emails don't sound like AI. I'm aware. That's the one joke about it you're getting.)

❌ How most people prompt

  • No voice context — Claude has never seen how you actually write
  • No real examples — it guesses "friendly newsletter voice" and lands on beige
  • One email at a time — every draft starts from zero, so every draft drifts
  • Nothing stops the slop phrases sneaking back in next month

✓ How this setup works

  • Claude reads your real sent emails straight from Gmail — your openers, your sign-offs, your rambles
  • It turns your voice into a reusable skill you invoke in any future chat
  • A personal banned list of AI tells — pulled from Wikipedia editors' own field guide — is locked into the skill as hard rules
  • Placeholders where your real stories go — nothing gets invented

Think of the skill as a recipe card for how you sound — one Claude keeps on file and follows automatically. Once it exists, Claude stops writing like a well-meaning intern and starts writing like you on a good day. That's the whole trick, and the setup takes about 15 minutes.

The maths on this

If you spend an hour every week staring at a blinking cursor trying to write one email, that's four-plus hours a month. This setup replaces it with one sitting: 15 minutes of setup (once, ever) and one batch session a month. The blinking cursor never gets you again.


The old way to do this was digging through your sent folder and pasting emails into Claude one by one. Not anymore. You connect Claude to your Gmail once, and it reads your real sent emails directly — the whole lot, rambles and all. Five minutes, mostly clicking "allow".

1

Open Claude's connector settings

Go to claude.ai → click your initials (bottom left) → Settings → Connectors. Using Cowork instead? Same thing — connectors work there too.

2

Connect Google Workspace / Gmail

Find Google Workspace (or Gmail) in the list, click Connect, and sign in with the Google account you actually send emails from. If your list emails go out from a different address, use that one.

3

Check it worked

Open a fresh chat and ask: "Can you see my Gmail? Read the subject lines of my last 3 sent emails." If Claude reads them back, you're connected. That's the whole setup.

Privacy, because you're about to ask

Claude only reads your email when you ask it to — connecting Gmail doesn't mean it's sitting there reading your inbox all day. You can disconnect anytime (same Settings → Connectors screen, one click). And if there are folders or labels you'd rather it never touched, just say so in the prompt — "skip anything in my Personal label" works fine.


This is the step that makes everything else work, so don't skip it. One prompt does the whole job: Claude reads your last 50 sent emails, works out exactly how you write, studies what makes AI writing sound like AI, builds your personal banned list — and packages the lot into a reusable skill called my-email-voice that you invoke in any future chat. Build it once, use it forever.

You have access to my Gmail. We're going to build a reusable Claude SKILL that writes emails in my real voice — with a personal banned list baked in, built from Wikipedia editors' own field guide for catching AI writing, so nothing you ever draft for me trips the tells that give AI away. STEP 1 — READ MY REAL EMAILS Read my last 50 sent emails. Skip receipts, order confirmations, forwards, calendar replies, and one-line "thanks!" messages — I only want the real emails where I'm actually talking to a person or my list. STEP 2 — EXTRACT MY VOICE From those emails, document with real quoted examples: - OPENING PATTERNS — how I actually start emails. Quote my real first lines. Note what I never open with. - SIGN-OFFS — how I close, exact wording, including any recurring PS habits. - PET PHRASES — words and expressions I use again and again, quoted exactly as I write them. - SENTENCE RHYTHM — short and punchy or long and rambly? One-line paragraphs? Where do I break for emphasis? How long are my paragraphs really? - PUNCTUATION HABITS — dashes, ellipses, exclamation marks, CAPS for emphasis, emoji (or none). - WARMTH LEVEL — how casual vs formal am I really, and does it shift depending on who I'm writing to? - WHAT I NEVER SAY — formal or corporate phrases that appear in typical newsletters but never once in my emails. STEP 3 — STUDY THE FIELD GUIDE Read this page: https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing It's the field guide Wikipedia's editors use to catch AI writing, and it catalogues the concrete tells. Extract the specific words, phrases and habits it lists. If you can't fetch the page, work from these seed examples pulled from it: - Puffery and fake significance: "stands as a testament", "plays a vital/pivotal role", "underscores its importance" - Overused AI vocabulary: "delve", "landscape", "tapestry", "showcase", "underscore", "fostering", "meticulous" - Negative parallelisms: "not just X, but Y", "not only... but also" - The rule of three: tidy triplets like "adjective, adjective, adjective" - Promotional adjectives: "vibrant", "rich", "profound", "groundbreaking" - Vague attribution: "experts argue", "industry reports", "observers have cited" - Em-dash overuse as an all-purpose punctuation crutch - Mechanical boldface on key terms throughout Then add the stock phrases you know AI assistants (including you) overuse in emails. STEP 4 — BUILD MY PERSONAL BANNED LIST Cross-check Step 3 against Step 2: which of the field guide's tells and stock AI phrases NEVER appear in my real sent emails? Those become my personal banned list. Be specific — actual words, phrases and habits, not vague categories. STEP 5 — OUTPUT THE SKILL Now output a complete, ready-to-save Claude skill named "my-email-voice" containing: 1. My voice rules from Step 2, written as blunt do/don't instructions. Don't flatter me — if I overuse something, note that too. 2. Three short excerpts from my real emails as reference examples of the voice. 3. My personal banned list from Step 4 as HARD CONSTRAINTS — if a draft contains anything on the list, rewrite it before showing me. 4. This rule: NEVER invent personal stories, client results, numbers, or details about my life. Where a real story or example belongs, insert [YOUR STORY HERE: describe what kind of story fits] and keep going. Format the whole thing as a skill I can save and invoke by name in future chats.

