Claude AI Projects Tutorial: How to Build Your Own Custom AI Assistant (2026)

Most people treat Claude like a search engine. Ask a question, get an answer, close the tab. I did the same for the first few months — until I started using Claude Projects with my clients and realized this feature alone changes what AI can actually do for a business. If you want a genuine custom AI assistant that knows your business context, works from your own documents, and follows your specific rules every session, this Claude AI projects tutorial walks you through everything you need to set one up in 2026.

What Are Claude Projects — and Why They’re Different from Regular Chats

A regular Claude conversation resets when you close the tab. Every new session starts blank. You re-explain your context, re-paste your prompts, and rebuild from scratch. For one-off questions, that’s fine. For repeat business work, it’s a daily time drain that adds up fast.

Claude Projects, available on Claude Pro ($20/month) and Claude Team ($25/user/month), gives you a persistent workspace. You write custom instructions once. You upload your reference documents once. Claude reads both every time you open that project — no re-pasting, no re-explaining, no context setup at the start of every session.

Here’s what a Claude Project actually gives you:

• A custom instructions field that defines Claude’s behavior and rules for this specific use case
• File uploads — PDFs, Word documents, spreadsheets, and text files that Claude treats as a knowledge base
• A shared conversation history Claude can reference within the project
• Complete separation between projects so your GHL work doesn’t bleed into your real estate research

I currently run six active Claude Projects. When I open my ‘GHL Onboarding’ project, Claude already knows my clients are real estate agents in Dubai, my standard onboarding workflow, and the FAQ document I’ve built over three years of teaching. That context is just there — every time, without setup.

Action item: Log into Claude.ai and check whether Projects appears in your left sidebar. If you’re on the free plan, you’ll need Claude Pro to access this feature.

Setting Up Your First Claude Project in 5 Minutes

Getting started is faster than most people expect. Here’s the exact process I walk my students through.

Step 1: Create the Project

In Claude.ai, click ‘Projects’ in the left sidebar, then click ‘New Project.’ Give it a specific, descriptive name. I use a format like ‘GHL Onboarding — UAE Clients — Apr 2026’ so I know exactly what it’s for at a glance. Add a brief description; this helps Claude understand the scope even before you write your full instructions.

Step 2: Write Your Custom Instructions

This is where most users underinvest. The custom instructions field is where you define who Claude is inside this project. Don’t write ‘help me with marketing.’ Write a real context block with your industry, client profile, standards, and what you don’t want. My templates are in the next section.

Step 3: Upload Your Knowledge Base

Click ‘Add content’ inside the project. You can upload PDFs, Word files (.docx), plain text (.txt), CSV files, or paste content directly. Claude handles files up to 10MB each, with a combined project limit of roughly 200,000 tokens — around 300 to 400 pages of text. For most business use cases, that’s more than enough.

Step 4: Test With a Real Question

Click ‘Start chat’ inside the project. Claude immediately has access to your instructions and uploaded files. Ask something that requires using your documents. You’ll feel the difference from a blank chat within the first response — it’s a fundamentally different experience.

Total setup time: under 10 minutes for a basic project. For a well-built project with tested instructions and organized documents: 30 to 45 minutes once, then zero setup time for every session after that.

Action item: Create one project today. Start with the task you repeat most often in your business.

Writing Custom Instructions That Actually Work: My Templates for Business Owners

Vague instructions produce vague results. ‘Be helpful and professional’ tells Claude almost nothing specific. The quality of your custom instructions determines roughly 80% of your project’s output quality — this is where to invest your setup time.

