GH-300 Skills measured as of 07 Aug 2026
AI / ML August 22, 2026 8 min read

GH-300 Exam Guide 2026: GitHub Copilot Certification Domains and Prep

GitHub’s Copilot certification was revised on 7 August 2026 and now tests Copilot CLI, Agent Mode and MCP. Here are the real domain weights — including why you will see two different domain counts online.

GH-300 GitHub Copilot certification exam guide and domain weights

GH-300 is the GitHub Copilot certification — a GitHub credential delivered through Microsoft's certification programme, with the official study guide hosted on Microsoft Learn. It was revised on 7 August 2026, and the revision matters: Copilot CLI, Agent Mode, MCP and Sub-Agents are now named objectives.

If you prepared before August, read the companion brief on what changed in the August 2026 update before you sit it.

What GH-300 certifies

The audience profile is unusually direct about scope. You are expected to have expertise in using GitHub Copilot to improve productivity, quality and security — covering responsible AI use, prompt engineering, Copilot features across the various plans, and privacy safeguards. You also need GitHub fundamentals and experience with at least one programming language.

Note what is not there: no requirement to write production code under exam conditions. GH-300 tests your judgement about an AI tool — when to trust it, how to steer it, what it does with your data, and how to govern it across an organisation.

Who this exam is really for

It is aimed at developers, but two of the six domains — responsible AI, and privacy and safeguards — plus the org-policy content inside the features domain are squarely administrator territory. Team leads and platform engineers rolling Copilot out to an organisation are arguably the best-served audience.

Exam format at a glance

Exam codeGH-300
LevelIntermediate
Passing score700 / 1000
Domains6
Maintained byGitHub
Skills updated07 Aug 2026

Microsoft does not publish a fixed question count or duration on the study guide itself; widely cited figures are around 65 questions in roughly 100 minutes, and around USD $99, but these are third-party numbers rather than official ones. Treat the Pearson VUE booking page for your region as authoritative on both time and price.

What is official: a score of 700 or greater passes, there is a free exam sandbox for previewing the interface, and localised versions of the exam are updated roughly eight weeks after the English version — so if you sit it in another language shortly after an update, you may be tested on the previous objective set.

The six domains and their weights

Use GitHub Copilot features25–30%
Use GitHub Copilot responsibly15–20%
Understand Copilot data and architecture10–15%
Apply prompt engineering and context crafting10–15%
Improve developer productivity with Copilot10–15%
Configure privacy, content exclusions and safeguards10–15%

One domain dominates and five are roughly equal. That shape should drive your study allocation directly.

Use GitHub Copilot features (25–30%)

The largest domain by a clear margin, and the one the August update expanded most. Four clusters:

  • Copilot in the IDE — enabling it, and triggering it through inline suggestions, chat, CLI and agent mode. Content exclusions for specific files or repositories also appear here.
  • Copilot CLI — what it is, installing it, key commands, using it interactively and in sessions, generating scripts and managing files. This is now a full sub-objective in its own right.
  • Features and capabilities — Agent Mode, Copilot Edits, MCP, Agent Sessions, delegating to Sub-Agents, code review, Spaces, Spark, pull request summaries, instructions files, and Copilot Chat's limits, options and commands including prompt file reuse.
  • Organisation-wide settings and policies — policy management, Copilot Code Review policies, feature availability across IDEs and github.com, audit log events, and managing subscriptions via the REST API.
Where the difficulty concentrates

That third cluster is the densest list of named features anywhere in the exam, and most of it is recent. If your Copilot experience is inline suggestions and chat, this is where you will lose marks. Spend disproportionate time on Agent Mode, Sub-Agents, MCP, Spaces and instructions files.

Use GitHub Copilot responsibly (15–20%)

The second-largest domain, and the one developers most often dismiss. Two sub-objectives: responsible AI principles — the risks and limitations of generative tools, ethical use, and identifying potential harms with mitigations — and validating and operating AI tools, meaning why AI output must be validated and what responsible operation looks like in practice.

Do not skim this. At 15 to 20 percent it outweighs prompt engineering, and the questions reward considered judgement rather than recall.

Understand Copilot data and architecture (10–15%)

How data is used, flows and is shared; how input is processed and prompts are built; proxy filtering and post-processing; the code-suggestion lifecycle; and the limitations of large language models and of Copilot itself. This is the domain that explains why the privacy controls in the last domain exist, so study the two together.

Apply prompt engineering and context crafting (10–15%)

Prompt structure and context, how Copilot determines context, zero-shot and few-shot prompting, and best practices — plus prompt engineering principles, prompt process flow and how chat history is used.

Smaller than most candidates expect. Useful daily, but only 10 to 15 percent of the paper.

