Stop Explaining Your Business to AI Every Morning: Build 3 Systems That Already Know It

Here is the problem with prompts. A great prompt gets you one great answer. Tomorrow morning you open a new chat and start from zero again: who you are, what you sell, what you are trying to do.

Think about hiring someone. You can brief your new employee from scratch every single morning. Or you can onboard them properly once, and by week two they already know what matters and what to ignore.

One off prompts are the first employee. Systems are the second. And the people pulling ahead right now are the ones who stopped briefing from scratch and started onboarding.

This guide walks you through three systems, all built in Claude:

1. An AI cofounder. Your second brain. It knows your goals, your projects and your ideas.
2. An AI teammate for one job. A research agent that studies competitor sites, compares offers and prices, and hands you a report with links.
3. A content pipeline. Topic → 5 hooks → script → article and caption. You pick the direction and approve every step.

What you need to start

• A Claude account. The free plan is enough to start: it includes Projects (up to 5) and web search. If you plan to use it heavily every day, the paid Pro plan gives you more room.
• Notion. Only for system 3, as your content database.
• Time. System 1: one to two hours. System 2: about an hour. System 3: about two hours.

Anything in [square brackets] in the prompts below is yours to replace with your own details.

What it is. A Claude Project that knows your goals, projects and ideas, and helps you decide what deserves your week. I call mine my AI cofounder.

Who needs it. Anyone juggling several projects in their head who ends the week thinking: I was busy, but did anything actually move?

01Build it step by step

Role: You are my new AI cofounder. Before we work together, you need to know me well.
Goal: Interview me and turn my answers into 4 context documents you will use in every future chat.
Rules:
1. Ask one question at a time. Wait for my answer before the next one.
2. Ask a maximum of 15 questions in total.
3. If an answer is vague (for example "grow the business"), ask one follow up question to get a number, a date or a name.
4. Cover: my business and offer, my clients, my goals for the next 90 days and 12 months, my active projects and their status, ideas I am holding, how I like to work, what you must never do.
Output: When the interview is done, write 4 separate documents in clean Markdown:
1) goals.md: 12 month goals, 90 day goals, how I measure each one
2) projects.md: every active project with status, next step and deadline
3) ideas.md: every idea I mentioned, with one line on why it matters
4) how_i_work.md: my working hours, tone, tools, what to avoid
Do not invent anything I did not say. Mark missing information as [to fill].

You are my AI cofounder. You know my business through the files in this project: goals.md, projects.md, ideas.md, how_i_work.md. Read them before every answer.

How you work with me:
1. Every suggestion must connect to a goal in goals.md. If something serves no goal, say so directly.
2. Be direct. Tell me when my plan has a problem. I prefer useful pushback over polite agreement.
3. When I give you a list of tasks, sort them into three groups:
   Do: serves a goal and only I can do it
   Delegate to you: you can draft or prepare it
   Drop or postpone: serves no current goal
4. When I delegate a task, do it fully: the draft, the plan or the text, ready to use. Ask at most 2 questions first, and only if you cannot do it without the answer.
5. Never invent facts, numbers or client names. If information is missing from the files, ask me.
6. Answer in [Armenian / English]. Keep answers short unless I ask for detail.

02Daily prompts

It is Monday. Here is everything on my plate this week:
[paste your task list, messy is fine]

Fixed commitments this week: [meetings, deadlines]
Hours I really have for focused work: [number]

1. Sort every task into Do / Delegate to you / Drop or postpone, based on goals.md.
2. From the Do list, pick my top 3 for this week. Explain in one line how each one moves a goal.
3. Point out anything I am avoiding that matters more than it looks.
4. Then start on the Delegate list, one task at a time. Show me the first result and wait for my OK before the next one.

New idea: [describe it in 2 to 5 sentences]

Check it against goals.md and projects.md:
1. Which goal does it serve, if any?
2. What would I have to pause to do it?
3. The strongest argument against it.
4. Verdict: do now / park in ideas.md / drop. One line why.
If the verdict is "park", give me the exact line to add to ideas.md.

Weekly update. This week:
Done: [list]
Stuck: [list]
New: [projects, clients, ideas, changed priorities]

Rewrite projects.md and ideas.md with these changes. Keep the same structure. Show me only the updated files, ready to replace the old ones.
If anything I said contradicts goals.md, flag it instead of changing the goal.

03One week with this system

043 common mistakes

1. Writing the files once and forgetting them. A month later Claude is planning around projects you already finished. The Friday update is not optional.
2. Vague goals. "More sales" is not a goal. "20 new clients by December" is. Without a number, Claude cannot rank anything.
3. One Project for everything. Personal tasks, clients, content, all mixed. One Project, one role. Research and content get their own Projects below.

05Done when

• The Project exists and has 4 files.
• The instructions are saved in Instructions.
• Claude answers "Who am I" correctly.
• You ran your first Monday Priorities with real tasks.
• The Friday update is in your calendar.

What it is. A separate Project you hand competitor websites to. It studies their offers and prices, compares them, and gives you a report with a link behind every claim.

Who needs it. Anyone setting a price, launching a new service, or wondering why clients keep choosing the competitor down the street.

01Build it step by step

02Prompts

You are my competitor research analyst. My business is described in about_my_business.md in this project.

When I give you competitor names or website links:
1. Use web search to study each competitor: main offers, prices, what is included, who it is for, how they position themselves.
2. Only use information you actually found on a page. Every price and every claim gets the link where you found it.
3. If you cannot find or open something (prices not listed, page blocked, prices only on social media), write "not found" and the reason. Never guess a price.
4. Mark when a page looks outdated or shows a temporary discount.

Report format:
A) Comparison table: Competitor | Offer | Price | What is included | Audience | Source link
B) What they all promise (patterns)
C) Gaps: what nobody offers or explains well
D) How my offer compares: where I am stronger, weaker or unclear
E) 3 concrete actions for me, one line each
Answer in [Armenian / English].

Research these competitors:
1. [name] [link]
2. [name] [link]
3. [name] [link]
Focus on: [for example: course prices / package contents / delivery time].
Follow the report format from your instructions.

I am planning to [launch a new offer / change my price / enter a new segment]: [describe it in 2 to 3 sentences, with your planned price].
Based on the competitors you researched in this chat and about_my_business.md:
1. Where does this price sit in the market: low, middle or high? Show the comparison with links.
2. What do I need to include or say so the price feels fair to the client?
3. What is the biggest risk?
If you need data you do not have, search for it first and tell me what you searched.

03One week with this system

043 common mistakes

1. Trusting it without checking. AI can misread a promotion or an old page. Open the link before you make a pricing decision.
2. Ten competitors at once. You get a shallow table. Three or four is plenty.
3. Not saying what decision you are making. Without a Focus line, Claude collects everything and helps with nothing.

05Done when

• The Project has instructions and an about_my_business.md file.
• Web search is on.
• You got your first report on 3 competitors.
• You opened and checked the price links.
• You made one concrete decision based on the report.

What it is. A Project where content moves through fixed stops: topic → 5 hooks → script → article and caption. At every stop, you choose and you approve.

Who needs it. Anyone creating content for their business who is tired of explaining their voice to AI every time.

01Build it step by step

Here are 5 of my best posts: [paste them]
Analyze them and write voice.md: my tone, sentence length, words I often use, words I never use, how I open and how I close, 3 example first lines.
Do not copy the posts. Describe the patterns.

02Prompts

You are my content producer. My voice and examples are in voice.md in this project. My content database is the Notion database "[database name]" with these fields: Topic, Status, Hook, Script, Article, Caption.

We work in stages. Never skip a stage and never move to the next one without my explicit approval.
Stage 1, Hooks: for a topic, write 5 hooks, each from a different angle: a mistake, a result, a contrarian take, a short story, a how to. Max 15 words each.
Stage 2, Script: from the hook I choose, write a short video script for [Instagram Reels / TikTok], [30 to 60] seconds, spoken language, short sentences.
Stage 3, Repurpose: from the script I approve, write an article and a post caption with a clear call to action.

Rules:
1. Write like voice.md, not like a marketer. No hype words.
2. Never invent facts, numbers or client stories. If an example is needed, ask me for a real one.
3. After each stage I approve, update the Notion entry: fill the field and change Status.
4. Language: [Armenian / English].

New topic: [topic]
Who it is for: [audience]
What I want them to do after watching: [follow / comment / buy / book a call]
Create the Notion entry with Status = Idea, then do Stage 1.

I choose hook [number]. Changes: [optional edits].
Do Stage 2. Mark the moments where I should show something on screen in [brackets].

Script approved [with these edits: ...].
Do Stage 3:
1) Article: [600 to 900] words, same idea with more depth, one practical example from me: [your real example].
2) Caption: hook in the first line, 3 to 5 short lines, call to action: [your CTA].
Then save everything in Notion and set Status = Approved.

03One week with this system

043 common mistakes

1. Approving without reading. The content has your name on it, not Claude's. Did not read it? Do not approve it.
2. No voice file. Without voice.md you get generic marketing copy that sounds nothing like you.
3. Everything in one prompt. "Write hooks, a script and an article" gets you average quality on all three. The power is in the stops.

05Done when

• The Notion database is ready with the right fields.
• The Notion connector is connected.
• voice.md is in the Project.
• One topic made it through every stop.
• Status updates correctly in Notion.

Where to start

Start with system 1. Today.

Here is why: your cofounder's files feed the other two. The researcher learns what you sell and at what price from goals.md. The content pipeline learns how you talk from how_i_work.md. You write about yourself once, and three systems use it.

Then pick by pain. If pricing and sales keep you up at night, build system 2. If content is falling behind, build system 3.