Saving it as a skill

When Claude hands you the finished skill, just say "save this as a skill" — in claude.ai and Cowork, Claude will store it as my-email-voice and you can invoke it by name in any future chat (if Claude asks where to put it, paste the output when prompted). From then on, "using my my-email-voice skill, draft..." is all it takes.

Quality in, quality out

The 50-email read is the magic. If your sent folder is mostly two-word replies, tell Claude to go back further or to focus on your newsletter sends. And it must be SENT mail — the way you think you write and the way you actually write are usually two different people.


This is the main event. Your my-email-voice skill is saved, so all you bring now is what's happening this month — and Claude drafts every email you'll send: weekly value emails, a promo sequence for whatever you're selling, and a re-engagement note for the quiet ones on your list.

You only need two things

1
Your my-email-voice skill — built in section 3. The prompt below invokes it by name; no pasting required.
2
Your month, roughly — topics you want to cover, anything you're promoting and when. Dot points are fine. "Week 1: client story about X. Week 2: how-to on Y. Weeks 3-4: promoting my workshop, doors close on the 28th." That level of rough is plenty.
Using my my-email-voice skill, draft a full month of emails for me. Load the skill first and follow every rule in it — the voice rules AND the banned list — before writing anything, and check every email against it before you finish. WHAT'S HAPPENING THIS MONTH: [PASTE YOUR MONTH HERE — topics, anything you're promoting, key dates. Rough dot points are fine, e.g. "Week 1: story about a client win. Week 2: tutorial on X. Weeks 3-4: promoting my workshop at $97, cart closes on the 28th."] DRAFT ALL OF THE FOLLOWING: 1. WEEKLY VALUE EMAILS — one per week for the weeks that aren't promo-heavy. One idea per email, under 300 words each. 2. A PROMO SEQUENCE for whatever I'm promoting above — three emails: the announcement, a value-then-pitch email, and a last-call email for the deadline. 3. ONE RE-ENGAGEMENT NOTE for subscribers who've gone quiet — short, warm, zero guilt-tripping. Its only job is to get a reply or a click. RULES FOR EVERY EMAIL: - Give me 3 subject line options: one curiosity-driven, one direct, one casual. - Match my-email-voice exactly — openers, sign-offs, rhythm, pet phrases, punctuation. If a line breaks a rule in the skill, rewrite it. - If any draft contains a phrase from the skill's banned list, rewrite it before showing me. - NEVER invent personal stories, client results, numbers, or details about my life. Where a real story or example belongs, insert [YOUR STORY HERE: describe what kind of story fits] and keep going. - One clear call to action per email. Not three. One. - No filler intros like "I hope this email finds you well" — the skill knows how I actually open. Present the emails in sending order, each labelled with its week, its job (value / promo / re-engagement), and the subject line options on top.

A few minutes later you're looking at your entire month: subject lines, drafts, the promo arc, the lot. The [YOUR STORY HERE] placeholders are deliberate — Claude flags where a personal story belongs instead of making one up. You drop in the real thing during the edit. Your stories stay yours; Claude just builds the scaffolding around them.

Why batch beats one-at-a-time

When Claude drafts the whole month in one go, it can see the shape of the month — so the value emails warm up the promo, the promo doesn't repeat itself, and nothing reads like it was written by four different people. One email at a time can't do that.


Real talk: the first drafts won't be perfect. Mine weren't — I changed maybe two lines per email. But changing two lines is editing, and editing is a completely different job to staring at a blank page every Tuesday night. Claude does the 80%, you do the 20% that makes it unmistakably yours.

1

Read each email out loud

Anything you wouldn't actually say out loud to a mate — change it. Your ear catches what your eyes skim past.

