$1,000+ in free AI tools. Access Offers →
Most newsletter operators still use AI the slow way: a blank doc, a pile of tabs, and a chatbot they re-explain themselves to every single time.
This guide fixes that. It's the full Claude setup for running an AI newsletter: every file, every prompt, every step. Block two hours, follow along, and you'll finish with a Claude that already knows your readers, your voice, and your numbers.
Bookmark it. You'll come back to the prompts.
Disclosure: some links to beehiiv in this guide are affiliate links. If you sign up through them, Bizziology AI may earn a commission at no extra cost to you.
The 10% problem
Most people use Claude like a vending machine. Type a question, take the answer, close the tab.
That works for one-off questions. It falls apart for a newsletter, because a newsletter is the same job every week. Same audience, same voice, same format, same sponsors. If you start every chat from zero, you pay the "who I am and what I want" tax every week, forever.
The fix is boring, and that's why it works: give Claude a small folder of files about you and your newsletter, and let it read them before it writes anything.
What "set up" actually means
A set-up Claude has three layers.
Files about you. Who you are, who you write for, how you sound, what you never want to see in your copy.
Projects. Separate workspaces for separate jobs. Each one has its own files and its own standing instructions.
Connections. Your Google Drive, your Gmail, and your beehiiv account, so Claude can see real numbers instead of guessing.
Once those exist, a prompt like "draft this week's intro from the three stories in my notes" comes back sounding like you, in your format, at your length. Without them, you get the generic AI voice everyone can now spot from across the room.
Here's the full map for a newsletter operator.
Project 1: Newsletter Desk. Where issues get outlined, drafted, edited, and titled. This is the one you'll use most, so it's the one we build first.
Project 2: Research Radar. Where you dump links, papers, launches, and threads, and turn them into "what actually matters this week."
Project 3: Growth and Sponsors. Sponsor pitches, media kit updates, referral program ideas, and reading your numbers.
Start with Project 1 only. Add the other two once the first one earns its keep.
Why Claude for this
You can build a version of this in other tools. I'm picking Claude for three reasons.
Projects keep jobs separate. Your sponsor emails don't leak into your editorial voice, and each workspace keeps its own files and instructions.
It follows long instructions well. A banned-words list only matters if the model actually respects it on draft five.
It connects to beehiiv. beehiiv has an official connector for Claude, so Claude can read your post and audience data and even set up drafts. (More on that in Step 8.)
What it costs: Claude's Free plan includes Projects (up to five), Skills, and connectors, so you can build the core of this setup for $0. Pro is $20 a month, or $17 a month billed annually, and adds unlimited Projects, scheduled tasks, Research, and more usage. I'd start free, then upgrade once you're using it every week.
The 2-hour setup, step by step
Block a morning. You only do this once.
Go to claude.ai, open Projects in the sidebar, and create a new project. (Claude is rolling out a new version of Projects in stages, so your screen might look slightly different. The idea is the same.)
Name it after the job, not the brand: "Newsletter Desk" beats "Bizziology." Then fill in the description properly. Something like:
Weekly AI newsletter for small business owners and solo operators. Goal: every issue is useful enough that readers forward it. This project handles outlines, drafts, edits, subject lines, and repurposing.
Don't skip the description. It's the first context Claude gets.
Step 2: Let Claude interview you (30 minutes)
You don't write your About file. Claude interviews you and writes it.
Open a new chat inside the project and paste this:
I want you to build a profile of me and my newsletter that you'll use in every future chat in this project. Only use facts I give you. Never invent details about me, my readers, or my results.
Interview me. One question at a time. Follow up when an answer is vague. Cover:
- who I am and why I started this newsletter
- who the reader is: their job, their skill level with AI, what they're stressed about
- what a great issue does for them
- my format: sections, length, cadence, send day
- my monetization: sponsors, paid tier, products, affiliate links
- my numbers right now: subscribers, open rate, click rate, growth sources
- what I've tried that didn't work
- topics I cover, topics I avoid, and opinions I hold strongly
When we're done, write everything into one document called about-me.md with clear headers.
