Claude Code for non-technical people: the setup, the first prompts and what to put in CLAUDE.md (2026)

Claude Code for non-technical people needs a quick setup, one message that tells it who you are, and a CLAUDE.md file that remembers it. You do not need to know how to code for any of it. The video above sums up the steps in under two minutes.

This guide is for people who have never opened a terminal: founders, designers, marketers, operations people and students who want a rough first version of an idea they can click through. Every Claude Code fact below comes from Anthropic's own documentation, read on October 24, 2026, with links to the pages.

What Claude Code actually does

You point Claude Code at a folder on your computer and talk to it in plain English. It reads the files in that folder, writes new ones, and runs the steps a programmer would run. Opening the page it builds in your browser is one of those steps. Anthropic's overview calls it "an agentic coding tool that reads your codebase, edits files, runs commands". For a beginner, "codebase" just means "the folder".

You can open it in three ways. The Claude desktop app's Code tab is the one to start with. The second is the terminal, the text window on your computer where you type commands. The web version at claude.ai/code works from code stored on GitHub and is better left for later.

It is not free. Anthropic's setup page says Claude Code "requires a Pro, Max, Team, Enterprise, or Console account. The free claude.ai plan does not include Claude Code access." Pro, the cheapest of those, has been enough for our prototypes; the pricing page has the current prices.

Set up Claude Code without the terminal (or with it, in five minutes)

Setup depends on one question: do you want to see a terminal or not? Both routes end in the same place: a session (one conversation with Claude Code) where you type plain English and Claude works in your folder.

The desktop app route (recommended if you have never used a terminal)

1. Install the Claude desktop app and sign in with your paid account.
2. Click the Code tab at the top of the sidebar. Anthropic's desktop quickstart says "you don't need to install Node.js or the CLI to use the Code tab". Those are tools programmers install; you can skip them.
3. Make a new empty folder on your computer, named after your idea, and open it from the Code tab.
4. Pick the permission mode. The app offers Auto, Manual, Accept edits and Plan. Start with Manual: Claude shows you each change and asks before making it. Switch to Accept edits once you trust it. A stop button interrupts Claude at any moment.

The terminal route

A terminal is the text window where you type commands. Anthropic wrote a guide for people who have never used one, which opens with "You can use Claude Code even if you've never used a terminal before."

1. Open the Terminal app on a Mac, or PowerShell on Windows.
2. Paste one line and press Enter. On a Mac:

curl -fsSL https://claude.ai/install.sh | bash

On Windows:

irm https://claude.ai/install.ps1 | iex

1. Go to your project folder: type cd and a space, drag the folder from Finder (or from File Explorer on Windows) into the window so its location appears, and press Enter.
2. Type claude and press Enter, then sign in when your browser opens.
3. Press Shift+Tab until the status bar (the strip at the bottom of the window) shows "manual mode on". One thing to know: since version 2.1.283, Claude Code in the terminal starts in Auto mode. In Auto mode a background check approves routine actions without asking you. Anthropic's permission modes page explains each mode.

One folder, one hello

Whichever route you took, send one of the first prompts Anthropic suggests in its quickstart:

what can Claude Code do?

Then the one from the terminal guide:

make me a simple webpage that says hello world

If Claude asks for permission, say yes. When it is done, double-click the HTML file it made to open the page in your browser, which is what Anthropic's guide says to do. That is your first working page. Next, tell Claude who it is talking to.

The first prompt to send: tell Claude who you are

Claude does not know you are not a programmer. Left alone, it answers you the way it answers an engineer: a list of ways to do everything, and the names of the tools. Anthropic's prompting guidance says to treat Claude "as a brilliant but new employee who lacks context on your norms and workflows". So give it the context, as the first thing you type in a new session:

I am not from the IT industry, explain everything as simple as you can. Don't dig into tech details but simply explain if needed, so I'm not overwhelmed.

What two people and one measured test showed

We first tried that message in two sessions with people who had no coding experience. Both sessions felt calmer to us than earlier tries without it. Two sessions are not enough to go on, so we measured it.

We ran three fresh Claude sessions (Claude Sonnet 5.5) through the same six beginner questions. The questions: where do I start; what is git and do I need it; what does this error mean; how do I put it online for a friend; what is an API key; can the form email me. The only difference was the first message: none, the simple message above, or a longer structured version. Jargon hits are technical words from a fixed list of 90; the reading grade is the US school grade the text is written at.

