00 Start

I don't explain context to my AI anymore. At all.

Want to hear something funny?

In the last month I haven't once explained to the AI who I am, what I sell or how I write. Not once.

No pasting case studies. No attaching audience files. No typing "my tone is friendly but expert" (we've all written that sentence at least once, admit it).

And no, the model didn't get smarter.

It's simpler than that... and that's the whole beauty of it: I have files it reads on its own.

And this is where it gets interesting.

Today you'll build the same thing for yourself. The starter version... right here, as you move through this page. One page, one action per step, and the prompts are already written — they fill themselves in with your answers. I've deliberately stripped it down to the bare minimum, so that tonight you look at your finished base and go: "Wait. That's all it took?"

What you'll walk away with

• A "style" file: the AI writes in your voice by default. Any task, no explanations
• An "audience" file: your people's pains and words, taken from real messages. They work in every sales text you'll ever write
• A "project" file: what you sell and how your work is set up. Done — you never explain it for the 100th time again
• A "my content" folder: your whole channel, broken down post by post. Articles, decks and websites get assembled from it later
• As a bonus: an understanding of how this same base powers websites, offers and talks... not just posts

Nothing here is theory I haven't run on live projects — this is the exact method I use every day.

00 Start

How many times this week did you explain the same things to your AI?

Three? Five? There was a time I did it ten times a day.

Look, does this feel familiar...

You open a chat. You paste a pile of context: product, cases, audience, tone of voice. The AI answers. You edit. You close the tab.

Tomorrow — all over again... You dig for yesterday's chat. The context gets muddier with every message. You're practically yelling at the AI to remember what you were even talking about. You feed it the same data again...

Sound familiar? Exactly.

Now here's the thing the smartest teams have already figured out: the era of prompt engineering is over. Done. While everyone else is collecting "100 best prompts" (you have that file saved somewhere too, don't you?)... they've moved on to context engineering: you hand the AI your context once, and from then on you just steer it.

How? I'll show you. It's easier than it sounds... much easier.

00 Start

A knowledge base is not a pile of useless folders in Obsidian. And definitely not "one magic prompt".

Let me explain the difference — it matters.

A prompt: a one-time command. Paste it, get an answer, paste it again tomorrow. A file attached to a chat lives for one conversation. Close the chat... and hello, we're starting from scratch.

A base... is a digitized piece of you. Expertise, cases, methods, style, your audience's own phrases. Built once. Plugged into any task. On its own.

I build far more than posts from my base. My website, my conference decks, my offers, the page you're reading right now... all of it grew out of the same files. Because what lives in the base isn't "content". It's expertise. And expertise is needed everywhere.

Today we'll do the first setup of your base using the simplest, most popular task: content. That's where the result shows fastest... within half an hour the AI will write a post in your voice. And the mechanics are exactly the same for every other task.

The starter version of the base: literally 3–4 files. And you don't even write or save them by hand — the AI drafts the text itself, and on the phrase "save this as an md file in my base" it files it where it belongs. From then on it reads them for every task, without you lifting a finger.

Here's what you need to start:

01 Tool

Where your base will live

Same method, same files. The only difference is which AI works with them. Pick the one you use more... and every instruction from here on will match your choice.

01 Tool

Create the folder. It's your base's home

The base lives in one folder on your computer. And for ChatGPT to see it in every chat, set up a "project" and attach the files from that folder to it. It's simple:

1. Create a folder on your computer called My base. Every file the AI builds next will live there: style.md, audience.md, project.md
2. Open the ChatGPT desktop app → in the sidebar, Projects → New project, name it My base
3. Attach the files from the folder to the project, and paste one line into the instructions field:

Before answering, always take the project files into account: "style" — how I write, "audience" — who I write for, "project" — what I do and sell. Never ask me to re-explain them.

01 Tool

Create the folder. It's your base's home

With Claude it's even simpler: the base lives in a regular folder on your computer. Cowork and Claude Code read its contents on their own... and they can create files in it on their own too.

This step is exactly two actions:

1. Create an empty folder called My base: on your desktop or in Documents (right click → "New folder")
2. Open that folder in Cowork (or Claude Code) and send the first message: "This is my knowledge base folder. Read files from here, and save here the files I ask for"

01 Tool

Fill in three fields. The prompts assemble themselves

These aren't base files yet — just three short answers for this page. Everything you type here gets written into the ready-made prompts ahead: instead of "[INSERT YOUR NICHE]" you'll be copying finished text. This page's little bit of magic.

