every time you open claude or chatGPT, you start from zero. no memory. no context. no idea who you are, what your business does, or how you talk.

then you spend the first 5 minutes of every conversation explaining the same things: your company, your audience, your tone, your rules. you do this dozens of times per week. hundreds per month.

the CLAUDE.md file fixes this.

it's a configuration file that claude code loads automatically at the start of every session. it tells claude everything about your project before you type a single word. your business context, your rules, your preferences, your workflows. every time.

anthropic's official documentation calls it "the single highest-leverage file in your development workflow." buildcamp's guide says it's "the highest-leverage file in your entire workflow" because it affects every session, every task, and every output.

most guides cover CLAUDE.md for developers. code style. test commands. build scripts. that's useful if you're an engineer.

this guide is for business operators. the people using AI to write content, manage clients, automate workflows, and run operations. you don't need to write code to use a CLAUDE.md file. you need to know what to put in it.

what a CLAUDE.md file actually does

dometrain's engineering guide explains the core function: "claude code excels at understanding general programming principles, but it might lack context for your business domain. project-specific jargon, obscure entity names, and acronyms confuse the agent."

the same applies to business context. claude writes great content. but it doesn't know your brand voice, your pricing, your client base, your competitors, or your internal terminology unless you tell it.

the CLAUDE.md file is where you tell it. once.

when you start a session, claude reads the file automatically and incorporates the context into every response. no prompting required. no "act as my marketing manager" preamble. it already knows.

builder.io's guide warns that "context is precious. every line in your CLAUDE.md competes for attention with the actual work you're asking claude to do." so you keep it tight. specific. actionable.

how to structure a CLAUDE.md for business operations

buildcamp recommends keeping the file under 300 lines. "shorter is better. some high-performing teams keep theirs under 60 lines."

here's the structure we use for client projects. not code projects. business operations.

section 1: company context (5-10 lines)

who you are. what you sell. who you sell to. one paragraph. specific.

example:

## company context
[company name] is a commercial insurance agency in tampa, FL.
we serve contractors, property managers, and small fleet operators.
annual premium volume: $8M. team of 12. using applied epic as AMS.
main carriers: hartford, travelers, progressive commercial

## company context
[company name] is a commercial insurance agency in tampa, FL.
we serve contractors, property managers, and small fleet operators.
annual premium volume: $8M. team of 12. using applied epic as AMS.
main carriers: hartford, travelers, progressive commercial

## company context
[company name] is a commercial insurance agency in tampa, FL.
we serve contractors, property managers, and small fleet operators.
annual premium volume: $8M. team of 12. using applied epic as AMS.
main carriers: hartford, travelers, progressive commercial

this changes every output. claude references your actual business, not generic "your company" language.

section 2: voice and tone (5-10 lines)

how you talk. how you write. words you use. words you avoid.

example:

## voice
write as a knowledgeable insurance advisor, not a sales rep.
tone: professional, direct, no fluff. explain coverage in plain english.
never say: "reach out," "touch base," "circle back," "synergy."
always use: specific coverage types, dollar amounts, real scenarios.
address the reader as "you." reference specific industries we serve

## voice
write as a knowledgeable insurance advisor, not a sales rep.
tone: professional, direct, no fluff. explain coverage in plain english.
never say: "reach out," "touch base," "circle back," "synergy."
always use: specific coverage types, dollar amounts, real scenarios.
address the reader as "you." reference specific industries we serve

## voice
write as a knowledgeable insurance advisor, not a sales rep.
tone: professional, direct, no fluff. explain coverage in plain english.
never say: "reach out," "touch base," "circle back," "synergy."
always use: specific coverage types, dollar amounts, real scenarios.
address the reader as "you." reference specific industries we serve

section 3: pricing and offers (5-10 lines)

what you charge. what's included. what's not. so claude never makes up numbers or misquotes your services.

## pricing
commercial package policies: $2,500-$15,000/year depending on class.
workers comp: quoted per payroll. minimum premium $2,500.
we don't write personal lines. if someone asks, refer to [partner agency].
all quotes require a completed acord application before binding

## pricing
commercial package policies: $2,500-$15,000/year depending on class.
workers comp: quoted per payroll. minimum premium $2,500.
we don't write personal lines. if someone asks, refer to [partner agency].
all quotes require a completed acord application before binding

## pricing
commercial package policies: $2,500-$15,000/year depending on class.
workers comp: quoted per payroll. minimum premium $2,500.
we don't write personal lines. if someone asks, refer to [partner agency].
all quotes require a completed acord application before binding

section 4: workflows and rules (10-20 lines)

the processes claude should follow. the rules it should enforce. the steps it should reference when helping with tasks.

