When your agent gives a wrong, vague, or unhelpful response, it's almost never a problem with the AI itself — it's a problem with the information it was given. This guide walks through every type of weak response and exactly how to fix it.
Type 1 — "I don't have information on that"
What's happening: The agent searched your knowledge base and found nothing relevant. The information simply isn't there.
How to fix it:
- Identify what the visitor was asking
- Go to your Knowledge Base tab and add a manual Q&A entry with the correct answer
- If it's a topic that should be on your website, add it to your website and re-scrape
- Re-test the question to confirm it's now answered
Type 2 — Vague or incomplete answer
What's happening: The information is in your knowledge base, but it's too vague for the AI to give a specific answer. This often happens when your website uses phrases like "competitive pricing", "varies by requirement", or "contact us for details".
How to fix it:
- Find the source content causing the vague answer (check the knowledge base)
- Rewrite it with specifics: real numbers, real timelines, real details
- Or add a manual Q&A pair that gives the specific answer directly
Before: "Our pricing is competitive and depends on your requirements."
After: "Logo design starts at $499. Brand identity packages start at $1,200. Custom quotes available on request."
Type 3 — Factually wrong answer
What's happening: The agent found content in your knowledge base but it's outdated or contradictory. This is the most damaging type of error — it actively misleads visitors.
How to fix it:
- Go to your knowledge base and search for the topic
- Find any entries with the old/incorrect information
- Delete them immediately — do not leave them disabled, delete them
- Add a new entry with the correct information
- Re-scrape your website if the outdated content was from a web page
Type 4 — Off-topic or unexpected response
What's happening: The agent is responding to questions outside its intended scope, or drifting away from your brand voice. This is usually a system prompt problem.
How to fix it: Add explicit scope boundaries to your system prompt:
Only answer questions related to [your business and services]. If asked about unrelated topics, politely redirect: "That's outside my area — but I can help with questions about [service]. What would you like to know?"
Type 5 — Response is too long
What's happening: No length constraint in the system prompt. The AI defaults to being thorough, not concise.
How to fix it: Add one line to your system prompt: "Keep all responses to 2–3 sentences unless the visitor explicitly asks for more detail." This instantly makes all responses cleaner and more readable.