Aug 20, 2026 · Free Guide

“CONCISE” Stop Claude Code From Writing Walls of Text

You ask a simple question. You get six paragraphs, a recap of what you just asked, and an offer to help further. Anthropic quietly shipped a fix. When I tested it my replies dropped from 2,328 characters to 1,462. That is 37% less reading for the exact same answer. Here is the whole setup.

The problem is not you, it is the default

Claude Code ships with a Default output style built for one thing: doing software engineering carefully. Careful means explaining. Explaining means length.

That is fine when you are debugging something gnarly. It is exhausting when you asked what a flag does.

Most people try to fix this by begging in every message. "Be brief." "Short answer only." It works for one reply, then it drifts right back. You are fighting the system prompt with a sticky note.

What an output style actually is

An output style rewrites Claude's system prompt. Not a suggestion inside your chat, the actual instructions Claude boots up with.

That is why it sticks and why prompting does not. There are five built in.

01
Default
The stock setting. Built for careful software engineering, which means it explains everything.
02
Concise
Result first. No preamble, no narration, no recap of what you just asked. This is the one you want.
03
Proactive
Acts instead of asking. Makes reasonable calls on routine decisions and keeps moving.
04
Explanatory
Teaches you as it works. Drops insights about why it made each choice.
05
Learning
Makes you write parts of the code yourself. Leaves TODO markers for you to fill in.

Concise is the one you want. Its own instructions literally say conciseness never means withholding requested information. You still get full detail when you ask for it. It just stops padding.

STEP 1

Turn it on everywhere

Do not hunt for the file yourself. Open Claude Code and paste this. It finds your paths, backs up your files, and checks for anything that would override it.

Set my Claude Code output style to "Concise" everywhere so it applies to every
new session, in this project and globally.

Do all of this yourself, do not just tell me the steps:

1. Find my user settings file at ~/.claude/settings.json. Create it as {} if it
   does not exist.
2. Find the project settings files in the current working directory:
   .claude/settings.json and .claude/settings.local.json. Create either one as
   {} if it does not exist.
3. Back up every file you are about to touch by copying it to <name>.bak first.
4. In each of those three files, add or update the top-level key:
   "outputStyle": "Concise"
   Preserve all existing keys exactly as they are. Do not reformat or reorder
   anything else. Do not touch the "permissions" block.
5. Validate each file parses as JSON afterward and tell me if any failed.
6. Check whether anything could override the setting and report what you find:
   - a managed settings file at /Library/Application Support/ClaudeCode/managed-settings.json
     (macOS), /etc/claude-code/managed-settings.json (Linux), or
     C:\ProgramData\ClaudeCode\managed-settings.json (Windows)
   - any installed plugin output style using "force-for-plugin: true"
   - an "outputStyle" key already set in ~/.claude.json
7. Confirm "Concise" is a valid built-in style in my installed version. Run
   claude --version and report it.

Then tell me exactly what you changed and remind me the style only loads at
session start, so I need /clear or a brand new session for it to apply.

Why one file is not enough

You will find guides telling you to just edit the global settings file. That is where I started, and it did not stick. Here is what nobody mentions:

  • Settings load in a stack. Managed policy beats project local, which beats project, which beats user. A stray project file you forgot about silently wins.
  • The style loads once, at session start. Editing the file mid-chat does nothing. You need /clear or a new session.
  • A plugin can override you. Any plugin shipping an output style with a force flag beats your setting completely.
  • Desktop and terminal read different files. The docs point desktop users at the project file, not the global one.

That is why the prompt above writes it in every place at once. You stop guessing which file your setup actually reads.

STEP 2

The instructions that do the heavy lifting

Concise makes Claude shorter. It does not make Claude talk like a person. For that you add a Response Style block to your CLAUDE.md file.

Copy this whole thing and paste it into Claude Code as one message. The instruction is on top, the block it needs to add is underneath.

Append everything below the line to the end of my CLAUDE.md file, exactly as
written. Keep every existing section in that file untouched. Do not reorder or
rewrite anything already in there. When you are done, show me the section you
added so I can confirm it landed.

---

# Response Style

Talk to me like I am smart but not technical. Sixth-grade reading level. Short sentences. Plain words.

## Structure every reply this way

1. The answer. One or two lines. What happened, or what I asked for. Nothing else first.
2. The details. Bullet points only. One idea per bullet. One line per bullet.
3. What I need to do. Only if I actually need to do something. Say it as a direct instruction: "Click X" or "Tell me if you want Y."
4. Also found (optional). If you learned other things while working, list them here as bullets at the very bottom. One line each. Then stop. Do not explain them. Let me ask if I want more.

## Rules

- Never start with a preamble. No "Great question," no "I've gone ahead and," no restating what I asked.
- Never narrate your process. I don't need to know which files you opened or what you tried first.
- No em dashes. Ever.
- One topic per reply. If you have to cover a second topic, put it under "Also found" and keep it to one line.
- No jargon. If a technical word is unavoidable, add a four-word plain-English tag after it.
- Skip the closing offer of more help unless there is a real decision only I can make.
- Do not pad. If the answer is one sentence, send one sentence.

## When you have a question for me

- Ask one question at a time.
- Give me the options as bullets.
- Tell me which one you recommend and why, in one line.

## When something goes wrong

- Say what broke in one line.
- Say what it means for me in one line.
- Say what you want to do next in one line.
- Do not paste error logs unless I ask.

## When I ask for real writing

Long is fine for drafts, scripts, posts, and documents. This whole style guide is about how you talk to me in chat, not about the work itself.

The last section is the part people skip, and it is the most important one. Without it Claude starts writing you two-sentence sales pages. You are setting how it talks to you in chat, not how it does the work.

STEP 3

Prove it actually worked

Close Claude Code. Open it again. Start a brand new chat and paste this:

Is a Concise output style active in your system prompt? Answer yes or no, and
quote the header line if it's there.

If it says yes and quotes "Concise Style Active", you are done.

If it says no, you skipped the restart. The style only loads at session start. There is no way around that one.

All Resources

Output Styles DocsThe official reference for all five styles
Settings & PrecedenceWhich file wins when they disagree
Claude CodeYou need this installed first
The ReelWhere this guide came from

The Next Step

Guides show you what. AI Builders shows you how.

Inside the community, I walk through every build live, including the stuff that doesn't make it into guides. Regular people (not developers) figuring out AI together, shipping real projects, asking me anything. No fluff, no theory, just the actual work.

Join AI Builders

skool.com/ten-fold