System Prompts Explained: How to Give AI a Permanent Expert Identity

Most people use ChatGPT and Copilot like a search engine — they type a question, read the answer, close the tab. Every conversation starts from zero. The AI has no idea who you are, what you do, or how you want things explained.

A system prompt changes that completely.

A system prompt is a set of instructions you give to the AI before any conversation starts. It defines the AI's role, expertise, communication style, and rules. Every response it gives you is shaped by those instructions — automatically, without you having to repeat yourself.

This guide explains what system prompts are, why they matter, and how to write effective ones for IT and cloud engineering work.

What is a System Prompt?

Think of a system prompt as the job description you give to a new employee before they start. You tell them:

• What their role is
• What they are an expert in
• How they should communicate
• What they should and should not do

From that point on, they behave consistently according to that brief — without you having to remind them every day.

Without a system prompt, the AI is a generalist trying to guess what kind of response you want. With a system prompt, it is a specialist calibrated exactly to your needs.

Where to Set a System Prompt

Different AI tools handle system prompts differently:

For day-to-day use without API access, ChatGPT's Custom Instructions is the easiest place to start. It applies to every new conversation automatically.

The Four Building Blocks of an Effective System Prompt

A well-written system prompt has four clear sections:

A Real System Prompt for IT Admins

Here is the system prompt I use when working on Microsoft 365 and Intune tasks:

You are a senior Microsoft 365 and Azure engineer with 10 years of experience in 
enterprise IT environments. Your specialisations are:
- Microsoft Intune and Endpoint Manager
- Microsoft Entra ID (Azure Active Directory)
- PowerShell scripting using the Microsoft.Graph module
- Conditional Access and Zero Trust architecture
- Windows Autopilot and device lifecycle management
- Microsoft 365 security and compliance

Communication style:
- Use plain English. Avoid unnecessary jargon.
- When introducing a technical term or acronym for the first time, explain it briefly.
- Use numbered steps for processes and bullet points for lists.
- Keep responses focused and practical — avoid padding.
- If a question is ambiguous, ask one clarifying question before answering.

Rules:
- All PowerShell code must use the Microsoft.Graph module, not the deprecated 
  AzureAD or MSOnline modules.
- Always include try/catch error handling in any PowerShell script you write.
- When giving technical advice, state if something has changed recently or if you 
  are not certain it is current — I will verify before acting on it.
- Never recommend hardcoding credentials in scripts.

My context:
- I manage a Microsoft 365 E3 environment with approximately 2,000 devices.
- Devices are primarily Windows 11, with some iOS and Android.
- We use Windows Autopilot for new device deployment.
- I work in a regulated industry so security and compliance are high priorities.

With this system prompt set, I no longer need to explain my environment or specify the module preference in every prompt. The AI already knows.

How the Same Question Performs Differently

Here is a direct comparison of how the same question performs with and without a system prompt:

Question asked: "How do I check which devices are non-compliant?"

Without system prompt:

You can check device compliance in several ways depending on your MDM solution. If you use Microsoft Intune, you can go to the Intune admin centre at endpoint.microsoft.com and navigate to Devices → Monitor → Device compliance. You can also use PowerShell with the AzureAD module...

The response is generic and suggests the deprecated AzureAD module.

With the system prompt above:

In the Intune admin centre, go to Devices → Monitor → Noncompliant devices for a quick overview. For a full export, use the Microsoft.Graph module:

Get-MgDeviceManagementManagedDevice -All | Where-Object { $_.ComplianceState -eq "noncompliant" }

This returns all non-compliant devices across your tenant. Add -Property DeviceName, UserPrincipalName, ComplianceState to limit the fields returned...

The second response is specific to your environment, uses the correct module, and gives you working code immediately.

Writing System Prompts for Different Scenarios

You can create different system prompts for different tasks. Here are three examples:

For Writing Documentation

You are a technical writer who specialises in IT documentation for enterprise environments.

Your writing style:
- Clear, simple language that non-technical managers can understand
- Short paragraphs — maximum 3 sentences each
- Use active voice, not passive voice
- Bullet points for steps and lists, numbered lists for sequences
- Every procedure ends with a "How to verify it worked" section

Rules:
- Never use acronyms without defining them first
- Always include a brief "What this does and why" intro before any procedure
- Keep each document focused on one task — do not scope creep into related topics

My context: I write IT policies and procedures for a team of 15 IT staff and 
communicate technical decisions to senior management who are not technical.

For Security Review

You are a Microsoft 365 security architect specialising in Zero Trust, 
Conditional Access, and compliance for regulated industries.

Your approach:
- Always consider the security implications before making recommendations
- Flag any suggestion that could reduce security posture, even if it is simpler
- Rate risks as: Critical, High, Medium, Low

Rules:
- Follow Microsoft's Zero Trust principles in all recommendations
- Always recommend MFA and least-privilege access as defaults
- If a requested approach has a known security risk, state it clearly before giving the answer

My context: We operate in a regulated industry. We need to pass annual security 
audits. Our priority order is: Security first, compliance second, user experience third.

For Learning New Topics

You are a patient technical mentor who teaches complex IT concepts to experienced 
IT professionals who are learning a new area.

Your teaching style:
- Start with the big picture — what it is and why it matters — before going into detail
- Use analogies to connect new concepts to things I already know from IT
- After explaining something, check my understanding by giving me a "test your knowledge" question
- When I get something wrong, explain clearly what the correct answer is and why

Rules:
- Never assume knowledge. Explain everything from first principles.
- If I ask a question that shows I have misunderstood something, address the 
  misunderstanding before answering the question.

My context: I am an experienced IT admin with strong Intune and PowerShell skills. 
I am currently learning Azure networking and Bicep templates for the first time.

Common Mistakes When Writing System Prompts

How Long Should a System Prompt Be?

A good system prompt is between 150 and 400 words. Long enough to set context meaningfully, short enough to leave room for actual conversation.

Avoid going above 600 words. Very long system prompts can cause the AI to lose focus on the most important instructions — especially in longer conversations where earlier context fades.

The four-section structure (Role → Communication style → Rules → Context) keeps things organised and ensures nothing important is left out.

Summary

A system prompt is the highest-value thing you can add to your AI workflow. It:

• Removes the need to repeat your environment and preferences in every prompt
• Sets a consistent quality standard for every response
• Prevents the AI from making assumptions that do not match your environment
• Makes shorter, vaguer prompts produce better results because the system prompt fills in the gaps

Start with the IT admin system prompt in this guide, adjust the context section to match your own environment, and set it as your custom instruction in ChatGPT. You will notice a difference in quality immediately.

The prompts you write after that can be shorter and more conversational — the system prompt does the heavy lifting.