Automate Knowledge Base Article Creation from Resolved Tickets

What This Builds

A workflow that automatically drafts KB articles from your resolved tickets — at the end of each day or week, resolved tickets are processed through AI, which extracts the troubleshooting pattern and produces a ready-to-review KB draft. You review and publish; you don't write from scratch. Over 3 months, your KB transforms from a sparse, outdated collection to a comprehensive, current library that genuinely deflects tickets.

Prerequisites

• Claude Pro account ($20/month) — needed for larger batch processing
• Access to your ticketing system's API or export function (ServiceNow, Jira, Zendesk, Freshdesk — all have this)
• Basic comfort with JSON or CSV data (you don't need to code, just understand the format)
• A KB platform to publish to (Confluence, IT Glue, SharePoint, Notion)

The Concept

Most KB articles go unwritten because the process requires two things at the same time that rarely align: 1) you just resolved a complex issue (you have the knowledge) AND 2) you have 30-45 minutes of uninterrupted writing time (you don't).

This workflow separates the two. The ticket resolution is your input. The Claude-powered pipeline runs at the end of the day while you're not writing anything. You just review drafts in the morning — a 15-minute task instead of a 2-hour one.

Build It Step by Step

Part 1: Identify Which Tickets Deserve KB Articles

Not every ticket becomes a KB article. Set criteria:

KB article candidates:

• Ticket required more than 30 minutes to resolve
• Issue appears 3+ times in the last 30 days
• Resolution involved non-obvious steps
• Issue affects multiple users or machine types
• You had to research the fix online (meaning it wasn't in your KB)

Not KB article candidates:

• Simple password resets (procedure doesn't change)
• Hardware failures (no troubleshooting steps, just swap hardware)
• One-off unique situations unlikely to recur

Part 2: Set Up Your Ticket Export

Option A — Weekly manual export: Export resolved tickets from the last week to CSV from your ticketing system. Most systems have a built-in report export. Filter for: resolved, closed, this week. Export with columns: Title, Description, Resolution Notes, Category, Time to Resolve.

Option B — Automated via ticketing system API: Use a simple API call (Zendesk, Jira, ServiceNow all have REST APIs) to fetch resolved tickets. This is more advanced — only pursue if you're comfortable with API calls.

For most analysts, Option A (5 min weekly) is sufficient.

Part 3: Create Your KB Draft Prompt Template

In Claude, create a prompt that processes multiple tickets at once. Save this as a text file for reuse:

I'm going to give you a batch of resolved IT support tickets. For each ticket that meets these criteria, draft a knowledge base article:
- Issue required research or non-obvious troubleshooting
- Resolution steps could be followed by another analyst
- Issue is likely to recur

SKIP writing an article for:
- Simple password resets
- "Replaced hardware" resolutions with no troubleshooting steps

For each qualifying ticket, write a KB article with these sections:
1. Title (clear, searchable, problem-focused)
2. Problem Description (what the user reports seeing)
3. Affected Environments (OS, app versions, conditions)
4. Resolution Steps (numbered, specific — include exact commands, paths, registry keys)
5. Verification (how to confirm the fix worked)
6. Common Variations (similar symptoms with same root cause)

Here are this week's resolved tickets:

TICKET 1:
Title: [paste]
Description: [paste]
Resolution Notes: [paste]

TICKET 2:
[continue]

Part 4: Run the Weekly KB Draft Session

Every Friday afternoon (or Monday morning — pick what works):

1. Export your week's resolved tickets to CSV
2. Filter for tickets with meaningful resolution notes
3. Open Claude Pro
4. Paste the prompt template with your tickets filled in
5. Claude generates draft articles for qualifying tickets
6. Copy drafts into a "KB Drafts" folder in Confluence/IT Glue (not published yet)

With 20-30 tickets per week, Claude typically generates 4-8 KB article drafts. The batch processing takes 3-5 minutes in Claude.

Part 5: Review and Publish (Monday Morning Routine)

Spend 15-20 minutes Monday morning reviewing the drafts from the previous week:

• Read each draft: does it accurately describe the issue?
• Check the resolution steps: are they specific enough? Any steps missing that only you know?
• Add any screenshots or specific paths you didn't include in your ticket notes
• Publish the ones that are good enough (80% quality is fine — imperfect KB articles are better than no KB articles)
• Delete or flag for revision the ones that need more work

Part 6: Track the Impact

After 2 months of consistent KB building, run a report in your ticketing system:

• How many tickets were closed with "resolved via KB article" as the solution?
• Are the ticket types covered by your new KB articles declining in volume?

This data makes the case to your manager for continued AI tool investment — and validates that the effort is worth it.

Real Example: Friday KB Processing Session

Scenario: You've had a busy week with 45 tickets. You export them to CSV. You scan the list and identify 8 tickets that had non-obvious resolutions. You paste them into Claude.

Input (abbreviated):

TICKET 23: "Outlook search not returning results"
Resolution: OST file corrupted. Renamed .ost file in AppData\Local\Microsoft\Outlook, Outlook rebuilt cache. [15 steps noted]

TICKET 27: "Can't connect to mapped network drive after password change"
Resolution: Credential Manager had stale credentials. Cleared old credentials for server in Credential Manager, remapped drive.

TICKET 31: "New MacBook not showing on domain"
Resolution: JAMF enrollment failed due to MDM token expired. Renewed MDM push certificate in Apple Business Manager, re-enrolled device.

[5 more tickets]

Output (per ticket): Full KB articles, each with problem description, affected environments, numbered resolution steps, and verification steps.

Monday morning review: 6 of 8 drafts are publishable with minor edits. 2 need screenshots added. You spend 20 minutes reviewing and publish 6 articles.

3 months later: Your KB has 60+ articles instead of 8. Users find answers themselves. "Outlook search" tickets drop by 70%.

What to Do When It Breaks

• Claude produces KB articles with wrong steps → Your ticket resolution notes were too vague. Add more specific steps in your ticket notes going forward — Claude can only work with what you give it
• Articles are too generic → Add your environment specifics to the prompt: "Our environment is Windows 11, domain joined, M365 Exchange Online — include these specifics in articles where relevant"
• Too many articles to review → Raise your criteria — only process tickets over 60 min to resolve, or that appeared 5+ times. Quality over quantity.
• Ticket export takes too long → Build the weekly export into your Friday end-of-day routine — 5 minutes while shutting down

Variations

• Simpler version: Instead of batch processing, do one KB article per day immediately after your most complex ticket — simpler, no export needed, but requires the daily habit
• Extended version: If your ticketing system has a Webhook or API trigger, you can automatically send resolved tickets to Claude via the API and have drafts appear in your Confluence space automatically — no manual export step needed

What to Do Next

• This week: Do a manual batch run using Claude free tier — pick 5 complex tickets from the past month and generate KB articles from them. See the quality.
• This month: Establish the weekly export → Claude batch → review routine. Commit to one month and track how many articles you publish.
• Advanced: Connect your ticketing system API to Claude's API to fully automate the ticket-to-draft pipeline — no manual export needed, drafts appear automatically for your review

Advanced guide for IT support technician / help desk analyst professionals. These techniques use more sophisticated AI features that may require paid subscriptions.