Claude already writes at about a fifth-grade level without being asked. What the message changed was the amount: roughly half the words, about a quarter of the jargon, no blocks of code to copy. With the simple message, Claude still ended two of the six replies with an open question ("Want to go with that, or would you prefer a different way?"); the structured version asked almost nothing back.

It also showed the catch. Asked about git, the "no tech details" session told the beginner to skip it entirely. The other two sessions recommended it, because git is the undo button for a whole folder. The same session dropped two warnings that the session with no first message gave: that secret keys must never go into the uploaded page, and that an email notification is not a list of subscribers. The structured version kept its own caveats instead, one sentence each: visitors' addresses pass through the form service, and the link is public. "Don't dig into tech details" sometimes removes the safe practice together with the detail.

This is a small test: one model, one run per version, six questions, and only the two in-person sessions involved real people. The sessions were told they were part of a writing test and ran no real commands, so the results are about wording, not about what Claude does with tools. The scores are proxies: a reading formula and a fixed word list measure simplicity, not understanding, so the direction is clear but the exact sizes are not. Studies of language models simplifying medical leaflets found the same trade-off. In a 2025 study Claude was the strongest simplifier and among the least accurate. In a 2023 study simplified texts kept about 80 percent of their key messages, so about one in five was lost.

The structured version

So keep the message, and add the guardrails it loses. This is the version we recommend, and the one that goes into CLAUDE.md:

I'm not a programmer and I'm building a prototype. For this whole session:
1. Use plain words. The first time you use a technical term, explain it in one short sentence.
2. Give me one recommended way to do things, not a list of options.
3. When something must be done in the terminal, do it yourself and tell me what happened, or give me the exact text to paste.
4. Keep answers short. End with the one next step I should take.
5. Ask me before doing anything that costs money, deletes something, or puts something on the internet.

In our test it produced the shortest replies and ended all six with a single next step. It offered to set git up itself, so the beginner would never need to type a git command. It asked for an OK before creating an account or publishing a page. It uses more technical words than the simple message because it names each one once and explains it. That is what you need to read the next error message.

Part of why one sentence works: Anthropic's guidance says explaining "why such behavior is important" lets Claude "generalize from the explanation", and that when you give Claude a role, "Even a single sentence makes a difference". In the simple message, "so I'm not overwhelmed" is that explanation.

What to put in CLAUDE.md so Claude remembers it

Put in CLAUDE.md the things you would otherwise repeat every session: who you are, how to talk to you, what to ask before doing, and what the project is. A CLAUDE.md is a plain text file in your project folder that Claude Code reads at the start of every session. Anthropic's memory page explains that each session "begins with a fresh context window", which means an empty memory. Write it once and you stop pasting the first message.

You do not have to create it by hand. Say "add this to CLAUDE.md" followed by the rule, and Claude edits the file. Two commands exist if you want them: /memory opens the file, and /init drafts one from whatever is already in the folder.

The CLAUDE.md template

The template is 30 lines. Paste it into Claude with the words "create a CLAUDE.md with the following text, exactly", then fill in the blanks under "This project".

# About me and how to work with me

I am not a programmer. I am building a prototype to test an idea, not production software.

## How to talk to me
- Use plain words. The first time you use a technical term, explain it in one short sentence, then just use it.
- Give me one recommended way to do things. Do not list options unless I ask "what are my options".
- Keep answers short. End every answer with the one next step I should take.
- When something must be done in the terminal, do it yourself and tell me what happened. If I must do it, give me the exact text to paste and say where.
- If something fails, tell me in one sentence what went wrong and what you will try next. Do not paste long error logs at me.
- Simple does not mean incomplete. Never drop a warning, a cost or a risk to keep an answer short.

## Ask me first
IMPORTANT: always ask before you:
- delete or overwrite files I made
- spend money or create an account anywhere
- put anything on the internet (deploying, publishing, sharing a link)
- install anything outside this project folder

## How we build
- Small steps. Build one thing, let me look at it, then the next.
- Before a bigger change, write a 3 to 5 line plan in plain words and wait for my "go".
- Save a snapshot (git commit) after every working step and tell me in one line what it saved. Set git up yourself; I never need to type git commands.
- Keep secrets (API keys, passwords) out of the code and out of anything published. Tell me where you put them.

## This project
- What it is: <one sentence>
- Who it is for: <who>
- To see it: <Claude fills this in>
- Done when: <one thing I can check>

Two things about the file. The one word in capitals is deliberate: Anthropic's best practices page says "If Claude keeps skipping one instruction, add emphasis such as 'IMPORTANT' to that line alone. If you emphasize many lines, none of them stands out." And "Ask me first" is advice, not a lock. Anthropic's permissions page puts it this way: "Instructions in your prompt or CLAUDE.md shape what Claude tries to do, but they don't change what Claude Code allows." The permission mode you chose during setup is the lock; the file tells Claude what you care about.

