Build Your Own AI Work Assistant: The Complete Guide to Customizing ChatGPT GPTs & Claude Projects

Build Your Own AI Work Assistant: The Complete Guide to Customizing ChatGPT GPTs & Claude Projects

Do you explain the same context, request the same tone, and paste the same template every time you start a recurring task? Set up a custom AI assistant once, and all of that repetition disappears. This guide walks you step by step through building your own work assistant with ChatGPT’s GPTs (formerly Custom GPTs) and Anthropic’s Claude Projects, and sharing it with your team.

Why You Need a Custom AI Assistant

Standard AI Chat vs. Custom AI Assistant

In one real case, a B2B sales team built a GPT for drafting proposals, and the time to write each proposal dropped from an average of 2 hours to 35 minutes.

Part 1: Build Your Own Assistant with ChatGPT GPTs

Before You Start

• A ChatGPT Plus or Team subscription ($20 or $25 per user per month)

• Gather the documents, templates, and guidelines you use repeatedly at work

• Define the role and tone of voice you want your assistant to have

Step 1: Open the GPT Builder

1. Log in at chat.openai.com.

2. Click “Explore GPTs” at the top of the left sidebar.

3. Click “+ Create” in the upper right.

4. You’ll see two tabs: “Create” and “Configure.” Select the “Configure” tab.

💡 The “Create” tab sets up your GPT through conversation, but for fine-grained control we strongly recommend configuring it directly in the “Configure” tab.

Step 2: Set the Basic Information

In the Configure tab, fill in the following fields.

• Name: Enter a clear name your teammates will understand right away.

  • Examples: B2B Proposal Drafter, Weekly Report Writer, Customer Email Reply Helper

• Description: Summarize what this GPT does in one line.

  • Example: A custom assistant that drafts tailored proposals for IT solutions sales teams

• Profile image: Generate one automatically with DALL·E, or upload your team logo. Click the pencil icon on the right → select “Generate with DALL·E” or “Upload photo.”

Step 3: Write the System Instructions — The Key Step

The "Instructions" text box defines your GPT's brain. Whatever you enter here controls the default behavior of every conversation.

5 components of effective instructions:

1. Role definition — "You are a ..."

2. Target user — "The user is a ..."

3. Core behavior rules — "Always ... / Never ..."

4. Output format — "Write in the ... format"

5. Tone and manner — "Use a ... tone"

Here is an example of instructions for a proposal-writing GPT built for a B2B sales team:

[Instructions example — B2B Proposal Drafter]

You are a senior proposal-writing specialist on a B2B sales team for IT solutions.

User: B2B sales reps at Korean IT companies (3–7 years of experience)

Core rules:

• When the user provides the client name, industry, and key needs, write a draft that follows the uploaded proposal template.

• Every proposal follows this structure: ① Client situation analysis → ② Problem definition → ③ Solution overview → ④ Expected benefits (with quantitative figures) → ⑤ Investment cost → ⑥ Implementation timeline

• Expected benefits must include at least one quantitative figure: ROI, cost reduction rate, or percentage of time saved.

• Use a professional yet warm business tone in Korean. Avoid excessive jargon and write at a level that is easy for client executives to read.

• Never use language that directly disparages competitors.

• Do not invent uncertain figures or facts. Mark anything you don't know as "[Needs verification]".

💡 The more specific your instructions, the better. Instead of "write it well," say something clear like "write it in a 6-section structure, include quantitative figures, and use a formal tone." Instructions of 500–1,500 characters work best.

Step 4: Upload Knowledge Files

Click the "Upload files" button under the "Knowledge" section.

Recommended files to upload:

• Samples of your best existing proposals (PDF/DOCX) — 2–3 files

• Company profile and solution brochures

• Proposal-writing guidelines

• A document summarizing pain points by industry for your most frequent clients

• Price list or quoting guidelines

Upload limits: up to 512 MB per file and up to 20 files in total.