Improve developer productivity (10–15%)

Code generation, refactoring and documentation; accelerating learning and reducing context switching; generating sample data and modernising legacy code. Then the testing and security half: generating unit and integration tests, identifying edge cases, writing assertions, and suggesting security and performance improvements.

Configure privacy, content exclusions and safeguards (10–15%)

Content exclusions and editor settings, ownership and limitations of outputs, enabling the filter for suggestions matching public code, and resolving issues with suggestions and content exclusions. Troubleshooting content exclusions is a recurring theme — know what an exclusion does and does not prevent, and how long changes take to take effect.

Why some sites list seven domains

You will find reputable-looking pages claiming GH-300 has seven domains. Here is what is going on.

Microsoft's “skills at a glance” summary list on the study guide contains an apparent duplicate: Use GitHub Copilot features (25–30%) is immediately followed by GitHub Copilot features (25–30%). Scrapers and summarisers copy the bullet list and report seven.

The arithmetic settles it

Six domains give a minimum total of 80% and a maximum of 110% — a range that brackets 100% correctly. Seven domains, counting the 25–30% entry twice, would give 105% to 140%. That is impossible. The detailed body of the study guide also has exactly six headed domain sections.

Practically this changes nothing about what you study — the duplicated entry has no unique objectives under it. But it is a useful signal about prep material: a site that lists seven domains has copied the summary list without reading the study guide.

Where candidates lose points

TrapWhy it catches peopleFix
Treating it as a coding examStrong developers over-index on productivity and under-study responsible AI and privacyThose two domains are 25–35% combined — budget accordingly
Stale feature knowledgeAgent Mode, Sub-Agents, MCP, Spaces and Copilot CLI are recent additionsRead the August 2026 changes brief
Confusing plans and featuresFeature availability differs across Copilot plans and across IDEs versus github.comBuild a small matrix of feature versus plan versus surface
Guessing at data flowThe architecture domain asks about proxy filtering and post-processing specificallyLearn the suggestion lifecycle end to end, in order
Assuming exclusions are absoluteContent exclusion behaviour and its limits are explicitly testableKnow what exclusions do not cover, and the propagation delay

A three-week study plan

Assumes you already use Copilot regularly. If you do not, add a week of simply using it daily first — several objectives are much easier to recall once you have felt the behaviour.

Week 1
The big domain. Work through every named feature in “Use GitHub Copilot features” and actually try the ones you have not: Copilot CLI, Agent Mode, Copilot Edits, Agent Sessions, Sub-Agents, MCP, Spaces, instructions files, prompt files. Then the organisation-side content: policies, Copilot Code Review policies, audit log events, subscription management by REST API.
Week 2
Responsible AI, data and privacy together. Responsible AI principles, harms and mitigations, and why validation is non-negotiable. Then the architecture domain — data flow, prompt building, proxy filtering, post-processing, the suggestion lifecycle — and finish with content exclusions, editor settings, public-code matching and output ownership. These three domains reinforce each other.
Week 3
Prompting, productivity, rehearsal. Prompt structure, context determination, zero-shot and few-shot, chat history. Then productivity: tests, edge cases, assertions, refactoring, legacy modernisation. Spend the back half of the week on timed practice questions and the exam sandbox.
Exam day
Read the qualifier. A high share of GH-300 questions turn on a single word — which plan, which surface, which mode. When two options both look correct, one of them is almost always right for a different plan or a different IDE.

If you want the schedule sized against your own availability, the study-time calculator will do it, and the certification roadmap shows how GH-300 sits alongside the Microsoft AI credentials.

Frequently Asked Questions

How many domains does the GH-300 exam have?

Six. Microsoft's at-a-glance list appears to show seven because one entry is duplicated, but the detailed study guide has six headed domains, and seven weighted ranges could not sum to 100 percent.

Is GH-300 a GitHub or a Microsoft certification?

Both, in effect. The certification is GitHub's and is maintained by GitHub, but it is delivered through Microsoft's certification programme and the official study guide lives on Microsoft Learn.

Do I need to be a strong programmer to pass GH-300?

You need working familiarity with at least one programming language and GitHub fundamentals, but the exam tests how you use and govern Copilot rather than your ability to write algorithms.

How long is the GH-300 certification valid?

GitHub Copilot certification is generally cited as valid for two years. Confirm the current validity on your Microsoft Learn transcript, since certification policies are revised periodically.

Practise GH-300 With Real Explanations

Timed practice questions across responsible AI, Copilot features, prompt engineering and privacy controls.

Browse All Exams
ExamCert

ExamCert Team

Certified IT professionals tracking the cloud, AI, and security certification landscape. Every exam brief is rebuilt against the official skills-measured document on the date shown above.

Practice Before You Book

500–1,000+ practice questions per exam with detailed explanations, across Azure, AWS, GCP, security, and AI certifications.