Hook

You've solved a tricky password reset issue. The fix took you an hour to figure out, but it involved one specific step that most people miss: checking if the user's mailbox was in recovery mode. Fast forward three months. The same issue comes back. You explain it again. Six months later, someone else explains it. A year later, your team has solved this problem five times, each person rediscovering the solution independently. You haven't documented it because "I'll get to it later." Meanwhile, AI can read your incident notes and your Slack conversation and draft a knowledge base article in 5 minutes. It won't be perfect, but it's 70% done. You edit it for tone, verify the steps are accurate, and publish it. Now the next person finds the solution in the KB instead of asking your team. The cost of not documenting is your team's time spent re-solving solved problems. The cost of not using AI to draft articles is your time spent writing from scratch.

Purpose

This lesson teaches you how to use AI to draft knowledge base articles from existing sources: resolved tickets, chat transcripts, incident notes, and tribal knowledge. You'll learn how to structure the AI prompt so it captures the essential information, verify and refine the draft, and set up a review workflow that keeps articles current. You'll also learn what makes an article searchable and useful to end users.

Why This Matters for IT Professionals

IT organizations run on tributes. Someone figures something out, remembers it next time, and the solution stays in their head. When they leave, the knowledge leaves with them. You can't scale if knowledge isn't documented.

A well-maintained knowledge base cuts ticket volume by 30-50%. Users self-serve. New team members onboard faster. Contractors don't waste time asking the same questions. Your team handles higher-complexity issues instead of re-explaining basics.

But writing KB articles takes time. Most teams don't write them because the ROI feels long-term and the effort feels immediate. AI changes the math. A draft takes 5 minutes. Refining it takes 20. Now the ROI is clearer: 25 minutes of work prevents the same question from coming up 20 times next year.

The key is making it easy to create articles (short prompts, AI drafting), easy to review them (structure, checklists), and easy to keep them current (versioning, review cadence).

Core Concepts

Key Insight: Source Material Shapes Article Quality

Where you get the information affects the article's quality:

• Resolved tickets: Usually have symptoms, troubleshooting steps, root cause, and resolution. Good source material. Missing: context about why this matters, when to use other solutions.
  - Chat transcripts: Messy, with tangents and corrections. But they show the thinking process and common user questions. Good source material for understanding confusion points. Need cleanup.
  - Incident reports: Structured, with timeline and impact. Good for "what went wrong and how we fixed it" articles. Missing: how-to-prevent guidance.
  - Verbal knowledge: "Ask Dave about the backup process." Hard to capture, but often the most useful. Requires interviewing the expert and writing from their explanations.
  - Existing docs: Old articles that need updating. Often outdated. Can be refreshed using new tools/steps.

The best articles combine multiple sources. A ticket shows the problem and solution. A chat transcript shows user confusion points. An interview with the expert clarifies why the solution works.

Key Insight: Article Structure Shapes Usability

A knowledge base article isn't just information. It's information shaped for a specific use case. Different use cases need different structures.

Common structures:

Troubleshooting/How-To:

1. Symptoms (what the user is experiencing)
2. Prerequisites (what they need to do this; what permissions, tools, etc.)
3. Steps (numbered, one action per step)
4. Verification (how to confirm it worked)
5. If that doesn't work (alternative steps or escalation)

Explanatory:

1. What is this? (definition, context)
2. Why should I care? (benefits, when to use it)
3. How does it work? (architecture, concepts)
4. Examples (concrete, relatable examples)
5. When to escalate (when to call support)

Procedural/Runbook:

1. Purpose (what this procedure does)
2. Prerequisites (tools, access, time required)
3. Steps (detailed, with warnings)
4. Verification (how to confirm success)
5. Rollback (how to undo if something goes wrong)

AI doesn't know which structure you need. You tell it upfront.

Key Insight: Tone Matters for End-User Articles

If the article is for end users (how to reset password, how to upload files), the tone should be friendly, jargon-free, and encouraging. If it's for IT staff (how to provision a VM, how to troubleshoot DNS), tone can be more technical.

AI often generates neutral, bland tone. You need to specify:

• Audience: "This is for non-technical end users" changes tone completely.
• Tone: "Friendly and encouraging" vs. "Technical and precise" vs. "Brief and to-the-point."
• Level of detail: "Assume no knowledge of command line" vs. "Assume familiarity with Linux."

Example:

• Bland AI version: "To reset your password, navigate to the password reset portal and enter your email address."
• Better version: "Forgotten your password? No problem. Here's how to reset it in 30 seconds without calling the help desk."