💡 Give files clear names such as proposal_sample_manufacturing_2024.pdf, and the GPT will grasp the context more accurately. Avoid names like document(1).pdf.

Step 5: Set Conversation Starters

You can set up to 4 clickable, button-style prompts that users see when they open your GPT.

Examples:

• Write a cloud migration proposal for a manufacturing client

• Convert my last proposal into a financial-sector version

• Write a follow-up email after a client meeting

• Summarize how our solution stands out ahead of a competitive pitch

Step 6: Configure Additional Capabilities

In the "Capabilities" section, toggle the features you need:

• Web Browsing: Enable it when the GPT needs to reference a client's latest news

• DALL·E image generation: Turn on if your proposals need diagrams

• Code Interpreter: Turn on if you need data analysis or charts

💡 Turn off features you don’t need. The fewer the features, the more your GPT stays focused on the core task.

Step 7: Save and Share

1. Click “Create” or “Update” in the top right.

2. Choose who can access it:

   • Only me: Personal use

   • Anyone with a link: For sharing with your team ← Most recommended

   • Public: Publish to the GPT Store

3. Click “Save” and you’re done.

Before/After:

• Before: Five team members each write proposals in their own style → big quality gaps, and a senior has to revise every one

• After: Everyone shares the same GPT → draft quality 80%+ more consistent, revision time down 60%

Part 2: Build Your Own Assistant with Claude Projects

What Are Claude Projects?

Claude Projects is a feature of Anthropic’s Claude that gives you a workspace where context is fixed per project and reference materials are always loaded when you chat. It is similar to GPTs, but stands out for its longer context (200K tokens) and precise, document-grounded answers.

Prerequisite: Claude Pro subscription ($20/month)

Step 1: Create a Project

1. Log in to claude.ai.

2. Find and click “Projects” in the left sidebar.

3. Click the “+ Create Project” button.

4. Enter a project name.

   • Examples: Weekly Sales Report Writer, Customer Q&A Assistant, Meeting Notes Organizer

5. Optionally, add a project description.

Step 2: Add Materials to Project Knowledge

Add materials in the Knowledge panel on the right side of the project screen.

How to add:

• Click “Add Content” → choose “Upload Files”: upload PDF, TXT, CSV, DOCX, and more

• Click “Add Content” → choose “Add Text”: paste text directly

Recommended materials to upload (weekly report writer example):

• 3–5 samples of strong past weekly reports

• Report template and writing guidelines

• Team KPI definitions

• This quarter’s goals and pipeline status

💡 Claude Projects excels at handling large documents. Even a proposal of 50+ pages or a customer list of 100+ rows can go into project knowledge, and Claude will draw on the whole thing when it answers.

Step 3: Set Custom Instructions

On the project screen, click the “Set custom instructions” area (or the gear icon).

Write project-specific instructions like the following:

[Example instructions — Weekly Sales Report Writer]

You are a specialist report writer for the sales planning team at an IT solutions company.

Role: You take the weekly activity notes a sales rep provides (number of meetings, key deal status, issues, etc.) and write a weekly sales report that follows the uploaded report template.

Output rules:

• Report structure: ① Key results this week (3-line summary) → ② Pipeline status (table) → ③ Major deal updates → ④ Issues & risks → ⑤ Next week's plan

• Show every figure's week-over-week change in parentheses. Example: 12 meetings (+3)

• Tone: concise and fact-based. No emotional language.

• Keep each section to 5 lines at most.

• Mark uncertain information with "⚠️ Needs verification."

Step 4: Start a conversation inside the project

1. Type your message in the chat box at the bottom of the project screen.

2. Every file you uploaded to the project knowledge and your custom instructions are applied automatically.

Real-world prompt example:

Here's this week's activity summary. Turn it into a weekly report.

