Hey folks, hoping to get some consensus here. I've been noticing our support platform's AI assistant keeps pulling up articles that are, frankly, obsolete. We migrated our entire deployment workflow from Jenkins to GitHub Actions over six months ago and updated the KB accordingly, but the AI still surfaces the old Jenkins pipeline guides at least 30% of the time when someone asks about "build failures."
It's causing real deflection issues—agents report tickets bouncing back because users try the outdated steps and hit a wall. Our deflection rate for deployment-related queries has actually *increased* by about 15% since this started happening, which is the opposite of what we wanted.
Here's a snippet from our internal agent feedback log:
```markdown
Date: 2024-10-26
Issue: AI suggested "Troubleshooting Jenkins Node Connectivity" for query: "GitHub Actions runner won't connect to private VPC."
Result: User attempted obsolete SSH key method, ticket reopened.
Impact: +45min resolution time.
```
I'm curious if anyone else is tracking similar metrics? More importantly, has anyone found a reliable way to "retrain" or weight the AI's search towards more recently updated content? Our KB articles have clear `last_updated` timestamps, so it feels like the AI should prioritize them.
We're considering a brute-force approach—maybe archiving anything older than a year—but that feels like we might lose some valid, evergreen content. Any pipeline wizards out there have a better automation strategy for this?
-pipelinepilot
Pipeline Pilot