Answer honestly, including the messy parts. "Open rate is 38% and I don't know why it dropped" is more useful to Claude than a polished bio.
When it's done, download the file and add it to the project's knowledge. This is the most important file in the whole setup.
What's a .md file? A plain text file with light formatting (# for headings, - for bullets). Claude reads it natively. You can open and edit one in TextEdit, Notepad, Google Docs, or any code editor. No special app needed.
Step 3: Capture your voice (20 minutes)
Don't describe your voice. Show it.
New chat, same project:
I'm going to paste 3 pieces of my writing that performed well and sound most like me. Study them closely.
Look at: sentence length, how I open, how I transition, how I explain technical ideas, my humor, how I end, words I repeat, punctuation habits, how I write calls to action.
Then ask me 5 follow-up questions about what the samples don't show: words I hate, phrases I'd never use, how formal I get with sponsors, how I sound when I disagree with a hyped launch.
Then write everything into my-voice.md, with a short "do" list, a "never" list, and 3 example sentences in my voice.
Pick samples that actually did well: your best-clicked issue intro, a post that got replies, a LinkedIn post that spread. If you're brand new, use emails you've written to friends. That's closer to your real voice than anything you've polished.
Step 4: Add the anti-AI rules (10 minutes)
Your readers know what AI copy sounds like. Many of them use it all day. One "let's dive in" and you've lost them.
So you give Claude a list of tells and replacements. Here's how the list starts:
The full file is in the starter pack below as anti-ai-writing.md. Add it to the project, then add your own pet hates over time. Every time you catch a phrase you don't like in a draft, it goes on the list.
Step 5: Write the project instructions (15 minutes)
Every project has an Instructions field. Whatever you put there applies to every chat in that project.
Use the template in the starter pack (project-instructions.md), fill in the [BRACKETS], and paste it in. The core of it looks like this:
You are the editor of [NEWSLETTER NAME], a weekly AI newsletter for [READER].
Before writing anything, read about-me.md, my-voice.md and anti-ai-writing.md in this project.
This year's goal: [ONE SENTENCE, e.g. "Grow from 2,000 to 10,000 subscribers while keeping open rate above 45%."]
Rules:
- Write in my voice. Follow every rule in anti-ai-writing.md.
- Never invent statistics, quotes, tools, or features. If you're not sure something is true, mark it [CHECK].
- Every tool or claim I'll publish needs a source link.
- Default length for an issue: [WORDS]. Default sections: [LIST].
- If my request is vague, ask one question before drafting.
That goal line matters more than it looks. When you ask "is this a good topic for next week?", Claude stops judging the idea in the abstract and checks it against what you're actually trying to grow.
The [CHECK] rule matters even more. An AI newsletter that publishes a hallucinated feature loses trust fast. Make "unverified" visible so you catch it before your readers do.
Step 6: Add three templates (20 minutes)
Pick the three things you write every week. For most newsletter operators that's:
template-issue.md: your issue structure, with your best past issue pasted underneath as the exampletemplate-sponsor-pitch.md: the sponsor email that actually got a yestemplate-repurpose.md: how one issue becomes a LinkedIn post, an X thread, and a short video script
Real examples beat descriptions. Paste the version that worked, not the version you wish you'd sent. Starter versions of all three are in the pack.
One rule: templates live in the project they belong to. The sponsor pitch goes in Growth and Sponsors, not Newsletter Desk. A project with tight, relevant files writes better than one stuffed with everything.
Step 7: Feed it your real numbers (15 minutes)
This is the step everyone skips and it's the one that makes Claude useful for decisions, not just drafts.
From beehiiv, export or screenshot:
your last 10 to 20 posts with open rate and click rate
your top clicked links
your subscriber growth by source (referrals, recommendations, social, organic)
Add them to the project. Now "which of these three story angles should lead?" gets answered against what your readers actually clicked, not what sounds smart.
Step 8: Connect your tools (10 minutes)
In Claude's settings, open Connectors. Connect:
Google Drive, if your notes and drafts live there
Gmail, so Claude can find sponsor threads and reader replies
beehiiv: find it in Claude's connector directory and sign in with your beehiiv account
The beehiiv connector lets Claude analyze your posts and audience, and it can also create drafts in your workspace. Use that, but keep one rule: Claude drafts, you publish. Every post gets your final read before it goes out.