The second version is warmer and more likely to be used.

Key Insight: Searchability Determines Whether People Find the Article

An article that exists but can't be found doesn't help anyone. Searchability depends on:

• Title: "How to Reset Your Password" is more searchable than "Password Management." Users search for what they're trying to do, not abstract concepts.
  - Keywords/Tags: "password," "reset," "forgotten," "account," "login." These should reflect how users would search.
  - First paragraph: Include the key term early. "If you've forgotten your password and can't log in, this article explains how to reset it."
  - Structure: Headings and subheadings help search engines and users scan quickly.

AI doesn't optimize for searchability unless you ask. Specify: "This article should be findable by searches for 'password reset,' 'forgotten password,' 'account locked.'"

Key Insight: Maintenance and Versioning Prevent Stale Articles

A KB article is living documentation. It gets out of date as systems change. You need a strategy for keeping articles current:

• Review cadence: When should you revisit this article? Every year? When you update the system it describes?
  - Owner: Who's responsible for keeping it accurate?
  - Version history: Track what changed and when. "Updated for Outlook 2024."
  - Deprecation: Mark articles that are no longer relevant. Don't just delete them (breaks links); mark them as "Superseded by [new article]."

Without maintenance, your KB becomes a graveyard of outdated guides that confuse users.

Practical Use Cases

Use Case 1: From Resolved Ticket to Knowledge Base Article

Scenario: Your help desk resolved a ticket about "Outlook keeps asking for password." The issue was subtle: the user had two Office 365 accounts, and Outlook was caching credentials incorrectly. The solution involved clearing the Office credential manager and re-authenticating. This is the second time this month you've seen it. Time to document.

What you have:

• A resolved ticket with steps the help desk took
• Chat messages where you discussed why it was happening
• A mental model of what causes this and how to prevent it

AI Prompt:

I need a knowledge base article for non-technical end users about fixing "Outlook keeps asking for password."

Here's a ticket summary:
[Paste ticket description and resolution steps]

And here's chat discussing the root cause:
[Paste relevant chat]

Create a KB article with this structure:
1. Title (user-friendly, searchable)
2. Symptoms (what they're seeing)
3. Before You Start (prerequisites, things they should know)
4. Steps to Fix (numbered, simple language, no jargon)
5. If That Didn't Work (alternative or escalation path)
6. Why This Happens (brief explanation, reassurance it's not their fault)
7. How to Prevent (settings or behaviors to avoid it in future)

Tone: Friendly and reassuring. Assume no technical knowledge. Avoid jargon. Include "You're not alone" messaging.

Include keywords in the first paragraph so this is findable by searches like "Outlook password," "Outlook keeps asking," "Outlook credential issue."

Output as markdown with clear headings.

What you get: A draft article that:

• Explains symptoms in user language ("Outlook keeps asking for your password every time you open it")
• Lists prerequisites ("You need access to your computer to change settings")
• Provides step-by-step instructions ("Click File, then Account Settings, then Clear Password Cache")
• Explains why ("Outlook sometimes caches the wrong account")
• Suggests prevention ("If you have multiple Office 365 accounts, here's how to set them up so they don't conflict")

Before AI: You'd spend 45 minutes writing this article from scratch (deciding structure, phrasing steps clearly, testing them, editing for tone).

With AI + 15 minutes of your review: You have a draft. You read it, verify the steps match your Outlook version, adjust tone if needed, publish.

What you must do:

1. Verify the steps actually work on your version of Outlook.
2. Read it from a non-technical user's perspective: does this make sense? Would I understand this if I didn't work in IT?
3. Test the steps yourself or have someone non-technical test them.
4. Set a "review date" (e.g., "review this article if we upgrade to Outlook 2025").

Time saved: 30 minutes per article. Over a year, documenting 20 of your common issues saves 10 hours. Over those 20 issues' lifetime, users self-serving rather than calling saves 100+ hours.

Use Case 2: From Chat Transcript to Procedural Article

Scenario: Your network team just finished a project to migrate everyone from the old VPN to the new VPN. Users are calling with questions. You've answered many in Slack. Time to document the migration steps.

What you have:

• Slack conversation where you explained the migration to different teams
• Screen shots of the new VPN client setup
• Known issues and workarounds discovered during migration

AI Prompt:

I need a procedural KB article for migrating from OldVPN client to NewVPN client.

Here's a Slack discussion where we guided users through the migration:
[Paste slack transcript]

Here are known issues that came up:
- Issue 1: NewVPN doesn't work with some legacy routers. Workaround: update router firmware.
- Issue 2: Dual-VPN doesn't work on Mac. Solution: disconnect OldVPN completely before using NewVPN.

