Table of Contents
It’s the fifth time this week you’ve typed “shorter, and no bullet points.”
Claude gave you 400 words, a header, six bullets, and a friendly little sign-off. You asked for two sentences. You correct it, Claude adjusts, and the next chat starts from zero again.
This is exactly what Claude custom instructions are for. You write your preferences once, and Claude applies them from the first message of every new conversation. No more retyping “keep it short” like a mantra.
The catch is that most people either skip the setting or fill it with “be helpful and concise,” which changes almost nothing. Specific instructions change everything.
Here’s what to put in them, what to keep out, and what to do when Claude still doesn’t listen. You can have yours set up in about ten minutes.
Where to Find Claude Custom Instructions in Settings
Here’s the confusing part. Claude doesn’t have a feature with that exact name.
Anthropic splits the idea into three tools: profile preferences that apply to every chat, project instructions that apply inside one project (paid plans only), and styles that change how responses are formatted and delivered. When someone searches for Claude custom instructions, they almost always want the first one.
To find it, click your initials in the lower left corner, open Settings, and look for the box asking what preferences Claude should consider in responses. The label shifts now and then, so search the page for “preferences” if you can’t spot it. Anthropic’s own explainer covers all three tools.
Save your text, then open a fresh chat to test it.
What to Put in Claude Custom Instructions
Think of it as the briefing you’d give a new freelancer on day one. You wouldn’t hand them your whole business plan. You’d tell them who you are, who you write for, and what drives you up the wall.
Four things do most of the work.
Who you are. One sentence about what you actually do. “I run an Etsy shop selling printable planners” beats “I’m a small business owner,” because Claude can now pick the right examples and vocabulary without being told.
Who’s reading. Claude writes differently for a busy parent on a phone than for a procurement manager at a desk. Say which one you’re writing for.
Your defaults. This is where “shorter, no bullets” finally lives. Use numbers, because “150 words unless I ask for more” can’t be misread and “keep it brief” can.
Your pet peeves. Name the exact phrases that make you wince. If Claude keeps opening with “Great question!”, tell it to stop.
Here’s a full set that fits in under 100 words:
I run a small Etsy shop selling printable planners to busy parents.
My customers skim on their phones, so write in short paragraphs.
Default to 150 words or fewer unless I ask for more.
Use plain text. No headers, and no bullet points unless I ask for a list.
Do not open with praise for my question. Do not use the words
"delve," "leverage," or "unlock."
If my request is missing key details, ask one question before answering.That last line matters more than it looks. It stops Claude from guessing when your request is thin, which saves you a round of “no, that’s not what I meant.”
Paste it in, swap your details, and you’ve done most of the work.
What Not to Put in Your Claude Custom Instructions
The most common mistake is treating the box like a junk drawer. Everything you might ever want goes in, and then nothing works well.
Task rules. “Write product descriptions in a warm tone” feels helpful, but it also applies the day you ask Claude to debug a spreadsheet formula. Task-specific guidance belongs in the chat or inside a project.
Whole documents. Pasting a 2,000-word brand guide buries the lines that matter. Put the guide in a project and keep the preferences box for rules that hold everywhere.
Rules that fight each other. “Keep it brief” plus “explain everything thoroughly” hands Claude a coin flip, and you won’t know which side it landed on.
Anything secret. No passwords, payment details, or client names covered by an NDA. This box is for how you like to work, not for anything you’d hate to see leaked.
Why Claude Ignores Your Custom Instructions (and How to Fix It)
Custom instructions are guidance, not a hard switch. Claude usually follows them, but a few things make it drift.
Your prompt overrides them. Say your preferences ban bullets, then you type “give me ten ideas as a list.” You’ll get a list, because the request in front of Claude weighs more than a standing preference.
The wording is vague. “Be professional” means something different to a lawyer than to a lifestyle blogger. Try “Write like a support email from a small business: warm, direct, no exclamation marks.”
The box is too long. Past a few hundred words, your most important rule starts competing with your minor ones. Cut anything that isn’t true in every chat, and put the rule you care about most on the first line.
If your trouble is uneven quality rather than ignored settings, that’s a different problem with a different fix. It’s covered in why Claude prompts give inconsistent results.
Custom instructions tell Claude who you are and how you like answers delivered. They don’t build the task itself, and that’s where most of the remaining time gets lost. If you’re tired of rebuilding that structure every time, JSON Prompt Mastery for Claude gives you 70+ pre-built structured prompts organized by business workflow, so you run the prompt and get the output without building anything from scratch.
How to Test Your Claude Custom Instructions in Ten Minutes
Don’t trust the first draft. Run it against real work.
Open a new chat and send three prompts you use every week: a customer email, a product description, a summary of something long. Read each answer next to what you’d have written yourself.
Wherever Claude missed, fix that one line. Wrong length? Add a number. Wrong tone? Add a sentence you’d actually write. Change one thing at a time so you know which edit worked.
Then leave it alone. Revisit it when your work changes, like a new client or a new product line, and skim it once a month either way.
The Takeaway
Claude custom instructions work when they’re short, specific, and true in every chat: who you are, who’s reading, your defaults, and what to avoid. Everything else belongs in the prompt or in a project.
Once that’s set, the next gain comes from the prompt itself. A structured task gets a more reliable answer than a loose request.