## workflows
new business submission: collect acord, loss runs (5 year), photos.
renewal process: start 120 days out. pull file, review, contact insured.
certificates: verify coverage, check additional insured requirements, generate.
claims: collect FNOL details, confirm coverage, submit to carrier, notify insured.

## rules
never promise coverage without underwriter approval.
always include "coverage is subject to policy terms and conditions."
quote ranges are estimates only, not binding

## workflows
new business submission: collect acord, loss runs (5 year), photos.
renewal process: start 120 days out. pull file, review, contact insured.
certificates: verify coverage, check additional insured requirements, generate.
claims: collect FNOL details, confirm coverage, submit to carrier, notify insured.

## rules
never promise coverage without underwriter approval.
always include "coverage is subject to policy terms and conditions."
quote ranges are estimates only, not binding

## workflows
new business submission: collect acord, loss runs (5 year), photos.
renewal process: start 120 days out. pull file, review, contact insured.
certificates: verify coverage, check additional insured requirements, generate.
claims: collect FNOL details, confirm coverage, submit to carrier, notify insured.

## rules
never promise coverage without underwriter approval.
always include "coverage is subject to policy terms and conditions."
quote ranges are estimates only, not binding

section 5: reference files (5 lines)

CLAUDE.md supports importing other files with the @path/to/file syntax. point claude to your existing documentation instead of duplicating it.

## references
see @docs/carrier-appetites.md for current carrier guidelines.
see @docs/class-codes.md for common contractor class codes.
see @templates/renewal-letter.md for standard renewal letter format.

## references
see @docs/carrier-appetites.md for current carrier guidelines.
see @docs/class-codes.md for common contractor class codes.
see @templates/renewal-letter.md for standard renewal letter format.

## references
see @docs/carrier-appetites.md for current carrier guidelines.
see @docs/class-codes.md for common contractor class codes.
see @templates/renewal-letter.md for standard renewal letter format.

why this matters more than prompting

prompting is one-shot. you write a great prompt, get a great result, and the context is gone next session.

CLAUDE.md is persistent. anthropic's documentation confirms that "your CLAUDE.md file becomes part of claude's system prompt. every conversation starts with this context already loaded, eliminating the need to explain basic project information repeatedly."

the practical difference:

without CLAUDE.md: "write a renewal letter for my insurance agency. we're in tampa. we serve contractors. use a professional tone. don't be salesy. mention that we've been their broker for 3 years. include their policy number XYZ-123..."

with CLAUDE.md: "write a renewal letter for johnson construction. policy expires april 15."

claude already knows your agency, your tone, your template format, your standard language. the output is better because the context is richer and more consistent than any prompt you'd write from memory.

how to set it up in 15 minutes

step 1: install claude code: npm install -g @anthropic-ai/claude-code

step 2: navigate to your project directory and run /init. claude generates a starter file based on what it finds. delete everything generic. keep what's useful.

step 3: add your business sections. company context, voice, pricing, workflows, rules. use the structure above.

step 4: test it. start a new session and ask claude to do something business-specific. check if it references your context without being prompted.

step 5: iterate. builder.io recommends updating CLAUDE.md whenever you find yourself correcting claude or repeating instructions. "the mistake won't repeat."

you don't need to be technical. the file is plain text. write it in markdown like you'd write notes for a new employee.

the compound effect

codewithmukesh's tutorial puts it simply: "create a CLAUDE.md file. this is the single biggest improvement you can make."

the compound effect is real. week one, you save 5 minutes per session on context. by week four, every output references your business correctly, uses your voice, follows your rules, and produces work that's 80% ready on the first pass instead of 40%.

multiply that across a team. every person using claude gets the same context. the same voice. the same rules. no training required. the file is the training.

we build CLAUDE.md files as part of every client engagement. your AI should know your business as well as your best employee. this is how you make that happen.

book a call → agentintegrator.io