A weekly content engine: one founder voice, ten posts

A workspace loaded with your voice that turns one or two raw thoughts into a week of posts that sound like you, not a marketing bot, in about 20 minutes a week.

What you'll have when you're done

A repeatable Friday habit: you dump one or two raw thoughts (or a meeting transcript) into a workspace that already knows how you write, and it hands back a week of content, an anchor post, a few short-form variations, a snippet or two, all in your voice. You edit for 15 minutes and schedule. The blank page disappears, and so does the "everything AI writes sounds generic" problem, because the engine is fed on you.

The problem is not ideas. It is the blank page and the bot voice

Most founders post inconsistently for one of two reasons. Either they sit down to write and freeze ("what do I even say?"), or they try AI, get back something that sounds like a LinkedIn motivational poster, and quit. Both are real. You have plenty to say, you just say it in meetings and DMs, not in a feed. And generic AI output is worse than silence. I went through this exact arc: months of posting nothing, then a stretch of posting AI slop I was quietly embarrassed by, then nothing again. The thing that broke the cycle was not more discipline. It was realizing I was asking the AI for the wrong thing, ideas I already had, instead of asking it to carry the ideas I was wasting in Slack threads.

The unlock is to stop asking AI to have ideas and start asking it to give your ideas mileage. Morning Brew's Alex Lieberman built an AI content machine that cut his production time from 20-30 hours per post to 3-4, after spending about 50 hours building the system. That ratio is the whole point: the setup is the work, then it compounds. The setup is teaching a Claude Project your voice.

What you need first

• 5-10 of your best past posts, plus a few meeting transcripts or voice memos where you sound like yourself.
• A short list of your actual beliefs/takes on your industry, the spiky ones.
• A Claude Project (or a CLAUDE.md voice file if you work in Claude Code).
• A scheduler (Typefully, Buffer, or the platform directly) for the back end.

Step-by-step

Step 1Build the voice workspace

Create a Project called "Content engine." Load your best posts, transcripts, and belief bullets as its knowledge. In the instructions, be blunt about voice: "Write like the attached posts. Short sentences. No hype, no emojis-as-personality, no 'in today's fast-paced world.' Spiky takes over safe ones." This is the 50-hour-equivalent step, except it takes you 30 minutes because you are pasting, not building.

The instructions that actually move voice quality are the negative ones, the banned list. Be specific about your personal tells and tics to kill: "Never open with a rhetorical question. Never use 'game-changer,' 'unlock,' or 'in a world where.' No three-part lists where one point would do. Do not end on a forced call-to-action." A model left to its own devices reverts to the average of the internet; your banned list is what drags it back toward you. Then test it before you trust it: give it one thought, read the draft, and if a line is not something you would say, add the reason to the instructions. Three or four rounds of that and the voice locks in.

Step 2Feed it raw input weekly

Once a week, give it something real to work from. Not "write me a post about leadership," that produces mush. Instead: dump a genuine thought ("here's what I think most founders get wrong about hiring their first salesperson, and why"), or drop in a Granola transcript from a call where you said something sharp.

Step 3Ask for a pack, not a post

Prompt it to spread one idea across formats:

From the thought/transcript above, draft a content pack in my voice:
1. One anchor post (the full argument).
2. A shorter punchy version of the same idea.
3. A 3-bullet "carousel" outline.
4. Two one-line snippets I could post standalone.
Keep every piece in my voice. Spiky over safe. No hype.

One idea, multiple shapes. That is how you get a week from a single thought.

Here is what a pack looks like, illustrative, from one dumped thought: "most founders hire their first salesperson too early, before they have sold the thing themselves."

Anchor post: Your first sales hire will probably fail, and it will not be their fault. You are handing them a sales motion that does not exist yet, asking them to discover your pitch, your objections, and your buyer all at once. That is not a job. That is your job, and you have not done it. Founder is salesperson #1. The hire comes after you have closed enough to know what you are handing over.

Short version: Do not hire a salesperson to find your sales process. Hire one to scale a process you have already run yourself. First you close. Then you delegate.

Standalone snippets:

• "Your first AE inherits your clarity or your chaos. Usually chaos."
• "If you cannot close it, you cannot delegate it. You can only abdicate it."

Notice what happened: one belief became four shippable pieces, none of them invented. The engine did not have the take, you did. It gave the take mileage across formats, which is the part you would never sit down and do by hand on a Friday.

Step 4Edit for 15 minutes, you are the final voice

The drafts get you 80% there. Your job is the last 20%: cut the line that is slightly off, add the detail only you know, sharpen the hook. This edit is non-negotiable, it is what keeps the content yours and keeps it off the "obviously AI" pile that platforms increasingly bury.

Step 5Schedule and pace it

Load the pack into your scheduler and space it out. The data says founders do best at 3-4 quality posts a week, not ten in a day, so stretch one pack across a week or two. Consistency beats volume.

How you'll know it's working

You actually post, consistently, which is the whole game and the thing you were not doing before. The content sounds like you (people who know you say "that's so you"), not like a brand account. And the weekly time cost stays around 20 minutes, low enough that it survives a busy week.

When it breaks

• It sounds generic. The Project is underfed. Add more real posts and transcripts; voice quality is a direct function of how much you it has seen.
• Every post sounds the same. You are feeding it the same kind of input. Vary the raw material: a rant, a transcript, a contrarian take.
• Posts feel like AI wrote them. You skipped the edit (Step 4), or the voice instructions are too soft. Tighten both.
• You burn out trying to post daily. Drop to 3-4 a week. The engine makes volume possible, not mandatory.
• The takes are all safe and agreeable. The model defaults to inoffensive, and inoffensive does not travel. Feed it your spikier beliefs explicitly, and add "if a draft would not make at least one person in my industry disagree, sharpen it" to the instructions.
• It keeps reaching for the same three examples. It is anchoring on whatever is freshest in the Project. Rotate the raw input (a new transcript, a different rant) rather than asking it to "be more creative," which just produces stranger mush.

Make it yours. The engine adapts to your channel and your altitude. A solo founder building in public wants raw, first-person, in-the-trenches takes. A CEO of a 200-person company posting on LinkedIn wants the same voice dialed to category-level points of view, fewer "here is my Tuesday" posts, more "here is where this market is wrong." Tell the Project which platform and which register, and feed it samples from people who post the way you want to, so it learns the shape, not just the words.

Where this fits in your harness

This is your top-of-funnel megaphone, and it shares plumbing with the rest of your stack. The Granola transcripts that feed it also feed customer-call case studies and your team updates. The same "load a Project with your voice" move powers personalized outbound. And one good recording can become a week of content via repurposing one recording.