I am [WHO YOU ARE]. I sell / do: [WHAT YOU SELL]. My audience is [YOUR AUDIENCE].

02 File 1 · Style

The "style" file: the AI is about to describe you better than you could

That's it, theory's over. From here on it's hands-on... and we start with the most addictive file of the three.

Here's what's about to happen. You'll hand the AI your posts, and it will break your style down to the molecules: favorite words, rhythm, how you open, how you close, what you never do. Details you've never noticed about yourself. Reading about yourself from the outside... quite the experience, fair warning.

Open your channel or blog and copy your last 8–10 posts into one document. Don't cherry-pick the "best ones" — take them in a row: your style lives in the everyday posts, not the polished ones.

Then hand them to the AI with this prompt:

Here are my posts. Study them and describe my writing style:
— sentence length and rhythm
— favorite words, turns of phrase and punctuation
— tone: how I address the reader
— what's characteristic of me: how I open, how I close
— what I never do

[PASTE YOUR 8–10 POSTS HERE]

Format the answer as an instruction for an AI — "write like this: …" — so that texts in my style could be written from it without me.

How to save the answer as a file

No "copy this into a text editor". One phrase, and done:

• In Claude (Cowork / Claude Code): just write "save this as an md file in my base". Claude creates the file in the folder itself. That's it, seriously.
• In ChatGPT: same phrase — in the desktop app the agent saves the file straight into your "My base" folder. Working in a regular chat? Then it hands you a ready file... drop it into the folder and attach it to the project.

The "audience" and "project" files ahead are saved with the same phrase.

03 File 2 · Audience

The "audience" file: your people's pains in their own words

Collect real phrasing: comments under your posts, DM questions, reviews. Not an "ideal client avatar" from your head (Sarah, 34, has a mortgage)... but living messages from living people.

I am [WHO YOU ARE]. My audience is [YOUR AUDIENCE].

Here are real messages from my audience — comments, questions, reviews:

[PASTE COMMENTS / DM QUESTIONS / REVIEWS HERE]

Extract from them:
1. The frequent pains — verbatim, in people's own words, no translation into marketing speak
2. Objections and fears — what stops them
3. The words and expressions the audience itself uses

Format it as an "my audience" reference for an AI: from it, it should be clear who to write for and about what.

Then the same phrase: "save this as audience.md in my base". The file appears on its own.

04 File 3 · Project

The "project" file: so you never explain what you sell for the 100th time

Describe your project in 10–15 lines. As is, no polishing: what you sell, what the stages of your work are, what your offer is, what makes you different.

Here's my project description — as is, unpolished.

I am [WHO YOU ARE]. I sell / do: [WHAT YOU SELL].

[ADD 10–15 LINES: what the stages of working with you are, what's inside the product, what it costs, what makes you different]

Structure this into a short "my project" reference for an AI: what I do, for whom, what and how I sell, how the process works. Don't invent or embellish anything — work only with what I've written. If something important is missing — ask me questions, as a list.

Same phrase: "save this as project.md in my base". If the AI asked questions, answer them and ask it to update the reference. These are the most useful two minutes of this step... this is exactly where the things you've never put into words about your own product come out.

04 File 3 · Project

Stop. Look at your base.

It's not empty anymore: style, audience, project. And it took less time than you usually spend on a single post.

Now the fun part. Ask the AI for the thing you've asked it dozens of times... only now, with the base:

Using my "style", "audience" and "project" files, write a post for my audience on the topic: [A TOPIC YOU'VE BEEN MEANING TO WRITE ABOUT]

And compare it with what it used to give you, without the base. The difference usually shows from the first paragraph: the text lands in your rhythm, speaks your people's words and doesn't invent someone else's product for you. This is where most people get their "oh — it actually works" moment.

05 Secret level

Trick 1: your whole feed into a content library

Most platforms will hand you a full export of everything you've ever posted, in a couple of clicks. No copying posts one by one.

Pick whichever matches where you actually post:

1. Instagram or Facebook — Accounts Center → Your information and permissions → Download your information → select "Posts", format JSON
2. X (Twitter) — Settings → Your account → Download an archive of your data
3. LinkedIn — Settings → Data privacy → Get a copy of your data → tick "Posts"
4. Substack or a blog — most have an export/CSV option under settings; otherwise copy your last 15–20 posts into one document
5. No export on your platform? Copying your last 15–20 posts by hand into one document works just as well — ten minutes, same result

Whatever you end up with — an export file or a pasted document — hand it to the AI, and it breaks it into a folder where each post lives separately:

Here's an export of my posts (or my last 15–20 posts, pasted below). Break it down into a "my_content" folder inside my knowledge base — one file per post. Don't merge everything into one document: I need a library I can later pull material from for Reels, carousels and new posts.