• 8 new meetings (3 manufacturing, 2 finance, 3 retail)

• Company A PoC in progress; results to be presented next week

• Bank B second meeting done; awaiting confirmation of budget approval

• Retailer C project lost (competitor pricing issue)

• Total pipeline: KRW 1.25 billion (≈ US$935,200) (-KRW 120 million (≈ US$89,800) week over week)

💡 In Claude Projects, you can hold multiple conversations within the same project, and every conversation shares the same knowledge and instructions. Write the report on Monday, revise it on Wednesday — all in the same project.

Step 5: Share with your team

1. Click the share icon (person with a plus sign) at the top of the project screen.

2. Enter your teammates' email addresses and set their permissions:

   • Can view: read-only (can use the project but not change it)

   • Can edit: can modify knowledge files and instructions

3. Click "Share" and an invitation is sent to your teammates.

Note: Team sharing is fully supported on the Claude Team plan ($30/month/user) and above.

Part 3: A Collection of Task-Specific Assistant Recipes

Recipe 1: Customer email reply assistant

Use case: Quickly draft replies to customer inquiries, complaints, and follow-up emails

[Key instructions]

You are a customer success (CS) manager at a B2B SaaS company. You write replies to customer emails. Keep the tone polite yet warm, and focus on solving the problem. Every reply follows this structure: ① Thanks/empathy → ② Status update → ③ Solution → ④ Next steps. Keep each reply concise, at roughly 150–250 characters.

Result: Writing time per email drops from 15 minutes to 3 minutes, saving about 4 hours a day at 20 emails daily

Recipe 2: Meeting notes → action item converter

Use case: Turn post-meeting recording transcripts or notes into structured meeting minutes

[Example prompt]

Below are my notes from today's client meeting. Please organize them into meeting minutes.

Attendees: Director Kim (our side), Manager Lee (Company A), Assistant Manager Park (Company A)

• Company A currently uses a legacy ERP and is considering a move to the cloud

• Budget is around KRW 200 million (≈ US$149,600); they want to decide within this year

• Many questions about security certifications (ISO 27001, CSAP)

• They requested a technical validation meeting next week

Result: Time to write up minutes drops from 30 minutes to 5 minutes, and action items are missed almost 0% of the time

Recipe 3: Competitor Analysis Research Assistant

Use case: Preparing competitive positioning materials before a sales pitch

Knowledge files to upload: your solution spec sheets, competitor comparison tables, and past win/loss analysis documents

Result: Competitive analysis prep time 2 hours → 20 minutes

Part 4: Operations and Optimization Tips

5 Practical Tips to Boost Performance

1. Update your instructions every two weeks. Fold in the problems you run into and any new requirements as you use the assistant.

2. Always include "don't" rules. Negative rules like "no disparaging competitors," "no fabricated figures," and "no profanity" raise quality significantly.

3. Put metadata in your file names. Include industry, year, and outcome in the file name, as in Proposal_Finance_2024_BankA_Won.pdf, and the AI picks up context more easily.

4. Keep GPTs and Claude Projects separate by purpose.

   • ChatGPT GPTs: Quick short-form drafts, emails, and simple brainstorming

   • Claude Projects: Long-document analysis, complex reports, and work that draws on large reference materials

5. Set up a team feedback channel. Create a #ai-assistant-feedback channel in Slack or Teams where teammates share what works and what doesn't, and your instructions improve three times faster.

💡 The most important principle: don't try to build a perfect assistant from the start. Build it to about 70% quickly, then improve it by 10% every week. That works far better.

Final Checklist

After you build your GPT or Project, run through the items below:

• Is the role clearly defined?

• Is the output format (structure, length, tone) specified in detail?

• Does it include at least three "don't" rules?

• Have you uploaded at least three reference files?

• Are the file names clear and descriptive?

• Do the conversation starters reflect real work scenarios?

• Have you shared it with at least one teammate and gotten feedback?

• Is an instruction update scheduled on your calendar within two weeks?

Expected Benefits at a Glance

Average daily time saved: 45–90 minutes → about 15–30 hours a month, or roughly 180–360 hours a year

That time turns into more client meetings, deeper relationships, and better results. Build your first AI assistant today. 30 minutes is all it takes.