GH-300: Exam Guide
How the Exam Wants You to Think
The GH-300 exam is for Developers, DevOps Engineers, and Tech Managers using GitHub Copilot. It values Responsible Use, Effective Prompting, and Human Oversight.
Answer Philosophy
- Human-in-the-Loop Always: If an answer suggests skipping human review of AI output, it's wrong. Always validate.
- Context is King: When two prompt answers are similar, pick the one with more specific context — file reference, persona, or example.
- Privacy by Design: When data privacy is in question, favor the more restrictive, privacy-safe option.
Keyword Detection Table
| If you see... | Look for this in the answer... |
|---|---|
| "AI produces wrong information" | Hallucination / validate output |
| "Copilot uses my project's code" | Context / grounding via open files |
| "Automate a multi-step task" | Agent Mode |
| "Edit across multiple files" | Edit Mode |
| "Run a CLI command with AI help" | GitHub Copilot CLI |
| "Reuse prompt across chats" | Prompt files (.github/copilot-instructions.md) |
| "Suggest code from my own docs" | Knowledge Base (Copilot Enterprise) |
| "What did users do with Copilot" | Audit log events |
| "Block specific files from AI" | Content exclusions |
| "Code that looks like training data" | Duplication detection |
| "Vulnerable code suggestion" | Security warnings (filters) |
| "Context from connected services" | MCP (Model Context Protocol) |
| "Public vs business vs enterprise" | Copilot plans comparison |
| "Behavior changes per model" | Temperature / model selection |
Exam Traps
Look out for these!
- Copilot Individual vs Business vs Enterprise: Individual has no org management. Business adds org-wide policy, audit logs, and content exclusions. Enterprise adds Knowledge Bases, Copilot Chat in GitHub.com, and custom models. Don't confuse what's available per plan.
- Inline suggestions vs Chat vs CLI: Inline = triggered by typing in IDE. Chat = conversational interface. CLI =
gh copilot explain/gh copilot suggestin terminal. Each has a distinct interaction model. - Agent Mode vs Edit Mode: Agent Mode autonomously runs multi-step tasks (including terminal commands). Edit Mode lets you review and selectively accept cross-file changes. They are not interchangeable.
- Content exclusions ≠ privacy guarantees: Content exclusions prevent Copilot from using specific files as context. They don't prevent users from pasting that content into chat manually.
- Duplication detection: When enabled, Copilot filters out suggestions that match code in public GitHub repositories — it does NOT prevent all code reuse, only public-repo matches.
- Copilot doesn't train on your code: Microsoft does not use your organization's code to train the foundation Copilot models. A common exam distractor claims it does.
Decision Quick Reference
"Which Copilot plan?"
Personal use, no org policies needed → Copilot Individual
Org-wide policies, audit logs, content exclusions → Copilot Business
Knowledge Bases, GitHub.com chat, custom models → Copilot Enterprise"Which Copilot mode/interface?"
Write code inline as you type → Inline suggestions (IDE)
Ask questions, iterate on code → Copilot Chat (IDE or GitHub.com)
Multi-step autonomous task → Agent Mode
Refactor across multiple files → Edit Mode
Terminal/CLI help → GitHub Copilot CLI"How to improve the suggestion quality?"
Bad output → Add more context (open files, persona, examples)
Hallucination → Validate output; add references/docs as context
Irrelevant suggestion → Use few-shot prompting (provide examples)
Context from external service → Use MCP connector"What should be done first?"
Always: Validate AI output before using in production
Never: Blindly ship Copilot-generated code without reviewFinal Strategy
- Domains 1 and 2 together are 40–50% of the exam — master responsible use and all Copilot features/modes.
- Know the plan differences cold: Individual vs Business vs Enterprise is a frequently tested decision matrix.
- Think like a developer, not a manager: This exam puts you in the seat of a practitioner actively using Copilot — not a policy decision-maker.