Create a separate file for each post. At the top of the file — a short header:
topic, format (story / how-to / case study / opinion / list), date, reactions and comments if you have them, the first hook line verbatim, 3–5 keywords for search.
Below that — the full text of the post, unedited.

File name: the date plus a short meaningful title, e.g. 2026-04-12_three-funnel-mistakes.md

Also create an _index.md file in the same folder: a list of all posts by topic — title, date, format, one line on what it's about. So one file shows the whole feed at a glance.

And a _what_works.md file: the top 10 posts by engagement, and a breakdown of what they have in common — topic, format, first line, length, delivery.

Important: process ALL the posts, discard nothing. Even a weak post can become a Reel or a carousel later.

[PASTE YOUR EXPORT FILE CONTENT, OR YOUR LAST 15–20 POSTS, HERE]

05 Secret level

Trick 2: your audience's pains from a chat, verbatim

If your audience lives in a WhatsApp group, or you've got DMs full of real questions, this pulls the whole dialogue out at once — the most honest source for the "audience" file in existence.

1. WhatsApp — open the chat or group → tap the name at the top → Export Chat → without media. You get a .txt file with the whole conversation
2. Instagram or Facebook DMs — the same "Download your information" export from Trick 1 includes your messages too — just tick "Messages" as well
3. No export available? Copy 15–20 real comments, DM questions or reviews by hand — that's plenty to work with

Hand whatever you end up with to the AI:

Here's a chat export (or a collection of comments/DMs/reviews) from my audience. Pull out every phrasing of pain, every question and request that keeps repeating — verbatim, in people's own words, preserving their vocabulary.

Group them by topic, from most frequent to least. This is an addition to my "audience" file — people wrote down what worries them themselves, with no polishing into an "ideal client profile".

[PASTE THE CHAT EXPORT, OR YOUR COMMENTS/DMS/REVIEWS, HERE]

Then say: "add this to audience.md in my base"... and the file gets several times sharper.

The chat and the feed... just two sources. Next, transcripts of calls, consults and webinars go into the base the same way. And then the AI starts analyzing not just your content, but how you sell.

06 The ceiling

Look at what you have. And at what it can grow into.

Three files, a content library folder and your audience's pains from a live chat. That's already more than 90% of the people around you have. Not nothing. Seriously, not nothing.

But I've got two pieces of news for you. The good one: it all works, you just saw it with your own eyes. And the second one... what you built today is about ten percent of what a base can actually do.

You can wire much more into it: ready-made algorithms for posts, threads and carousels, reviews of your sales calls, rules by which it updates itself. Here's how it looks side by side:

The right column... is no longer "copy-paste". It's architecture: folders laid out so the agent doesn't get lost, and rules by which it knows what to open and where to save. I got to this architecture through months of trial and error. Not because it's hard... but because nobody showed me the logic upfront.

07 People

What a base looks like when it grows into architecture

You've seen the quotes along the way. Here's what stands behind them: what people actually put into their bases once they stop assembling them by hand.

Different niches, different tasks... same mechanics. The one you just built with your own hands. Only taken all the way.

08 Finale

You have a base — not a "do it later" note

Right now your base runs at 10% of its potential. Let it take over 80% of your routine

Look at what just happened: in an hour you built a base and saw the result with your own eyes. Well... that's only the beginning. About ten percent of what it can handle.

With a couple of weeks of building it out, the base turns into a tool that — from a single request, standing on your expertise:

• assembles carousels and Reels
• writes viral threads
• builds presentations
• designs nurture sequences and funnels
• digitizes your calls
• assembles websites, guides... and whatever else you dream up
• writes hooks, new angles and ideas
• plus any other task you decide to digitize

One tool = 20+ working hours saved.

This isn't theory: my own base carries my content, analytics, call reviews and talk prep today. The starter version you built tonight is the foundation — everything above grows from these same files, piece by piece, as you keep adding to it.

Use the base starting tomorrow

• Start every new text with "use my base files"
• Once a month, drop a fresh channel export into my_content/: new posts simply get added as files
• A new Reel or carousel? Start with "look into my_content and build … from the post about …"
• Heard a new phrasing of a pain from a client? Straight into "audience"
• Your answers are saved in this browser: come back, and every prompt will already be filled in