Update your knowledge base whenever your course content changes significantly, and do a light review after every cohort you run. A monthly check of what your agent was asked — and how it answered — catches gaps before they become recurring problems.
The Knowledge Base Is Not “Set and Forget”
A lot of educators build their knowledge base once, connect the agent, and never look at it again. That works fine for the first few months — but as your course evolves, your community grows, and AI tools themselves change, the content your agent pulls from starts to go stale. An agent answering questions about a tool you stopped using six months ago is worse than no agent at all.
Think of your knowledge base like a staff handbook. You do not rewrite it every week, but you absolutely update it when the policies change, when a new tool is introduced, or when you notice the team consistently asking about something that is not covered. The same trigger points apply.
When to Update — The Four Triggers
The first trigger is a content change: you update your course, change a process, switch tools, or revise your pricing. Any of those should prompt a knowledge base review to make sure the agent’s answers still match reality.
The second trigger is end-of-cohort. After each live cohort, review the questions your agent was asked that it struggled to answer confidently. Those are your next articles. Students are your best content brief — they ask exactly what is missing.
The third trigger is an incorrect answer you catch or are told about. Fix the relevant article immediately. Do not let a known error sit in the knowledge base while the agent keeps repeating it.
The fourth is a scheduled monthly check — fifteen minutes, skim the conversation logs, look for patterns in what was asked versus what was answered. This catches slow drift before it becomes a problem.
What This Means for Educators
The total maintenance burden is lighter than most educators expect. A well-built knowledge base with fifty solid articles might need one new article a week and a monthly audit — perhaps two hours a month total. That is a fraction of the time the agent saves you in answered student messages.
The key mindset shift: your knowledge base is a living document, not a finished product. Every cohort you run gives you new material. Every question your agent cannot answer well is a publishing prompt. Over time the base gets richer, the agent gets better, and the maintenance gets easier.
The Simple Rule
Update on change, review after each cohort, fix errors immediately, and do a monthly spot-check. Four triggers, not four hundred. Keep that rhythm and your agent stays accurate without consuming your time.