Launching a chatbot feels like finishing a project. It is actually the start of one. Prices change, policies change, products come and go, and a chatbot knowledge base that was perfect in January can be quietly wrong by March. The good news is that keeping it healthy does not need a big effort. A short, repeated routine does most of the work. Here is one you can copy.
Why Knowledge Goes Stale
A grounded chatbot answers from what you have written. That is its strength, because it will not invent things. It is also its weakness, because it will repeat anything you wrote, including what is no longer true. An outdated answer is more dangerous than a missing one. A missing answer leads to "I don't know" and a human reply. An outdated answer gives a visitor a wrong fact with full confidence.
Staleness usually comes from a few predictable sources:
- Prices, fees and offers that changed on the website but not in the bot.
- Policies such as returns, cancellations or delivery times.
- Opening hours around holidays and seasonal changes.
- Products or services that were added, renamed or discontinued.
- Contact details such as phone numbers, emails and addresses.
The Monthly Routine
Block out forty-five minutes on the same day each month. Treat it like paying a bill: dull, regular, essential.
Step 1: Review the Unanswered Questions (10 minutes)
Open the list of questions the bot could not answer. Sort them into questions you can now answer, questions that need a person and noise. Write a Q&A pair for each real, repeatable one. This is the single most valuable step because it comes straight from visitors.
Step 2: Check the High-Risk Facts (10 minutes)
Open a short list of facts that cause real trouble when wrong, and verify each against your website and internal records. A useful starter list:
| Fact | Check against | Risk if wrong |
|---|---|---|
| Prices and fees | Pricing page, invoices | Disputes, lost trust |
| Delivery or turnaround times | Operations, courier terms | Complaints, refunds |
| Returns and cancellations | Policy page | Legal or financial exposure |
| Opening hours | Google profile, website | Wasted visits |
| Contact details | Contact page | Lost leads |
Step 3: Look for Duplicates and Conflicts (10 minutes)
As the knowledge base grows, you end up with two answers to nearly the same question, written at different times and saying slightly different things. Search for key words, such as "delivery" or "refund", and read the matches together. Merge duplicates and delete the older version. Conflicting answers confuse the bot as much as they would confuse a new employee.
If your tool has a search box and a category filter, use them here. Categories such as Pricing, Delivery and Booking make a review much quicker.
Step 4: Test Ten Real Questions (10 minutes)
Ask the chatbot ten questions the way a customer would type them. Include a couple of awkward phrasings and at least two questions you know it cannot answer. Check that the answers are correct and that the unknown questions get an honest "I don't know" and a route to a person. For a fuller checklist, see how to test a chatbot before launch. It is useful after launch too.
Step 5: Write Down What Changed (5 minutes)
Keep a small change log: date, what you changed, why. It sounds fussy, but it saves arguments later ("Why does the bot say that?") and helps a new team member understand the history.
Triggers for an Immediate Update
Do not wait for the monthly routine when something important changes. Update the knowledge base on the same day as:
- A price or fee change
- A new or changed policy
- A product launch or discontinuation
- A holiday closure or a temporary change in hours
- A new service area or delivery rule
A simple way to make that stick is to add "update the chatbot" to the checklist you already follow when you change the website.
Keep Answers Easy to Maintain
Answers stay easy to maintain when each fact lives in one place, dates are avoided, categories are used and the language is plain.
- One fact, one home. If your delivery time appears in ten answers, changing it means ten edits. Where possible, write the fact once and keep other answers short.
- Avoid dates inside answers unless they must be there. "Orders placed before 2 pm ship the same day" ages better than "We are closed until the 14th".
- Use categories. They make reviews faster and gaps easier to spot.
- Prefer plain language. Simple answers are easier to check and harder to get wrong. For writing tips, read how to write Q&A pairs that make a chatbot useful.
Who Should Own It
Name one person. Knowledge bases drift when everyone assumes someone else is checking them. The owner does not need to write every answer, but they should run the monthly routine and make sure changes elsewhere in the business reach the bot. In a very small team, that person is often the founder or the person who handles customer emails.
A Sample Change Log Entry
A change log entry needs only the date, what changed, why and who changed it. Four short fields are enough to settle most later questions about why the bot says something.
- Date: the day you made the change.
- What changed: for example, "Updated delivery time answer from 3 to 5 days to 4 to 6 days."
- Why: for example, "New courier from this month."
- Who: the person who made the edit, so questions go to the right place.
Signs Your Knowledge Base Needs Attention Now
Your knowledge base needs attention now when visitors correct the bot, when low ratings appear on routine questions, or when your website says something different from the bot. Any one of these is a signal to stop and review.
- A customer says the bot's answer did not match what they were told elsewhere.
- The same topic keeps appearing in your unanswered list even after you added a pair.
- You changed something on the website and cannot remember updating the bot.
- Your team starts correcting the bot's answers by hand in the inbox.
Conclusion
Keeping a chatbot accurate is a small, regular job: review unanswered questions, verify the high-risk facts, remove duplicates, test ten real questions and log your changes, plus immediate updates when something important changes. Assign one owner, and forty-five minutes a month will keep your bot trustworthy.
Key takeaways
- An outdated answer is worse than a missing one because the bot repeats it with confidence.
- Run a forty-five-minute monthly routine: unanswered questions, high-risk facts, duplicates, ten test questions and a change log.
- Update immediately after price, policy, product or hours changes.
- Write each fact once and use categories so reviews are quick.
- Give one named person ownership of the knowledge base.
Quick Summary
Chatbot knowledge base maintenance is a monthly routine of about forty-five minutes: review unanswered questions, verify high-risk facts such as prices, policies and hours, merge duplicate or conflicting answers, test ten real questions and log what changed. Update immediately when prices, policies, products or opening hours change. An outdated answer is worse than a missing one, so name one owner for the knowledge base.
Want to see grounded answers on your own site?
Start free, add your Q&A pairs and paste one script tag. No credit card required.

Founder & CEO
Writes about grounded AI chatbots for websites: how they work, how to measure them and what to check before you buy.
Topics
Get new posts by email
Guides on AI chatbots for websites, straight to your inbox.