What to leave out

Shorter is better, and Anthropic is blunt about it: "Bloated CLAUDE.md files cause Claude to ignore your actual instructions!" The memory page sets a target of under 200 lines; the template above is 30. Its test for every line: "Would removing this cause Claude to make mistakes?" If not, cut it.

Leave out your business plan, rules you have not needed yet, copies of documentation, and anything secret. Add a line only when Claude makes the same mistake a second time, which is one of the triggers Anthropic's memory page gives. Claude also keeps its own notes, called auto memory, on by default and stored outside your folder. It decides what to keep, so a rule you care about goes in CLAUDE.md where you can see it.

Starter prompts for your first prototype

These are in the order a first project goes. Paste them as they are, filling in anything in angle brackets.

1. Describe the idea and ask for a plan, not code.

I want to build: <one sentence>. It is for: <who>. It is done when: <one thing I can check>. Before you build anything, tell me in 5 plain lines how you would do it and what I will need (accounts, money, anything to install). Then wait for my "go".

2. Build the smallest version that runs.

Build only the first step from your plan. When it works, tell me how to look at it in my browser, then stop and wait.

3. Save and continue.

That works. Save a snapshot and do the next step.

4. When you do not understand the reply.

Explain that again as if I have never used a computer for work. One short paragraph. What do I actually need to do?

5. When something breaks.

Something is wrong: <what you see, in your words, or paste the message>. Tell me in one sentence what happened, fix it if you can, and tell me what you changed.

6. Before anything goes online or costs money.

Before you publish anything, list what will be public, what accounts I need, and whether anything costs money. Then wait for my "go".

7. Teach it a rule mid-session.

Add this to CLAUDE.md: when you show me a link to open, put it on its own line.

8. End of a session.

Write a 5-line summary of what we built today, what is not finished, and the first thing to do next time. Put it at the bottom of CLAUDE.md under "Where we left off".

Prompts 1 to 3 follow the loop Anthropic recommends on its best practices page: explore, plan, implement, commit (look, plan, build, save). Its warning is "Letting Claude jump straight to coding can produce code that solves the wrong problem."

When it gets overwhelming: five habits

Every unexplained word and every choice Claude asks you to make is one more thing to hold in your head. These five habits keep that number down.

1. Start fresh between tasks. Type /clear when you move to something unrelated. Your CLAUDE.md stays; the clutter goes.
2. After two failed corrections, stop correcting. Anthropic's advice: after two corrections on the same thing, run /clear and write a better first prompt.
3. Plan before you build. Press Shift+Tab until the status bar shows "plan mode on", or start a message with /plan. Claude reads and proposes but changes nothing until you agree.
4. Stop and undo. Press Esc to stop Claude mid-sentence; it keeps what it has done so far. Press Esc twice, or type /rewind, to go back to an earlier point. Anthropic's best practices page notes this "isn't a replacement for git", which is why the template asks Claude to take real snapshots.
5. Ask the tool itself. /help lists the commands, /powerup teaches features in short lessons, and /usage shows how much of your plan you have used.

When words are not enough: record your screen and talk

Sooner or later you will see the problem and fail to name it. "The thing on the right is wrong" is a sentence Claude cannot act on. To describe it, you would need the technical words you asked Claude to spare you.

Record it instead. With the Vidmatic Chrome extension you record your screen, circle the part you mean and say what is wrong. Then you hand the recording to Claude Code. Connecting is one pasted line, then /mcp and a browser sign-in, with no key to copy. The screen recording for Claude Code page has the steps and the Vidmatic quickstart walks you through your first recording. Claude gets a timed transcript of what you said plus images of the moments you circled. They arrive over MCP, the standard that Claude Code uses to read outside sources. It does not watch the video; our post Can Claude Code read a video? explains the difference.

Two things to know: circling needs the Chrome extension, and Vidmatic stores the recording. For "that box, that color, that button" problems, a transcript plus pictures is enough.

Start with one small thing

Before you build the idea, get to a page that says hello, with your first message sent and a CLAUDE.md in the folder. After that, every session starts with Claude already knowing how to talk to you, and you can spend the time on the idea. When you are ready for more, our guide to the best MCP servers for Claude Code covers what to plug in next. The screen recording MCP server page shows the one line that connects Vidmatic.