Airgentic Help
This page covers the practical tasks that platform administrators and content operators do most often. It focuses on recurring actions rather than one-time setup.
| Task | Frequency | Who typically does it |
|---|---|---|
| Testing with Admin Chat | As needed | Admins, operators |
| Reviewing conversations | Daily or weekly | Content owners, operators |
| Checking flagged conversations | Daily | Operators |
| Running data syncs | After content updates | Operators, admins |
| Managing curated answers | As needed | Content owners, admins |
| Monitoring usage | Monthly | Admins |
| Reviewing search performance | Weekly or monthly | Content owners |
What it is: Admin Chat lets you test your service exactly as users experience it, without affecting analytics or consuming quota.
Why it matters: You should always test changes before making them live. Admin Chat shows you exactly what users will see.
How to do it:
The Trace Log shows:
- Which agent handled the question
- What search query was used
- Which pages were retrieved
- The full prompt sent to the AI
- Any function calls made
When to use it:
- After editing prompts or agent settings
- Before setting an agent status to Live
- When investigating reported issues
- When testing curated answers
Related: Admin Chat
What it is: Customer Service Insights shows all conversations that have occurred, with filters for finding specific types.
Why it matters: Regular review helps you understand what users are asking, spot recurring problems, and identify improvement opportunities.
How to do it:
What to look for:
- Questions that were not fully answered
- Topics that come up repeatedly
- Patterns in negative feedback
- Content gaps (users asking about things not in your knowledge base)
Recommended frequency: At least weekly during active use; daily during initial launch period.
Related: Customer Service Insights
What it is: Conversations are automatically flagged when they likely need attention: thumbs down ratings, negative sentiment, or failed answers.
Why it matters: Flagged conversations are the highest-priority items for review. They indicate something went wrong.
How to do it:
The flag triggers are:
- User gave a thumbs down rating
- Conversation had negative sentiment detected
- AI failed to answer the question
Recommended frequency: Daily during active use.
Related: Customer Service Insights
What it is: Data sync pulls content from your website and updates the search index so the AI can answer questions using current information.
Why it matters: If your website content changes, the AI won't know about it until you run a sync.
How to do it:
When to sync:
- After significant website content updates
- After publishing new pages that the AI should know about
- After fixing content that was causing poor answers
- On a regular schedule (configurable in the Web Crawl Schedule section)
What gets synced:
- Pages that match your crawler scope settings
- Content extracted according to your field mappings
- Uploaded documents (if any)
Related: Data Sync, Crawler Settings
What it is: Curated answers override or guide the AI for specific questions where you need exact control.
Why it matters: Some questions need precise answers that the AI might not generate consistently from general content.
How to do it:
Best practices:
- Phrase the question as users would ask it (the system matches by intent)
- Use curated answers for facts not in your content, conflicting sources, or required exact wording
- Review Insights to find questions that need curated answers
- Test after adding to confirm the override works
Related: Curated Answers
What it is: The Usage Dashboard shows your subscription plan, entitlements, and how much of each allowance you've consumed.
Why it matters: You need to stay within your plan allowances and plan ahead if usage is growing.
How to do it:
What to watch for:
- Meters approaching 90% (shown in yellow)
- Meters over limit (shown in red)
- Trends that suggest you may exceed limits before renewal
Recommended frequency: Monthly, or more often if usage is high.
Related: Usage Dashboard
What it is: Search Insights shows what users are searching for, what results they're clicking, and trends over time.
Why it matters: Search performance directly affects answer quality. If users search for things and don't click results, there may be content gaps or poor categorisation.
How to do it:
What to watch for:
- High-volume searches that aren't resulting in clicks
- Searches for topics you haven't covered
- Unexpected search terms that suggest confusion
Recommended frequency: Weekly or monthly.
Related: Search Insights
These tasks are done occasionally rather than routinely:
Adjust settings like the service title, welcome message, no-answer message, or chat input placeholder.
Where: Service Configuration → General, User Interface, Agent Messages
Related: Service Configuration
Add, edit, or remove users who have access to the admin console.
Where: Account → Manage Users
Related: User Management
Configure search result categories, boost rules, or synonyms.
Where: Search Settings
Note: Changes may require a reindex to take effect.
Related: Search Settings
Add PDFs, Word documents, or HTML files that aren't available via web crawl.
Where: Data Sync → Upload Documents
Related: Upload Documents
Update the instructions that control how agents behave.
Where: Configure Agents → Edit the agent → Agent Prompts
Note: Test thoroughly in Admin Chat after making prompt changes.
Related: Editing an Agent, Editing a Prompt
Here's a suggested routine for content owners and operators:
Contact your platform administrator or Airgentic support when: