Set up Claude in one weekend: the C.A.S.T method

If Claude has ever felt average to you, it is almost never the model. It is that you opened a brand new chat and asked a stranger to read your mind. Claude knows nothing about you yet: not your voice, not your audience, not your work. So it gives you the average of the internet, because that is all it has to go on. Spend one weekend fixing that and the same Claude suddenly feels twice as smart. Same model, completely different output.

🧭 The C.A.S.T method at a glance

Four moves, in order. Skip one and you are back to typing prompts into a stranger forever.

1. 🔵 C, configure. Turn on Claude's settings and connect your apps, so it can actually do things instead of just talking.
2. 🟡 A, architect. Build a Project and load your knowledge into it, so Claude remembers who you are across every chat.
3. 🟢 S, specialize. Save your repeated tasks as Skills, so you stop re-explaining the same process every week.
4. 🌱 T, target. Train it on your niche, your voice, and your audience, so everything it writes sounds like you.

🗂️ First, the 3 places Claude actually works

Most people only ever use the first one, and all three are on every plan, including free.

The one most people skip is Projects, and it is the one that changes everything. That is step A.

🔵 C is for configure (about an hour)

The boring-but-critical part: switch on Claude's capabilities so it can do more than chat. Exact paths, current as of this weekend:

1. ⚙️ Open claude.ai, click your initials (bottom left), then Settings, then Capabilities.
2. 🧠 Turn on "Generate memory from chat history". This is free on every plan, and it is how Claude starts remembering your role, projects, and preferences. (The related "Search and reference chats" toggle is the paid one, that is the split people confuse.)
3. 🛠 Make sure "Code execution and file creation" is on. It usually is by default, and it is what lets Claude make real files (spreadsheets, decks, PDFs).
4. 🌐 Web search is a per-chat toggle: click the slider icon in the chat input box and switch Web search on when you want live answers.
5. 🔌 Connectors live under Customize in the left sidebar (not Settings), then the Connectors tab. This is how Claude plugs into Google Drive, Gmail, Notion, Canva and more. You click connect, log in, approve, done. No code.
6. ✍️ While you are in Settings, fill in "Instructions for Claude": two or three lines about who you are and how you like answers. It applies to every conversation.

🟡 A is for architect (build your Project)

This is the magic step. A Project is Claude's long-term memory of you. Go to claude.ai/projects, click + New Project, then build it in four moves:

1. 🏷️ Name it. Pick a clear focus: "content engine", "client X", "my brand". One Project per job. (Small gotcha: Claude cannot see the name or description, they are for you. Anything Claude should know goes in instructions or knowledge.)
2. 📋 Instruct it. Click Set project instructions and write the standing brief: Claude's role, your tone, the output format you want, what to avoid. Every chat inside the Project follows it.
3. 📤 Upload your knowledge. The step that does the magic. Use the + on the knowledge panel and feed the brain: your brand voice doc, audience notes, content pillars, your best past posts, transcripts, style guides. Files up to 30MB each, and the more real context, the more it sounds like you.
4. 🧪 Test it. Run one real task inside the Project, tweak the instructions, ship. Thirty minutes of setup, hours saved every month after.

Copy-paste starter for your Project instructions:

You are my content and marketing partner. You know my brand inside out.

ABOUT ME: [one line on who you are and what you do]
MY AUDIENCE: [who you make content for, their main pain point]
MY VOICE: [3 to 5 words, e.g. warm, plain-spoken, no jargon, a little funny]
MY CONTENT PILLARS: [the 3 to 5 topics you post about]

RULES:
- Write in my voice using the docs I have uploaded as the source of truth.
- Keep it clear and direct. Explain anything technical the way a sharp friend would.
- No long dashes. Use commas, periods, or brackets instead.
- When you are missing context, ask me one sharp question instead of guessing.
- Always give me the actual draft, not a description of what you would write.

Official walkthrough if you want it: what are Projects.

🟢 S is for specialize (save your repeated tasks as Skills)

If you have explained the same process to Claude twice, that is a Skill waiting to happen. A Skill is a saved set of instructions Claude loads automatically whenever the task comes up. You will find them under Customize, then Skills (the Add button lets you create one with Claude, write the instructions yourself, or upload a skill file).

The easiest way to make one is to just talk it through, Claude formats it into a proper Skill for you. Worked example for captions:

Help me build a Skill for writing my Instagram captions.

When I give you a topic or a finished reel, write a caption that:
- Opens with a scroll-stopping first line, all lowercase
- Stays casual and friendly, like texting a friend
- Uses 2 to 3 emojis, never more
- Ends with one clear call to action
- Never uses long dashes

Ask me for the topic, then give me the caption. Save this as a Skill
called "caption writer" so I can reuse it.

Two more Skills worth making in your setup weekend: a weekly content report (paste your numbers, get the same structured recap every Sunday) and a brand brief formatter (turns a messy sponsor email into a clean brief every time). One tip: the short description on a Skill is how Claude decides when to use it, so make it specific ("writes Instagram captions in my voice") rather than vague ("helps with content").

Official guides: what are Skills and create a Skill through conversation.

🌱 T is for target (train it on your niche and voice)

The polish pass that separates "good AI writing" from "that sounds exactly like me." One honest framing first: you are not training the model. You are building Claude a permanent briefing pack: who you are (instructions), how you sound (style), who you serve (project knowledge), and what it should remember (memory). That is good news, it takes a weekend, not a data science team.

1. 📝 Feed it your voice, not a description of it. Do not tell Claude "I'm casual and warm." Show it. Run the voice-profile prompt below on 3 to 5 of your best pieces, and also try Claude's Styles feature (in any chat: the style menu, then "Create & edit styles", then "Add a Writing Example"), it builds a reusable style straight from your writing.
2. 👥 Give it your real audience. Run the audience-brief prompt below, save the result into your Project knowledge, and every future chat knows exactly who you serve.
3. 🔁 Correct it like a new hire. When the tone is off, be specific ("I'd never say leverage, I'd say use") and ask it to remember that. Specific corrections stick, vague sighs don't.
4. 🎯 Lock your format. If your reels follow a structure or your carousels are always 7 slides, add it to the Project instructions once so it never forgets.

The voice-profile prompt (paste with your samples):

I'm going to paste 3-5 pieces of my writing below. Your job is to reverse-engineer my voice into a reusable style guide, not to compliment the writing.

Analyze only patterns that appear in at least 2 samples (ignore one-off quirks so you don't turn my voice into a caricature). Cover: sentence length and rhythm, tone, vocabulary I favor, words and phrases I would never use, punctuation habits, how I open and close pieces, paragraph length, and how I use humor if at all.

Then output "MY VOICE GUIDE" in under 400 words, structured as:
1. Voice in one sentence
2. Always do (5-7 rules, each with a short example from my samples)
3. Never do (5-7 rules, including banned words and phrases)
4. Rhythm notes (sentence and paragraph patterns)
5. Sample line: one sentence written in my voice on a topic I haven't covered

The guide must end with this line verbatim: "Match these patterns at their natural frequency. Do not exaggerate them."

After the guide, rewrite this paragraph in my voice so I can check calibration: [PASTE ANY GENERIC PARAGRAPH]

Here are my samples:
[PASTE 3-5 OF YOUR BEST PIECES]

The audience-brief prompt (run inside your Project):

You're helping me build an audience brief that will live in this project's knowledge, so every future chat understands exactly who I serve without me re-explaining.

Interview me first: one question at a time, 8-10 questions total, covering my niche, who my audience is, what they've already tried, what they're afraid of, what they want most, how sophisticated they are on this topic, and what I offer them. Push back if my answers are vague ("everyone interested in AI" is not an audience).

Then produce "AUDIENCE BRIEF" with these sections:
1. Niche one-liner (who I help + what outcome)
2. Primary audience: situation and sophistication level
3. Top 5 pain points, ranked, in their own words
4. Top 3 goals and what achieving them looks like
5. Objections and false beliefs to address
6. Vocabulary: terms to use, terms to avoid
7. Content no-gos (topics, framings, and tones that would break trust)

Write it so a stranger could create on-brand content for me from this document alone. When I approve it, I'll save it to project knowledge. Start with question 1.

⏱️ Your one-weekend blueprint

1. 🔵 Hours 1 to 4, configure. Turn on the capabilities above and connect the two or three apps you live in.
2. 🟡 Hours 5 to 10, architect. Build one Project, write your instructions, upload your knowledge, run one test task.
3. 🟢 Hours 11 to 16, specialize. Save 2 Skills for your most repeated tasks (captions, scripts, repurposing, whatever you do weekly).
4. 🌱 Hours 17 to 22, target. Feed it your niche, your voice examples, your audience, and lock your formats.
5. 🟠 Hours 23 to 24, ship. Stress-test it with real work and smooth the rough edges, using this prompt:

Time to stress-test this setup with real work, not a toy example.

Here's an actual task from my week: [PASTE A REAL TASK, e.g. "write the caption for Thursday's post about X"]

Do three things, in this order:
1. Produce the deliverable, using my voice guide, audience brief, and everything else in this project.
2. Audit your own output: list where you applied a specific rule from my setup (quote the rule), and flag every place you had to guess because my setup didn't cover it.
3. Recommend exactly 3 edits to my setup (instructions, project knowledge, or memory) that would have made your first draft better, worded so I can paste them straight in.

Don't soften the audit. Every guess you flag is a hole I can patch this weekend instead of discovering it in three months.

🎓 A few things I wish someone told me first

• 💡 Upload more than you think you should. Your brand voice lives in your actual posts, transcripts and old work, not in one thin doc.
• 🧱 One Project per job. A "content engine" Project and a "client X" Project will both be sharper than one messy catch-all, and each Project keeps its own separate memory.
• 🪄 Skills are for tasks you are sick of re-explaining. Typed the same instructions twice? That is the signal.
• ✍️ Show, never describe. "Match this voice" plus five real examples beats three paragraphs describing your voice.
• 🔧 Treat it like onboarding a teammate, not a magic box. The first outputs will be 80 percent right. Correct it, save the note, and it gets better fast.

✅ The honest bit