Create a KB article with this structure:
1. Title (describing the migration)
2. What's Changing (explain why, what benefits they get)
3. Timeline (when old VPN stops working)
4. Prerequisites (access, permissions, what to prepare)
5. Migration Steps (numbered, clear)
6. Verification (how to confirm it worked)
7. Known Issues and Workarounds (with known issues I listed above)
8. Rollback (if something goes wrong, how to go back to old VPN temporarily)
9. Questions? (who to contact)

Tone: Professional, clear, supportive. Assume basic tech skills but not VPN expertise.

Output as markdown.

What you get: A comprehensive migration guide that walks users through the process, surfaces known issues before they hit them, and provides a rollback path if something goes wrong.

Before AI: 1.5 hours (organizing the slack history, structuring the article, writing from scratch, coordinating with team to verify).

With AI + 30 minutes of review: A published guide that prevents 20+ support calls during migration.

What you must do:

1. Verify the migration steps match your actual process (versions, tool names, exact UI).
2. Coordinate with the team: "Does this match what we did? Did I miss anything?"
3. Test the rollback process: Can users actually go back to old VPN if needed?
4. Prepare a deprecation date: "After [date], the old VPN client will no longer work. Migrate now."

Use Case 3: Interview-Based Article Creation

Scenario: Your backup process is complex and somewhat idiosyncratic. Only one senior admin fully understands it. You want to document it so other team members can handle backups if needed. You don't have a ticket or chat. You need to interview the admin.

What you do:

1. Record a 30-minute conversation with the admin explaining the backup process, common issues, recovery procedures, and edge cases.
2. Transcribe the conversation (AI can do this if you use a tool that transcribes).
3. Use the transcript as source material for the article.

AI Prompt:

I need a KB article for our IT staff about our backup process and recovery procedures.

Here's a transcript of an interview with our senior backup admin:
[Paste transcribed conversation]

Extract and organize this into a KB article with this structure:
1. Overview (what this documents, why it matters)
2. System Components (the backup infrastructure: what tools, where are backups stored, how long are they retained)
3. Daily Backup Process (what happens automatically; what the admin monitors)
4. Recovery Process (how to restore from backup; step by step)
5. Common Issues and Solutions (things that go wrong and how to fix them)
6. Emergency Procedures (if the backup system itself fails)
7. Contact Information (who to escalate to)

Tone: Professional, technical. Assume IT staff knowledge, but explain our specific setup clearly.

Output as markdown with clear headings and code blocks where appropriate.

What you get: A comprehensive backup runbook that captures the senior admin's knowledge in a form others can follow.

Before AI: 2+ hours (writing the article from notes and interviews, with multiple reviews with the admin).

With AI + 1 hour of review and refinement: A published runbook.

What you must do:

1. Have the senior admin review the draft: "Is this correct? Did I miss anything important?"
2. Verify technical details: storage locations, retention policies, recovery time objectives.
3. Test recovery: Walk through the recovery steps and confirm they work.
4. Set up maintenance: "When we change the backup system, we update this article."

Examples

Example 1: Weak vs. Strong Article Drafting Prompt

Weak prompt:

Write a KB article about password resets

What you get: Generic article about password management, probably too long, probably too technical for end users, probably not focused on the problem.

Strong prompt:

Write a KB article for non-technical end users titled "How to Reset Your Forgotten Password."

This is for users who have forgotten their password and can't log in to their account. They can use our self-service password reset portal.

Steps:
1. Go to [url]
2. Enter your email
3. Check your email for a reset link (if you don't see it in 5 minutes, check spam)
4. Click the link and create a new password
5. Log in with your new password

If the email doesn't arrive, troubleshooting:
- Check spam folder (sometimes ends up there)
- Wait 5 minutes (it might take time)
- Try again (sometimes our email system is slow)
- If you still don't receive it, contact help desk with your email

Tone: Friendly, reassuring, jargon-free. Use "you" language. Include reassurance ("This is normal, here's how to fix it quickly").

Output as markdown with emojis or icons if appropriate. Include a "Need more help?" section with help desk contact.

What you get: A focused, user-friendly article that addresses the specific problem, includes troubleshooting, and guides them to help if needed.

Example 2: Article Review and Refinement

AI Draft:

How to Reset Your Domain Password

If you've forgotten your domain password or need to change it, you can use the password reset service.

Navigate to the password reset portal and authenticate using alternative methods (your email or security questions). Enter your new password and confirm it. The password must be at least 12 characters and contain uppercase, lowercase, digits, and symbols.