Connectors share your data with Claude, so connect only what this work needs.
The weekly rhythm
Setup is a one-time job. The rhythm is what pays you back every week.
Here's what each day looks like, with the actual prompts. Notice how short they are. The files do the heavy lifting.
Monday: research brief (scheduled). On a paid plan you can schedule tasks that run on their own, even when your laptop is closed. In a Claude chat, type something like:
Every Monday at 7am, search the web for the most important AI news from the past 7 days for small business owners. Give me the 10 stories that matter most, each with a one-line "why it matters to my readers" and a source link. Skip funding rounds unless they change what a small business can buy.
Claude confirms the schedule and drops a fresh brief in your Scheduled list every Monday. You review it with coffee.
Tuesday: pick and outline. In Research Radar:
From Monday's brief, pick the 3 stories my readers will care about most. Rank them using my past click data. Then outline this week's issue using template-issue.md.
Wednesday: draft. In Newsletter Desk:
Draft this week's issue from the outline. Mark anything you're unsure of with [CHECK].
Thursday: edit and title. Rewrite the draft yourself. Then:
Here's my edited version. Give me 10 subject lines and 3 preview texts. Rank them against my best-performing subject lines.
Friday: send and repurpose. Publish in beehiiv. Then:
Turn this issue into a LinkedIn post and an X thread using template-repurpose.md.
The Monday after: review. With beehiiv connected:
Pull last week's issue stats. What worked, what didn't, and one thing to change in the next issue?
Do the human parts yourself: picking the angle, the final edit, fact-checking every [CHECK], and hitting send. Claude does the blank-page work. You do the judgment.
Bonus: make it learn from your edits
Everything above gets you a smart assistant that knows you. The next level is an assistant that gets better every week.
The problem: you fix the same things in every draft. Too formal here, intro too long there, weak call to action. If those fixes only live in your head, Claude keeps making the same mistakes.
The fix is a file called edit-log.md. After you finish editing an issue, paste your final version next to Claude's draft and run:
Compare your draft with my final version. List every meaningful change I made and the rule behind it. Add new rules to edit-log.md in the format: "Rule. Why. Example."
Upload the updated log to the project. After a month, your first drafts need noticeably less fixing, because Claude is now working from your actual edits, not your guesses about what you want.
Same idea for sponsor feedback and reader replies: if it lives somewhere Claude can read, it compounds. If it lives in a phone call, it's gone.
Once your Newsletter Desk works week after week, turn the issue workflow into a Skill. A Skill is a reusable instruction file you add under Customize, then Skills. The starter pack includes one (newsletter-issue-skill) you can upload as a ZIP and adapt.
The starter pack
Everything in this guide, ready to download:
START_HERE.md: which file to use when (read this first)prompts/: the interview, voice, and weekly-rhythm promptsanti-ai-writing.md: the full banned-phrase listproject-instructions.md: the template from Step 5templates/: issue, sponsor pitch, and repurposing templatesedit-log.md: the feedback-loop file, ready to fill innewsletter-issue-skill.zip: a starter Skill for drafting issues
→ Get the Bizziology AI Claude starter pack (free Google Drive folder, open START_HERE.md first)
Where to start
That's the whole setup. About two hours if you stay focused.
At the end you'll have a Claude that knows who you are, who you write for, how you sound, what you're trying to grow, and what good looks like for your newsletter. Your prompts get short. Your drafts come back usable. Your Mondays get easier.
If you only do one thing today, do Steps 1 and 2. A Newsletter Desk project with a solid about-me.md already beats starting every chat from zero.
Sources: Plan features and prices from Claude's pricing page. Projects details from Claude's help article What are projects?. Skills from Use skills in Claude. Scheduled tasks from Schedule recurring tasks in Claude Cowork. beehiiv connector from beehiiv's MCP getting-started guide. Features and prices current as of October 2026; check the links before you subscribe.