1. Overview: 3 Universal Steps

Regardless of whether you use ChatGPT, Claude, Gemini, Copilot, or LeChat, the process is the same. All five tools let you create a dedicated project space with a system instruction and uploaded files. Here are the 3 steps:

1. Create a Project / Space — Give your project a name (e.g., "AI Workshop — Customer Service"). This becomes your permanent workspace where all conversations and files are collected. In ChatGPT and Claude it's called "Projects", in Gemini it's "Gems", and in Copilot and LeChat it's "Agents".

2. Write a System Instruction — Tell the model who it is, how it should respond, and what it should focus on. The more specific you are, the better the answers. Think of it as a "job description" for your AI. Include role, tone, format, and constraints.

3. Upload Context & Data — Attach documents, data, guidelines, or examples. This gives the model a knowledge base to work from — so it responds based on your material, not general knowledge. Most tools support PDF, Word, Excel, and CSV.

> PRO TIP: Use the same system instruction and the same test data across all tools. This way you can make a fair comparison and see which model gives the best answers for your specific task.

> IMPORTANT: Each model has strengths and weaknesses. Set up the same project in at least 2 tools so you can compare and choose the best "AI assistant" for your task.

2. Tips That Make You Faster

Shift + Enter: Finish Writing Before You Send

When you press Enter in the chat window, most LLMs send your message immediately. But if you press Shift + Enter, you create a line break instead — without sending. This lets you write a complete, structured prompt with multiple paragraphs before the model starts working.

Why it matters:

• You can build a complete prompt with context, task, and format in one field.
• The model gets your entire instruction at once — and gives a significantly better answer.
• You avoid sending half-finished messages and using extra messages for corrections.

> Works in ChatGPT, Claude, Gemini, Copilot Chat, and LeChat. On mobile: use the enter button for line breaks.

Meta-Prompting: Let the AI Build Your Instruction

Instead of writing the perfect system instruction from scratch, you can ask the model to do it for you. It's called "meta-prompting" — you prompt the AI to write the prompt.

Example — type this in the chat:

"I want to create a project that helps me with [your task]. Write a system instruction for me that I can copy-paste into the project's instruction field. Include role, tone, format, and constraints."

The model generates a detailed instruction you can use directly. Customize it, test it, and iterate. It works in all 5 tools.

3. Copilot — Agents

copilot.microsoft.com | Model: Microsoft 365

1. Open Copilot Studio / Agent Builder — Go to Copilot Studio or use Agent Builder in Microsoft 365 Copilot. Click "Create new agent".

2. Describe Your Agent — Write a short description (up to 1,024 characters). The AI automatically generates name, instructions, and tool suggestions.

3. Configure Instructions & Knowledge — Refine instructions (up to 8,000 characters). Add knowledge sources: SharePoint sites, websites, or internal documents.

4. Enable Tools & Publish — Enable web search, code interpreter, or other plugins. Publish the agent to Teams, web, or internal use.

> Copilot is strongest when you already use Microsoft 365. Link directly to SharePoint sites as knowledge sources.

4. ChatGPT — Projects

1. Create a New Project — Click "Projects" in the left menu. Click "+ New Project". Give it a descriptive name. All new chats in the project automatically inherit your instructions.

2. Add Project Instructions — Click the three dots at the top and select "Project settings". Write your system instruction: role, tone, format, constraints.

3. Upload Files & Context — Drag files into the project (PDF, Word, Excel, CSV, images). ChatGPT indexes them automatically.

4. Start Conversation & Iterate — Test with a concrete task. Ask for adjustments. ChatGPT remembers context in your session.

5. Claude — Projects

1. Create a New Project — Go to claude.ai and click "Projects" in the menu. Click "+ New Project" and name it. You can create up to 5 projects for free.

2. Write Project Instructions — Click "Set project instructions" above the upload area. Write detailed instructions.

3. Upload Knowledge Base Files — Upload documents, data, and examples. Claude can handle up to 1,000,000 tokens of context — approximately 2,500 pages of text.

4. Use Artifacts for Output — Claude can create "Artifacts" — standalone documents, code, visualizations — that you can edit and export directly.

> Claude is strongest for long, nuanced tasks. Upload entire reports — the model handles 1M tokens. Use Artifacts for structured output.

6. Gemini — Gems

gemini.google.com | Model: 3.1 Pro

1. Create a New Gem — Go to gemini.google.com and click "Explore Gems" in the left menu. Click "New Gem".

2. Write Instructions — Write detailed instructions. Tip: Click "Expand with Gemini" to let the AI help you improve your instructions.

3. Add Files & Knowledge Sources — Upload documents as context. Gems support Google Drive integration.

4. Test & Save — Use the preview panel to test. IMPORTANT: Click "Save" when satisfied — the preview does NOT auto-save.

> Leverage Gemini's massive 2M token window for large amounts of context — 10x more than competitors.

7. LeChat — Agents

1. Create a New Agent — Click "Agents" in the left menu and select "Create Agent". Choose a template or start from scratch.

2. Configure Name & Instructions — Write detailed instructions. LeChat has no character limit on instructions.

3. Add Knowledge Sources & Libraries — Link external data sources (Google Drive, Gmail, SharePoint) or attach your own Libraries with documents.

4. Enable Tools & Share — Enable web search, code interpreter, image generation, or Canvas. Share the agent via link or API.

> LeChat is the only EU-based tool on the list. Perfect for organizations where data locality and GDPR are critical.

8. Quick Comparison

> Use this table as a starting point. Then test the same task in 2-3 tools to see which model performs best.

9. When Should You Use What?

Each model has its strength. Here's a quick guide:

• Writing & long documents → Claude — Claude's 1M context and strong language understanding make it ideal for long reports, analyses, and nuanced written work.
• Research & fact-checking → Gemini — Gemini's 2M token window and native web access make it superior for research and source-based work.
• Quick productivity & all-round → ChatGPT — ChatGPT's broad plugin ecosystem, image generation, and coding make it a strong daily tool.
• Microsoft 365 integration → Copilot — Already have Microsoft 365? Copilot integrates directly with Word, Excel, Teams, and SharePoint.
• GDPR & EU compliance → LeChat — Mistral is EU-based and GDPR-compliant by default. Ideal for organizations with strict data requirements.

> Start here: Choose 2-3 tools. Create the same project in each. Test the same prompt. Compare results. Choose your primary AI assistant — and keep 1-2 as a "second opinion".

---

🎓 Want hands-on training? Book our Copilot Workshop or Copilot Agent Workshop.

📬 Subscribe to the newsletter "AI, Built Human" on Substack — weekly insights on AI in practice.

---

Stefano Vincenti · AI Advisor & Trainer · aitrainer.dk · External Lecturer, IT University of Copenhagen · Cofounder & CTO BotTellMe · Partner, TryZone