After resetting, wait 10 minutes for the change to propagate across systems. You can then log in with your new password.

If you encounter errors, contact the IT support team at [email protected].

Issues:

• Tone is formal and technical ("authenticate using alternative methods")
• Not scannable (no emphasis, hard to follow)
• Doesn't address common questions
• Missing troubleshooting

Your revision:

Forgot Your Password? Here's How to Reset It

It happens to everyone. Here's how to get back into your account in 5 minutes.

Step by Step

1. Go to [password reset portal link]
   2. Click "I forgot my password"
   3. Enter your email address
   4. Check your email for a reset link
   - It usually arrives in 1-2 minutes
   - If you don't see it, check your spam folder
   5. Click the link in the email
   6. Create a new password:
   - Must be at least 12 characters
   - Include: uppercase, lowercase, numbers, symbols (example: MyNewPass123!)
   7. Click "Save password"
   8. Log in with your new password

The email didn't arrive?

• Check spam folder, sometimes our emails end up there
  - Wait a few minutes, we're usually quick, but can be slow sometimes
  - Try the reset again, you can request another link
  - Still stuck?, Call help desk at [number] or email [support email]. We'll get you back in.

After you reset

It takes about 10 minutes for your password change to work everywhere. If you can't log in right away, wait a bit and try again.

Questions?

Contact IT support: [email/phone]

Improvements:

• Friendly opening ("It happens to everyone")
• Clear numbered steps
• Explains what "reset link" means and where to find it
• Acknowledges common problems (email in spam)
• Provides alternatives if email doesn't work
• Sets expectations (10-minute propagation delay)

Example 3: Handling Outdated Articles

Scenario: You have a KB article "How to Connect to VPN with the VPN Client." But last year you switched to a new VPN system, and the article is now wrong. You could delete it (breaks bookmarks and links), but better to refresh it.

AI Prompt:

I have an outdated KB article about connecting to VPN. The old process involved installing the Cisco AnyConnect client. We've since switched to OpenVPN with a new client.

Here's the old article:
[Paste old article]

Here's what the new process is:
[Describe new process]

Refresh this article for the new VPN system. Keep the same structure and tone, but update all steps and references to the new system. At the top, add a note that this article was updated on [date] for the new system.

Output as markdown.

What you get: An updated article that preserves the original structure (users who found the old link can still use it) but with current information.

Anti-Patterns

Anti-Pattern 1: Publishing AI Drafts Without Review

Don't publish an AI-drafted article without reading it. Tone might be off, steps might not match your actual system, details might be wrong.

Anti-Pattern 2: Not Measuring Article Impact

Don't assume your KB is helping. Measure it: "Do fewer people call asking about password resets after we publish a KB article?" If articles aren't reducing ticket volume, they're not helping enough to justify maintenance.

Anti-Pattern 3: Letting Articles Go Stale

Don't just publish an article and forget about it. Set review dates ("Review this annually or when we update the system"). Old articles confuse users more than no articles.

Anti-Pattern 4: Writing Articles No One Can Find

Don't optimize articles for writing and completeness; optimize for search and usability. "How to Reset Your Password" is more findable than "Password Management Procedures." Users search for what they're trying to do.

Anti-Pattern 5: Over-Automating Article Creation Without Human Touch

Don't just dump AI-drafted articles into your KB. They need human review for tone, accuracy, usability, and searchability. The AI does 70% of the work; you do 30% to make it great.

Human Judgment Checkpoints

Before publishing a KB article:

• 
  Is the structure right for the audience? (How-to for users, runbook for IT staff, explanatory for learning)
• 
  Does the tone match the audience? (Friendly for end users, technical for IT staff)
• 
  Are all the steps accurate for your system? Test them if possible.
• 
  Is it easy to find? (Searchable title, relevant keywords, first paragraph includes key terms)
• 
  Does it have a review plan? When should you revisit this to keep it current?
• 
  Have I linked related articles? Help users discover more if they need it.

Key Takeaways

• Source material shapes article quality. Use resolved tickets, chat transcripts, incident reports, and interviews. Combine multiple sources for best results.
  - Specify structure upfront: troubleshooting, explanatory, or procedural. AI doesn't know which unless you say.
  - Tone matters for usability. Articles for end users should be friendly and jargon-free. Articles for IT staff can be technical.
  - Searchability determines usefulness. Title and keywords should reflect how users search. Include key terms early in the article.
  - Plan for maintenance. Set review dates, assign owners, deprecate rather than delete outdated articles. A stale KB confuses users.