Tuning FAQ
Frequently asked questions about tuning your AI agent's responses.
Knowledge Base
Q: Why is keeping the knowledge base updated important?
Updating the knowledge base ensures that the AI remains effective and relevant, avoiding the "garbage in, garbage out" problem. Only use clear, accurate, and current information.
Q: Should I update the knowledge base or add human feedback?
It depends on the problem:
If the knowledge base has outdated or incorrect information, update the knowledge base
If the knowledge base is correct but the AI still gives a wrong answer, add human feedback
Human feedback is faster to add and takes priority when there's conflicting information, but a clean knowledge base is still essential.
Human Feedback
Q: How do I use human feedback to improve responses?
Human feedback lets you correct specific AI answers. When the AI gives an incorrect or unsatisfactory response, you can provide the correct answer directly, and the AI will learn from it.
Find the conversation in the dashboard under "Conversations"
Use the "Provide Answer Feedback" option
Submit the correct response
See the human feedback guide for details.
Q: Can I edit feedback I've already provided?
Yes. Find the conversation in "Conversations" or go to the FAQs page to manage previous feedback. You can edit or delete the FAQ that was generated from your feedback.
Q: Does feedback create a new FAQ every time?
Yes, by default feedback creates a new FAQ to help the AI respond better to similar questions.
If your feedback is a general instruction rather than a specific correction, add it as a guideline instead.
FAQs
Q: How are FAQs sorted in the dashboard?
By last-modified date.
Q: Can FAQs include images or GIFs?
Yes. When providing feedback, mention the media you want included. The AI will incorporate it into the generated FAQ. You can also manually add media URLs by editing the FAQ.
Q: Can I export all FAQs to CSV?
Not directly in the dashboard. Contact the Alhena support team and we'll provide a CSV export.
Q: Can I assign ID numbers to FAQs?
This feature is not currently available.
Q: Why are two FAQs flagged as conflicting?
The AI flags FAQs that are very similar. You can keep both, but it's easier to maintain just one so you don't have to update multiple places.
Q: What exactly is the Answer field in an FAQ? Is it conversation-specific, or does it become a global reply once saved?
Once saved, the Answer field becomes a global reply. During the initial FAQ creation, the AI considers the specific conversation context, but after saving, the answer is conversation-agnostic and available globally.
Q: Can I add hints that point to articles instead of full answers?
Yes. When providing feedback, include the article title you want referenced. The AI will review that article when generating the FAQ.
Q: How do I keep FAQs updated?
Keeping FAQs current is a manual process. When reviewing conversations, you'll see which FAQ was used as the source. If you notice an outdated answer, click the FAQ to edit it directly.
Guidelines & Personality
Q: What are guidelines and personality settings?
Personality controls the AI's tone, name, and identity
Guidelines are rules for how the AI should behave in specific scenarios
See the personality and guidelines guides for details.
Q: Can I customize the identity and tone of the AI?
Yes. The AI's identity and tone can be customized in the "Personality" section of the dashboard, allowing it to align more closely with your company's branding or communication style.
Q: When should I use guidelines vs human feedback?
Use human feedback for correcting specific answers
Use guidelines for broad behavioral rules that apply across many conversations
Q: How do I make the AI avoid certain types of questions?
Add a guideline. Examples:
If your answer has potential of steering financial decisions, mention the disclaimer - "This is not financial advice. Please do your own research before making any financial decisions."Never provide investment advice or discuss strategies for making money or getting rich.
Q: What should I consider before adding a new guideline?
Decide if you really need a guideline or if human feedback will suffice. Consider whether the guideline overlaps with existing rules or is beyond the AI's capabilities. Ensure it's necessary and will effectively enhance the AI's performance.
Q: How do I write effective guidelines?
Ensure guidelines are actionable, concise, and directly influence the AI's interactions. Avoid overloading the AI with too many rules, which can confuse it and diminish its effectiveness. See the guidelines guide for best practices.
Troubleshooting
Q: The AI says "Sorry I couldn't find an answer to your question"
Check if the Answer Sources in the dashboard contain the needed information
If not, update your knowledge base with the correct information
If the knowledge base is correct, provide human feedback to guide the AI
Check that the query isn't restricted by a guideline you've set
Q: The AI gave an incorrect answer. How do I fix that?
Use human feedback to correct it. Find the conversation in the dashboard and use "Provide Answer Feedback" to submit the correct response. See the human feedback guide.
Q: The AI included a broken link (404) in its answer
Update your knowledge base to remove or replace the outdated link
Provide human feedback for that specific answer so the AI stops using that link
Q: The AI isn't following my guidelines
This usually means the guidelines are unclear or conflicting. Make sure your guidelines are:
Clear and specific
Not contradicting each other
Within the AI's capabilities
See the guidelines guide for writing effective guidelines.
Q: Can I see what part of a document was used by the AI to generate an answer?
Identifying the specific part of a document that the AI used can be challenging. The AI synthesizes information from multiple document chunks (up to 20 relevant pieces) to construct a response. Any single paragraph in the AI's response may be a hybrid, incorporating elements from several different sources. You can view "Answer Sources" in the dashboard to get a general sense of which documents influenced the response.
Q: Why did the AI refuse to answer a question or block a user?
If the AI detects attempts to trick or manipulate it, it may refuse to answer or block the user. If this was in error, please contact support.
Last updated