How to build your outbound system
Enter the password
This one is for workshop members. Pop the password in to take a look.
Your outbound prospecting machine
Journalists, podcast hosts, event organisers and editors are all reachable by email. This page turns that into a system: scheduled tasks that research the right people, draft messages in your voice, and tell you when there's something to send. You stay the sender. Claude does the legwork.
- Make your connections →
- Decide your sending email address →
- Set up Google Sheets and notifications →
- Build your /outreach skill →
- Then, and only then, your scheduled tasks
What do you want more of?
Tick everything you want. Each one becomes its own scheduled task with its own Google Sheet. Two or three is a strong start. You can always come back for more.
And if you want more
More options to build your profile and secure opportunities.
The setup, before anything else
Four decisions, made once. This is steps 1 to 3 of the order of operations: connections, sending address, Google Sheets and notifications. The skill comes next, and the scheduled tasks come last.
Which email address are you sending from?
Every draft your scheduled tasks write will sit in this inbox, and the reply watcher will read it.
Make your connections
In the Claude desktop app, open Settings, then Connectors. Connect these before you build anything:
Where do notifications go?
This is your layer of control. Drafts ready, replies in. Pick one, and every scheduled task prompt on this page will reference it by name.
The spreadsheet: Google Sheets, specifically
Google Sheets is the one option that lets Claude open the same file every single run. Create one folder in your Google Drive called Outreach. Inside it, your tasks will each keep one sheet per outreach type, named identically every time. No other spreadsheet tool will work here.
How outbound works here
Six rules. They exist to stop you building one giant broken task. Read them once now and they will save you hours later.
One outreach type, one scheduled task
Never merge them. Podcasts, journalists and event organisers need different research, a different message and a different spreadsheet. A merged task does all three badly.
One Google Sheet per task, in one fixed place
Claude must open that sheet, never create a new one. If it cannot reach the sheet, it stops and asks you for access. It never makes a duplicate. This is written into every prompt on this page.
Claude drafts. You send.
Every message lands in your drafts folder, in your voice, waiting for you. Claude never sends the email. Thirty seconds of your judgement on each one is what keeps the quality high and the relationships yours.
Set the run time for when you are at your desk
Scheduled tasks run on your computer, so the computer needs to be awake for the task to run. Pick a time you are reliably there: with your morning coffee works for most people.
Notifications are your layer of control
Drafts ready, replies in. Your chosen tool pings you when there is something worth looking at. You never have to come back to this page once your tasks are running.
A second task watches for replies
The reply watcher checks the sending inbox, spots responses, and logs them back to the same sheet. Nothing falls through the cracks because nothing depends on your memory.
One message shape, thirteen variations
Casual, short and to the point. It reads like an email to a friend. Every outreach type on this page is a variation on this shape. Get this right and everything you automate inherits it.
Subject lines: the label rule
The formula is expert title plus availability. The reader should know what the email is before they open it.
The ban list
The whole system fails if the messages sound like AI. These are hard bans, written into your /outreach skill on the next tab, checked on every draft.
What gets you ignored
- Your email looks identical to the forty other cold emails they got today. Deleted on pattern recognition alone, unread.
- Three paragraphs about you before one word about them. Nobody scrolls to find out why a stranger emailed.
- A compliment they can tell was written by a machine. Worse than no compliment, because now they distrust the rest.
- An ask in the first line. You became work to them before you became useful to them.
The out-loud test
Read it out loud. Would you say this to a friend? What would you think if you got this email? If either answer makes you wince, rewrite before you automate.
Your examples
These are the variations for the outreach types you ticked on tab 1. Same shape, different emphasis. The one line under each example is the rule that makes that type different.
Write one as a human first
This is the step that stops this entire page producing AI-flavoured outreach. The message you write here becomes the reference example inside your /outreach skill, the real sample of your voice the model holds onto for every draft it ever writes. Take it seriously and everything downstream sounds like you.
Claude invents a realistic target person and coaches you through the draft.
No pitch, no ask. Just the email you would actually send to get on their radar. This exact text goes into your skill.
One thing missing: your handwritten message
The skill file below still has a placeholder where your reference example should be. Do the human draft exercise on tab 3 first. Your handwritten message is what stops everything on this tab sounding like AI.
Your /outreach skill
The research changes for every outreach type. Your voice never does. So the voice gets built once, as a skill, and every scheduled task calls it. When you want to change your tone, you change one file, never thirteen prompts. Every task prompt below says "use the /outreach skill" instead of trying to describe your voice from scratch.
Your skill file is ready below, filled in with your details from the panel underneath and the message you wrote on tab 3. Download it, or copy it into a file called outreach.md.
Run this in Claude after installing the skill. It walks through your details, fills anything still in square brackets, and tests the skill on one real message before you automate anything.
Part 2 · Fill this in once
Everything you type here gets injected into the skill file above and every prompt below, so they come out ready to paste, no placeholders left in them.
One prompt per outreach type you ticked
Every prompt is built on the same six-part framework: role and goal, research, spreadsheet, draft, notify, schedule. The research step is what changes. The voice never does, because step 4 always calls /outreach. To set one up: copy the prompt, tell Claude to create a scheduled task with it, and confirm the run time.
The reply watcher and the weekly review
Whatever you ticked, these two run alongside. The reply watcher is what makes the whole system feel alive: someone says yes, and you know within the hour.
Build and ship
A one-time checklist in build order. Once your tasks are running, notifications take over and there is no reason to come back to this page. That's the point.