Telling Claude to "be concise" barely works. It shortens the sentences and keeps all the same waffle or word-salad. What actually works is telling it exactly what to stop doing: no intro, no recap, just answer the question. Save that once in your settings and every chat does it. The rule below is the one to paste.
To stop Claude writing an essay for every question, paste a brevity rule into the Instructions for Claude box in your settings so it applies to every chat. The rule tells it to answer first, drop the intro and the recap, keep the substance, and only go long when you ask. In Claude Code, set the Concise output style with /config.
1. Right now, tell it the shape you want
The quickest fix costs one line. Before or after your question, tell Claude the shape of the answer: "answer in three sentences", "just the list, no intro", or "give me the answer first, then stop". A hard shape lands far better than "be concise", which Claude reads as slightly shorter sentences while the padding stays. This sorts one message. The next step sorts every message.
2. Make it stick: one rule in your settings
The Instructions for Claude box applies to every conversation you start, so a rule written here is the version that truly sticks, on every plan. Click your initials at the bottom-left, open Settings, go to Profile, and find Instructions for Claude. Paste the rule below and save.
Answer first, in the fewest words that fully cover it. Never cut the meaning or the explanation to be short. Say the same thing with less padding. Cut the narration, keep the substance. Do not restate my question, announce your plan, or recap each step you took. Give me the outcome, the decision, and anything I have to act on. No warm-up like "Great question", "You're right" or "Let me help with that", and no summary at the end of what you just said. Short by default: a simple question gets one to three sentences of plain prose. Use a heading, a table or a bullet list only when it carries real structure, never as decoration. Say it plainly. Skip the hedging and the boilerplate. Mention a caveat only when it changes what I should do next. If a yes or no answers it, say yes or no first, then one line of why at most. Never trade being right for being short. Keep full error messages, warnings, and anything I need to act on safely. When I actually want the long version, I will say "full version" or "explain", and then you can go as deep as it needs.
This works because naming the exact habits to drop beats a vague "be concise". "Be concise" reads as a mood, while "no warm-up, answer first, no summary" is a set of rules it can actually follow.
3. In Claude Code, use the Concise output style
If you work in Claude Code, the terminal tool, it has a built-in Concise output style that leads with the result and skips the preamble and narration, while still keeping full error reports and safety warnings. Run /config, choose Output style, and pick Concise. Then start a new session, because the style loads at startup and a mid-chat switch does nothing until you do. It needs Claude Code v2.1.237 or later.
Claude has also had a built-in Concise response style in the chat app, but where you switch it on has been moving around and may now sit under Skills, so check your own app rather than trusting a set of clicks. The settings rule above is the part that reliably works either way.
If you use ChatGPT instead
Same idea, different menu. Open Customize ChatGPT from your profile, and in the traits box ask for something like "efficient and direct, no preamble", or paste the brevity rule above straight into its custom instructions. Set once, and it carries across your chats the same way.
The catch worth knowing
Short is not always better. Pushed too hard, a concise answer will quietly drop the caveat, the risk, or the reason behind a suggestion, and you feel finished sooner than you should. That is exactly why the rule keeps an escape hatch, so when a decision or a piece of code actually matters, type "full version" and let it open back up. Brevity for the everyday, depth on demand. In a very long chat it can drift back to its old habits, so a quick "shorter" pulls it straight back.
A few quick questions
Why is Claude so wordy in the first place?
Researchers and practitioners point to how it was trained: people rating its answers tended to score longer, more thorough replies as more helpful, so it learned to over-explain. It is a habit picked up in training, not something you broke, and you can turn it down.
Will making Claude concise make its answers worse?
It can, if you push too hard. A very short answer sometimes drops the caveat or the reason that actually mattered. That is why the rule above keeps an escape hatch: tell it "full version" when you want the depth back, and it opens up again.
Does this work on ChatGPT too?
Yes. It has its own version in the settings. Open Customize ChatGPT and set the traits to something like efficient and direct, or paste the same brevity rule into its custom instructions box. The idea is identical: tell it once, and it stays.