Why Custom Instructions Matter
Most people start every Claude conversation from zero. Custom instructions mean Claude already knows how you like to work before you type a single word.
Think about the best colleague you've ever worked with. They knew your communication style. They didn't over-explain things you already understood. They anticipated what you'd need next.
Custom instructions turn Claude into that colleague. You set them once, and every single conversation from that point forward starts with that context baked in.
Without instructions: You get a generic assistant that writes long paragraphs, hedges every answer, and never follows up with what to do next.
With instructions: You get a working partner that matches your pace, speaks your language, and thinks one step ahead of you.
How to Set Them Up
Takes about 60 seconds. Here's where to find it.
In Claude chat (claude.ai):
- 1Click your name in the bottom left corner
- 2Click Settings
- 3Find Personal preferences
- 4Type or paste your instructions and hit Save
That's it. These apply to every new conversation automatically.
If you use Claude Code (the terminal version), add your instructions to your CLAUDE.md file instead. Same idea, different location.
Step-by-Step Coaching Mode
Stop Claude from dumping the entire answer on you at once.
By default, Claude gives you everything in one go. A 15-step process, a full strategy, an entire implementation plan. It's overwhelming and you end up not doing any of it.
This instruction changes that completely.
What changes: Instead of a wall of text, Claude walks you through things like a patient colleague sitting next to you. It checks in. It waits for you. It adapts if something doesn't work at step 3 instead of ploughing ahead to step 12.
Learning new tools, setting up systems, following technical guides, anything where the order of steps matters and you want to actually do the thing instead of just reading about it.
Ask Before Assuming
Claude will fill in the blanks with guesses if you let it. This instruction makes it ask instead.
Ever given Claude a vague brief and gotten back a fully formed answer that was completely wrong for your situation? That's because Claude assumed your industry, your audience, your budget, your timeline. It filled in every gap with its best guess.
What changes: Claude pauses and asks smart questions before diving in. It acts more like a consultant doing discovery than an intern racing to deliver a first draft. The output quality jumps because the input is better.
This pairs perfectly with the step-by-step instruction above. Claude asks the right questions first, then walks you through the execution one step at a time.
The "Take More Off My Plate" Closer
This is the one that changed how I use AI completely.
Most AI conversations end too early. Claude answers your question, you say thanks, conversation over. But the best use of AI isn't answering the question you asked. It's telling you what to do next that you hadn't thought of yet.
What changes: Every response ends with a clear list of what to do next. Not vague suggestions. Specific actions you can take immediately, systems you could build to never do this manually again, and things you should hand off to someone else.
It turns Claude from a question-answering machine into a proactive partner that's always looking for ways to save you time.
I used to close Claude after getting my answer and move on. Now every conversation surfaces things I didn't know I should be doing, automations I didn't think to build, and tasks I was holding onto that someone else could handle. It's like having a chief of staff who never forgets to follow up.
Cut the Fluff
Stop Claude from writing essays when you need a paragraph.
Claude's default mode is thorough, which really means verbose. You ask a simple question and get 400 words of context you already know, caveats you didn't ask for, and a conclusion that restates the opening.
What changes: Claude respects your time. Answers come faster and with less scrolling. You still get depth when you need it, because you can always say "explain more." But the default becomes concise instead of comprehensive.
Challenge Me When I'm Wrong
AI that just agrees with everything you say isn't useful. It's dangerous.
By default, Claude is agreeable. You propose a bad idea and it helps you execute the bad idea enthusiastically. That's not a partner. That's a yes-man.
What changes: Claude starts flagging problems before you commit to them. It suggests alternatives when your first instinct isn't the strongest option. It saves you from the mistakes you'd only catch in hindsight.
This doesn't make Claude argumentative. It makes it useful. There's a difference between pushback that wastes your time and pushback that saves you from a bad call.
Know My Context
Tell Claude who you are so it stops writing for a generic audience.
If Claude doesn't know your role, your industry, or your experience level, every answer is written for a beginner with no context. You waste time skipping past explanations you don't need and translating generic advice into your specific situation.
What changes: Claude tailors every answer to your actual situation. A marketing manager gets different advice from a solo founder. Someone comfortable with automation gets different guidance from someone who's never used AI before.
"I'm a physiotherapist running a 4-person clinic. I'm new to AI but comfortable with technology. My priorities right now are reducing admin time and improving patient follow-up. Skip the basics and focus on what's actionable for a small healthcare practice."
The Full Instruction Set (Ready to Copy)
Here's everything in one block. Copy the whole thing into your Claude settings, then edit the "About me" section to match your situation.
These instructions aren't set in stone. Use them for a week, notice what's working and what isn't, then update them. The best custom instructions are the ones you refine over time as you figure out exactly how you work best with AI.
That's all 6 instructions.
Copy them, paste them into Settings, done. Better AI starts now.
Follow @reallyusefulai on Instagram for more practical AI guides.
More free guides at reallyusefulai.co/guides