Here’s the framework I teach in my courses:

1. Role definition — Who is Claude in this project? What’s the specific function?
2. Business context — Industry, location, client type, typical goals and problems
3. Tone and format rules — Response length, formatting preferences, language to use or avoid
4. Document usage guidelines — How Claude should use uploaded files and whether to cite sources
5. Explicit off-limits — Topics to decline, tools not to recommend, what not to share

Here’s a real example from my active GoHighLevel client support project:

‘You are a GoHighLevel support assistant for real estate agents in the UAE. My clients are primarily Dubai-based property agents using GHL for CRM, lead nurturing, and marketing automation. When answering questions, reference the uploaded SOP documents first before falling back on general GHL knowledge. Keep answers under 200 words unless technical detail requires more. If a question falls outside GHL scope, say so clearly and suggest the user contact Sawan directly. Never recommend competing CRM platforms.’

That instruction names the geography, the tool, the response length standard, the citation rule, and a hard limit. Instructions like this produce dramatically more useful output than a generic system prompt ever would.

For anyone building GoHighLevel workflows for clients, you can paste your complete workflow documentation into the project so Claude understands your exact automation setup when answering support questions.

Action item: Write your custom instructions using the five-point framework. Invest at least 20 minutes — it pays back in every session going forward.

Use Case 1: A GoHighLevel Client Onboarding Assistant

This is the project I use most actively. When I bring on a new client for GHL training, there’s a detailed onboarding process: account setup, sub-account creation, snapshot import, pipeline configuration. It’s easy to get wrong and generates a high volume of support questions.

I built a project specifically for this. I uploaded my full onboarding SOP — a 47-page PDF — my FAQ document with over 80 common questions, and a reference sheet of my standard pipeline names and stages. The custom instructions tell Claude to treat these documents as the primary source of truth before falling back on general GHL knowledge.

When a new client messages me with ‘how do I set up the calendar integration,’ I open the project, paste their question, and Claude gives a precise step-by-step answer that matches my exact configuration — not generic GHL documentation, but my specific workflow and setup.

One of my recent clients, a property developer from Abu Dhabi, completed his entire account setup independently using answers from this assistant. That saved me roughly four hours of WhatsApp back-and-forth in a single week. For multiple clients across a month, that time saving compounds significantly.

Action item: If you use GHL with clients, export your SOPs and FAQ documents to PDF and build this project. Even a basic version reduces your support load meaningfully.

Use Case 2: A Real Estate Market Research Assistant

Dubai’s property market moves fast. Prices in certain communities shift within a quarter. My real estate agent clients need current data, and most don’t have time to read multiple market reports every week before client meetings.

My market research project gets regularly updated uploads: Dubai Land Department data, CBRE quarterly reports, and community-specific pricing sheets. The custom instructions tell Claude to always present data in table format, cite the specific document it’s drawing from, and flag any information older than 90 days as potentially outdated.

When an agent asks ‘what’s the average price per square foot in Jumeirah Village Circle right now,’ Claude pulls from the uploaded reports and delivers a structured answer with source attribution. It only works with what I’ve uploaded — but for meeting prep, it’s dramatically faster than manually searching through PDFs.

One of my Dubai clients reduced his pre-meeting research prep from 45 minutes to under 10. Across a full working week, that reclaimed time is significant.

If you’re comparing Claude’s document analysis strength against other AI tools for research-heavy work, my 2026 AI chatbot comparison covers the differences in detail — Claude’s long-document handling gives it a clear practical edge in this specific context.

Action item: Identify the 3 to 5 documents you reference most often and upload them to a dedicated research project this week.

Use Case 3: A Content Repurposing Assistant for YouTube-to-Blog

I publish YouTube tutorials regularly on AI tools and GHL. Turning those into blog posts used to take me two to three hours per video. Now it takes about 20 minutes.

My content project has instructions specifying my writing style: first-person, practitioner tone, short paragraphs, UAE-focused audience, and a required CTA linking to sawankr.com courses. I also uploaded three of my best-performing blog posts as style references — real examples of how I write, not just a description of it.

When I want to repurpose a video, I paste in the transcript and ask Claude to draft a full post following the style guide already stored in the project. Because the instructions are permanently there, I don’t explain anything each time. Just paste the transcript and request the draft.

