Custom GPT: Build a Team IT Knowledge Base Assistant

What This Builds

You'll create a Custom GPT that acts as a team knowledge base for your IT environment — loaded with your runbooks, network diagrams (as text), script descriptions, and common troubleshooting procedures. Any team member can query it in plain English: "How do we handle a new employee onboarding in HR?" or "What's the runbook for SQL Server failover?" instead of hunting through Confluence or SharePoint. The institutional knowledge that currently lives in senior admins' heads becomes queryable by the whole team.

Prerequisites

• ChatGPT Plus subscription ({{tool:ChatGPT.price}}) — Custom GPTs require Plus
• Runbooks, SOPs, network documentation, or script descriptions to upload (even rough notes work)
• 2 hours for initial build; ongoing as you add documentation

The Concept

A Custom GPT is like a specialized coworker who's been given a stack of your documentation to read before starting work. Every conversation with the GPT has that documentation loaded as context — so when a junior admin asks "how do I reset a user's MFA?" the GPT answers based on your company's specific procedure, not a generic Microsoft tutorial.

Think of it as turning your documentation from a filing cabinet (you have to know where to look) into a coworker (you describe what you need and it finds the right answer).

Build It Step by Step

Part 1: Prepare Your Documentation

Before building the GPT, spend 30 minutes organizing what to upload. The GPT is only as good as the documentation it has.

What to include:

• Runbooks and SOPs (even rough bullet-point versions)
• Network topology descriptions (text description of VLANs, subnets, key servers)
• Common troubleshooting procedures
• Script inventory and descriptions (not the scripts themselves — just what they do and when to use them)
• Vendor contact information and account numbers
• On-call escalation procedures

Format for maximum effectiveness:

• Give each document a clear title at the top
• Use consistent headings: "Procedure," "Prerequisites," "Steps," "Contacts"
• Add environment-specific details (actual server names, IP ranges, domain names)

Note on security: Don't upload documents containing passwords, actual credentials, or sensitive personal information. Describe the process without embedding secrets.

Part 2: Access the Custom GPT Builder

1. Log in to {{tool:ChatGPT.url}} with your Plus account
2. In the left sidebar, click Explore GPTs
3. Click Create in the top right
4. Select the Configure tab for manual control

What you should see: A configuration panel with fields for Name, Description, Instructions, and a Knowledge section.

Part 3: Write Your System Instructions

In the Instructions box, paste and customize this template:

You are the IT knowledge base assistant for [Company Name]'s IT team. You have access to our team's runbooks, SOPs, network documentation, and operational procedures.

Your role: Help IT team members and help desk staff find procedures, troubleshooting steps, and operational information quickly.

How to answer:
1. Always base answers on the uploaded documentation first
2. If the documentation doesn't cover something, say so clearly — don't make up procedures
3. For critical operations (server restarts, firewall changes, AD modifications), always cite which runbook to follow and recommend a second reviewer
4. For questions about passwords or credentials, never provide them — refer the user to the password manager
5. For escalation questions, always reference the current on-call schedule

Our environment:
- Domain: [domain name]
- Key servers: [describe key servers — domain controllers, file servers, email, etc.]
- Monitoring: [which tools]
- Ticketing: [which system]
- Location: [office locations]

Who uses this GPT:
- IT team members (senior and junior)
- Help desk staff for tier 1-2 escalations
- On-call admins during incidents

Part 4: Upload Your Documentation

In the Knowledge section:

1. Click Upload files
2. Upload your runbooks, SOPs, and documentation (PDF, Word, or plain text)
3. Add a brief description to each file if the Knowledge section allows it

Recommended file structure:

• runbooks-server-operations.pdf
• runbooks-network-operations.pdf
• helpdesk-procedures.pdf
• onboarding-offboarding-procedures.pdf
• environment-overview.txt (server names, IP ranges, key contacts)
• incident-escalation-contacts.pdf

What you should see: Files listed in the Knowledge section. Each file is searchable in every conversation with this GPT.

Part 5: Set Conversation Starters

In the Conversation starters section, add prompts that show team members how to use it:

• "How do I onboard a new employee in IT?"
• "What's the runbook for an email server outage?"
• "Who's on call this week and what's the escalation path?"
• "How do I reset a user's MFA?"

Part 6: Test with Real Scenarios

Before sharing with the team:

1. Click Save and open the GPT
2. Test with 5 scenarios the team actually encounters:

   • "A user can't connect to VPN — what are the troubleshooting steps?"
   • "How do we offboard a terminated employee?"
   • "The backup job failed last night — what do I check first?"
3. If the GPT answers from your documentation, it's working
4. If it gives generic answers not based on your docs, the docs may need to be clearer or the instructions need to direct it to cite sources

What you should see: Answers that reference your specific procedures, not generic Microsoft/Cisco documentation.

Part 7: Share with the Team

1. In the GPT settings, change visibility from Only me to Only people with a link
2. Copy the link and share with your team in your IT chat channel
3. Brief the team: "This has our runbooks loaded. Use it as a first stop before escalating tickets or calling senior admins."

Real Example: Handling a "Can't Connect to VPN" Ticket

Setup: You've uploaded a VPN troubleshooting runbook that covers the 5 most common causes: wrong client version, expired certificate, split-tunnel misconfiguration, ISP blocking, and account lockout.

Tier 1 help desk query: "User says they can't connect to VPN. They're working from home. What do I check?"

GPT output: A step-by-step triage guide that mirrors your actual VPN runbook — starting with quick checks (is the account locked out? what error message does the client show?), escalating to technical checks (client version, certificate validity), and ending with escalation criteria ("If steps 1-5 don't resolve it, escalate to Tier 2 with these details collected...").

Time saved: Tier 1 resolves 60-70% of VPN tickets without escalating. Senior admin on-call receives escalations with the basic triage already completed.

What to Do When It Breaks

• GPT gives generic answers, not from your docs → Your documentation may be too vague. Add more environment-specific details (actual server names, exact menu paths for your tools) to the uploaded files.
• GPT says "I don't have information on that" for something you know you uploaded → The specific document may not have been processed correctly. Try re-uploading as plain text instead of PDF, or split the file into smaller sections.
• GPT makes up procedures that aren't in your docs → Add to your instructions: "If a procedure is not documented in the uploaded files, say 'I don't have a documented procedure for this — escalate to [contact]' rather than improvising."
• Team members get wrong answers → Create a feedback process: when someone gets a wrong answer, they report it and you update the corresponding runbook. The GPT improves as documentation improves.

Variations

• Simpler version: Use a regular Claude Project instead of a Custom GPT — same concept, slightly different interface, and doesn't require sharing a link.
• Extended version: Create separate GPTs for different domains — one for Windows/AD procedures, one for network procedures, one for help desk — with relevant documentation in each.

What to Do Next

• This week: Build the GPT with your 5 most-used runbooks and test with your team
• This month: Track which queries get wrong or incomplete answers — use them to identify documentation gaps
• Advanced: Connect the GPT to your ticketing system via API so it can log when it was queried and what procedures it recommended

Advanced guide for network administrator professionals. These techniques use more sophisticated AI features that may require paid subscriptions.