2

Fill the [YOUR STORY HERE] placeholders

Drop in the real story, in your own words, straight off the top of your head. Two or three sentences is enough — messy beats polished here.

3

Keep a "she changed this" list

Every time you edit a line, jot down what you changed and why. "Swapped 'delighted' for 'chuffed'." "Cut the second CTA." "My openers are shorter than this." Takes seconds per edit.

4

Feed the list back into the skill

Before next month's batch, tell Claude: "Update my my-email-voice skill based on these edits I made to last month's drafts." And when an AI-ish phrase sneaks through, it's one line to kill it forever: "Add 'I'd be delighted to' to my-email-voice's banned list." The skill gets sharper, so next month's drafts need even fewer edits.

This is the compounding bit

Month one, you might edit two lines per email. Month three, it's one. Because the feedback goes into the skill — not a doc you have to remember to re-paste — Claude carries every correction into every future draft automatically. Most people skip this step. Don't be most people.


my-email-voice is built for one job: your emails, in your voice. This one's the general-purpose cousin — paste ANY text at it (a caption, a bio, a proposal, that thing ChatGPT wrote for you last year) and it strips the AI tells out. Same rulebook as before: Wikipedia editors' own field guide for spotting AI writing, no Gmail access needed.

We're going to build a reusable Claude SKILL called "ai-humanizer". Its job: take any text I paste and strip out the tells that make writing sound machine-made — without touching the facts. This skill does NOT read my email or anything else; it only works on the text I hand it. STEP 1 — STUDY THE FIELD GUIDE Read this page: https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing It's the field guide Wikipedia's editors use to catch AI writing. Extract the concrete tells it lists — specific words, phrases and habits. If you can't fetch the page, work from these seed examples pulled from it: - Puffery and fake significance: "stands as a testament", "plays a vital/pivotal role", "underscores its importance" - Overused AI vocabulary: "delve", "landscape", "tapestry", "showcase", "underscore", "fostering", "meticulous" - Negative parallelisms: "not just X, but Y", "not only... but also" - The rule of three: tidy triplets like "adjective, adjective, adjective" - Promotional adjectives: "vibrant", "rich", "profound", "groundbreaking" - Vague attribution: "experts argue", "industry reports", "observers have cited" - Em-dash overuse as an all-purpose punctuation crutch - Mechanical boldface on key terms throughout Then add the stock phrases you know AI assistants (including you) overuse. STEP 2 — OUTPUT THE SKILL Output a complete, ready-to-save Claude skill named "ai-humanizer". Whenever I give it a piece of text, the skill must do all three of these: 1. DIAGNOSE — flag every tell it finds. Quote the offending line and name the tell ("negative parallelism", "puffery", "vague attribution", and so on). 2. REWRITE — return the text rewritten plainly: varied sentence lengths, concrete words instead of abstractions, no puffery. Keep every fact, name and number EXACTLY as given. Never invent specifics, anecdotes or details that weren't in the original. 3. RETURN BOTH — the marked-up diagnosis first, then the clean rewrite underneath. HARD RULE to bake into the skill: the goal is writing that sounds like a person because it reads plainly and specifically — not because it's been padded out or had synonyms swapped in. If the text is already clean, the skill should say so and leave it alone. Format the whole thing as a skill I can save and invoke by name in future chats.

Saving it

Same as before — when Claude hands you the finished skill, say "save this as a skill" and it lives in claude.ai (or Cowork) as ai-humanizer, ready to invoke by name.

How the two fit together: run ai-humanizer over anything, run my-email-voice when it's an email that has to sound like you.


Here's everything this setup needs — and everything it doesn't. Spoiler: the list on the right is why you haven't done this already, and none of it is actually required.

A Claude account with Gmail connected (claude.ai or Cowork)
15 minutes, once, to build the skill
Your month's topics, roughly (dot points are fine)
One batch sitting a month
Copying and pasting your emails anywhere
A copywriter
Tech skills or code of any kind
Writing from a blank page, ever again

Your order of operations: connect Gmail today (section 2, five minutes), run the skill-builder prompt (section 3) and save my-email-voice, then run the batch prompt (section 4) whenever your next email is due. Do the 20% pass, schedule everything in whatever platform you already use, and close the laptop. See you next month.

Start by connecting Gmail

If you only do one thing after reading this page, connect Gmail and run the skill-builder. It takes 15 minutes, you do it once, and it upgrades every single thing Claude ever writes for you — emails, captions, sales pages, all of it. It's the highest-leverage 15 minutes on this page.

Ready to go deeper?

Take your skills to the next level with these resources.