The Problem with Using AI Out of the Box
Most people use Claude or ChatGPT the same way every time: open a new chat, re-explain who they are and what they need, get a generic answer, close the tab.
The AI has no memory of your business. It doesn't know your service offering, your pricing, your tone of voice, your usual client questions, or how you prefer to communicate. So every conversation starts from zero, and you spend the first few exchanges re-establishing context that never changes.
Claude Projects solves this. It's a persistent workspace where you upload your business context once — and Claude uses it automatically in every conversation. No re-explaining. No generic outputs. An AI that responds like it's been working with you for months.
Here's how to build one that actually works.
What Claude Projects Does
A Project is a dedicated space in Claude.ai with two components:
Project Knowledge: Files and documents you upload that Claude reads and retains. These inform every conversation in the project. Your services list, pricing, past proposal examples, FAQ document, brand voice guide, client intake form — anything Claude should know about your business.
Custom Instructions: A brief you write telling Claude how to behave in this project. What role it's playing, how it should communicate, what to prioritise, what to avoid.
Together, these turn a general-purpose AI into something that behaves like a well-briefed team member who knows your operation.
Step 1: Gather Your Context Documents
Before you touch Claude, spend 30 minutes pulling together what you'd want a new employee to read on their first day.
The essentials:
Services document. One page covering every service you offer, what's included, what's not, and any key constraints. If you do custom quoting, include your typical pricing ranges. Specificity matters — "website design from $2,500" is more useful to Claude than "we do web design."
FAQ document. Write down the 10–15 questions you get asked most often and your actual answers. These can come from emails, intake forms, DMs — wherever you're repeating yourself. This is usually the highest-ROI document in the whole project.
Tone and voice guide. Three to five sentences describing how you write. Formal or informal? Long or short sentences? Any words or phrases you always use? Any you never use? If you have existing emails or proposals you're proud of, paste 2–3 examples in here.
Standard templates. Your proposal structure, your welcome email, your standard follow-up. These don't need to be polished — rough templates are fine. Claude will use them as a reference for format and content.
Optional but useful: a bio or background doc (who you are, how long you've been operating, what types of clients you work best with), your process document (how you work with clients step by step), any competitor positioning (why clients choose you over alternatives).
Don't try to be exhaustive. Four good documents are better than twelve mediocre ones. Claude works best with clear, specific information — not long, vague company overviews.
Step 2: Write Your Custom Instructions
This is where most people under-invest, and it shows in the outputs.
Your custom instructions tell Claude what it's doing in this project. Think of it like a system prompt for your specific use case. Here's a template to work from:
You are a business assistant for [Your Business Name], a [type of business] based in [location].
Your role: Help draft client communications, proposals, emails, and internal documents using the business context provided.
Tone: [2-3 sentences from your voice guide. E.g. "Direct and friendly. Short sentences. No jargon or corporate language. We talk to clients like they're smart adults, not like they need hand-holding."]
Always:
- Reference the services document when discussing what we offer
- Use the pricing ranges from the services doc, never guess or make up numbers
- Match the format of the templates provided when producing similar documents
- Sign off emails as [Your Name]
Never:
- Promise timelines or deliverables not mentioned in the project documents
- Use phrases like "Certainly!" or "Absolutely!" — just respond directly
- Add filler paragraphs. If the answer is two sentences, write two sentences.
When you don't know something: say so clearly and suggest what information you'd need to answer properly.
Adjust everything to fit your situation. The "Never" section matters more than people think — it's where you prevent the generic AI behaviours that make outputs feel impersonal.
Step 3: Test With Real Tasks
Once you've uploaded your documents and written your instructions, run three tests before you rely on it for anything that goes to a client.
Test 1: A client enquiry response. Paste in a realistic enquiry email and ask Claude to draft a reply. Does it reference your actual services? Does it match your tone? Does the pricing mentioned match your document?
Test 2: A proposal request. Give it a one-paragraph brief for a new project and ask for a proposal outline. Does the format match your template? Is the scope realistic given what you offer?
Test 3: An FAQ question. Ask something from your FAQ document. Does it give your actual answer, or a generic one?
Where it gets things wrong, the fix is almost always in the source documents — not enough specificity, or the instructions aren't clear enough. Update the doc or the instructions and test again.
Step 4: Build the Habit
The project only saves you time if you use it consistently.
The mistake most people make is using the Project for big tasks and going back to a new chat for quick ones. That's backwards. Quick tasks — short emails, one-paragraph summaries, a draft response to a client question — are exactly where the persistent context pays off most. Thirty seconds to get a draft you edit in two minutes, vs. five minutes writing from scratch.
Keep the Project open in a pinned browser tab. When a task comes up that involves writing, drafts, or client communication — go to the Project first.
What This Looks Like in Practice
Here's a real example of the difference.
Without the Project:
"Write an email to a potential client who asked about my social media management service. My rates are around $800–1200/month depending on scope. Keep it warm but professional."
Claude produces something generic. You edit heavily to make it sound like you.
With the Project:
"Draft a reply to this enquiry: [paste email]. They're asking about social media management for a small Brisbane retail shop."
Claude already knows your services, your pricing, your tone, and your typical proposal structure. The draft it produces sounds like you, references your actual service, uses your pricing range, and is formatted the way you'd format it.
You read it, make two small edits, and send.
That's the difference. Not magic — just context that doesn't have to be re-entered every time.
A Note on What to Keep Out
Don't put genuinely sensitive information in Project Knowledge. Passwords, bank details, private client data, anything you'd be uncomfortable with if there was a breach. Claude Projects are reasonably secure but you're still uploading to a cloud service.
For most business contexts — services, pricing, tone, templates — this isn't a concern. But if you work in legal, financial, or healthcare and handle confidential client information, be selective about what goes in.
The goal is a project that knows your business well enough to be useful, not a complete data dump. Start lean and add more as you identify gaps.
The Setup Time vs. the Return
Setting up a useful Claude Project takes 2–3 hours: gathering documents, writing instructions, testing, and refining.
The payback is immediate. Every client email, proposal draft, FAQ response, and internal document you produce through the project is faster and needs less editing than starting from scratch. For anyone doing regular client-facing writing, that's daily value.
Build it once. Use it every day.