The output always needs editing. But I’m starting from a structured first draft with the right voice and format, not a blank page. One specific result: I repurposed a 22-minute GHL automation tutorial into a 1,800-word blog post in 18 minutes. That post now ranks for three keywords and drives consistent traffic to my courses page.

Action item: Upload two to three of your best existing content pieces as style references. This single step improves output quality more than any prompt adjustment.

How to Upload Files and Train Your Project on Your Own Documents

Claude Projects support PDF, Word (.docx), plain text (.txt), CSV, and Markdown files. A few practical notes from using these regularly across multiple active projects:

• PDFs with complex tables sometimes lose formatting during parsing. For critical data tables, paste the content as plain text instead of relying on the PDF version.
• Large PDFs over 200 pages are better split into focused sections. Claude handles targeted, chunked documents better than one massive combined file.
• CSV files are the best format for spreadsheet data. Claude reads them cleanly and can analyze the data within them effectively.
• Version control is manual. Delete the old version before uploading the updated one — otherwise Claude may draw from both versions simultaneously.

Think of project files like a briefing pack for a new employee. Include what they need to do the job well. Don’t dump everything you’ve ever written. Keep it focused on what’s directly relevant to this project’s function.

Action item: After two weeks with a new project, audit the uploaded files. Remove anything outdated or off-topic — it reduces noise and improves response accuracy.

Managing Multiple Projects: My Organization System

I run six active Claude Projects across different business functions. Without a system, managing them becomes disorganized fast. Here’s what I use.

I apply a three-part naming convention: [Function] — [Audience or Context] — [Month Updated]. For example: ‘GHL Support — UAE Agents — Apr 2026’ and ‘Content — YouTube Repurpose — Apr 2026.’ The date tells me immediately whether a project needs a refresh without opening it.

Each project has a short plain-text README uploaded as the first file. It describes what the project is for, lists what other documents are uploaded, and notes when each was last updated. When I return to a project after a few weeks away, I re-orient in seconds.

I do a quarterly review of every project. Market data goes stale. SOPs get revised. Uploaded documents that were accurate in January may be misleading by April. A quarterly review keeps projects accurate rather than quietly drifting toward outdated information.

For a broader view of how Claude fits into my overall AI tool workflow alongside other platforms, my breakdown of ChatGPT Plus vs Pro covers the tradeoffs that inform which tasks I route to which tool.

Action item: Name your projects with the three-part convention from day one. Retroactive reorganization is time you don’t need to spend.

Claude Projects vs ChatGPT Custom GPTs: Honest Comparison

I use both tools. Here’s an honest side-by-side for business owners deciding where to invest their setup time:

My honest take: for document-heavy work — SOPs, market reports, internal knowledge bases — Claude Projects performs more consistently in my experience. The context handling on long documents is noticeably more accurate and reliable. For tasks requiring real-time web data or external client-facing sharing, ChatGPT has practical advantages Claude does not currently match.

For my UAE-based clients building internal support tools and knowledge bases, I recommend starting with Claude Projects. The instruction-following is tighter and the document analysis is more reliable for the type of work most of them are doing.

Action item: Run the same document-based query in both a Claude Project and a Custom GPT with identical instructions. For long-document tasks, the quality difference is often immediately apparent.

Getting Started: Build Your First Custom AI Assistant This Week

This Claude AI projects tutorial covers the complete process — from creating your first project to managing a full system of custom assistants across different business functions. You don’t need a developer or any technical background. You need clarity about what repetitive work you want to stop doing manually and what documents Claude needs to replace that work well.

Start with one project. Pick the task you repeat most. Build it properly — focused documents and specific custom instructions. Use it for two weeks. The habit forms quickly once you experience the difference between a properly configured project and a blank chat that forgets everything each session.

If you want structured, hands-on guidance building this for your specific business — GoHighLevel agency, UAE real estate operation, or content workflow — I cover Claude Projects in depth in my AI tools course at sawankr.com/courses, alongside GHL automation, Canva workflows, and content creation systems built specifically for the UAE market.