Help

Overview
Overview What to Expect Pre-Session Checklist Your Onboarding Session
Platform Mental Model Admin & Operator Essentials Guardrails and Governance What to Use When Roles and Responsibilities Post-Session Next Steps Optional Deep-Dives Deep-Dive: Agents and Prompts Deep-Dive: Search and Knowledge Deep-Dive: Functions and Secure Services Deep-Dive: Handoff and Escalation Deep-Dive: Voice Mode Deep-Dive: Analytics and Improvement Facilitator Guide
Overview Account Overview User Management Usage
Overview General User Interface Agent Messages Voice Mode Analytics
Admin Chat UI Design Curated Answers Search Settings Conversational Intelligence Data Sync Upload Documents
Overview General Crawl Scope Image Extraction Field Mappings
Overview Creating an Agent Editing An Agent
Overview Editing a Prompt User Segmented Prompts
Overview Writing Your First Function AirScript Functions AirScript Reference
Overview Privacy Security
Analytics API Documents Ingestion API
Human Handoff
Overview Customer Service Insights Categories And Themes Search Insights
Overview Identity Provider Widget SharePoint Integration How It Works SharePoint Checklist Implementation Checklist
Admin Console Authorisation
Methods Script Tag Gtm Analytics AI Question Input Fields
Overview Optimiser Monitor
Contact Support

Site Search AI Assistant

Airgentic Help

The AI Search Configuration Assistant on Search Configuration helps admins draft or refine search.conf — the descriptor that controls indexed fields, filters, scopes, result cards, and search UI behaviour.

The assistant is evidence-driven: it reviews your current configuration, optional reference website URL(s), and your instructions, then returns a proposal for review. Nothing is saved until you click Apply proposal.

What The Assistant Can Safely Change

The assistant can propose changes to descriptor-backed search settings in search.conf, including:

Area Examples
Fields New keyword, date, hierarchy, or text fields when there is evidence from HTML metadata, JSON-LD, URL patterns, or existing category data
Filters Enabling facets on existing or new fields
Scopes Adding or adjusting search tabs, scope filters, visible tabs, default sort, and scope-specific facets
Result cards Choosing built-in layouts such as media_card, compact, or tile_card, and modest badge changes
UI / API flags in the descriptor Search theme, overlay template, Ask tab, AI answer, chat agent, and backend search API enablement when explicitly requested

The assistant is conservative by design:

  • It preserves existing valid configuration unless there is clear evidence or an explicit instruction to change it.
  • Standard fields such as title, description, image, date, page_type, and format are assumed to exist and are not reinvented unless necessary.
  • It prefers configuring scopes from existing category values rather than duplicating category logic inside search.conf.
  • It does not edit ingest.conf, auto_mappings.yaml, or service.conf in the current version.

What Requires Extra Caution

Some descriptor changes can trigger reprocessing or reindexing:

Change Typical impact
Adding a new field with HTML/XPath/JSON-LD extraction Usually requires filter HTML processing so values are extracted during crawl
Renaming a field or changing a field type Can require a full reindex because indexed data no longer matches the schema
Changing result-card field references or scope filters that depend on indexed values May require reprocessing depending on the change
Scope-only or UI-only changes Often need no reindex, but you should still preview search after applying

The assistant shows:

  • Validation status — whether the proposed descriptor passes platform validation
  • Diff summary — what would change compared with the current descriptor
  • Reprocess impact — plain-language guidance on whether background ingest jobs are likely to be queued when you apply the proposal

Review these sections carefully before applying.

  1. Open Search Configuration and click AI Assistant.
  2. Provide one or more reference website URL(s) and/or clear design instructions.
  3. Optionally enable Browse additional pages when the site structure is not obvious from the homepage alone.
  4. Review the returned proposal:
  5. design summary
  6. changed vs kept sections
  7. validation and reprocess impact
  8. warnings and test checklist
  9. Click Apply proposal only when the result looks correct.
  10. Use the Preview tab and representative queries to verify scopes, filters, and result cards.
  11. Trigger or wait for any queued reprocessing jobs before rolling the experience out broadly.

Good Instructions To Provide

Examples that work well:

  • “Add a Courses tab filtered to the Courses category, with date sorting and delivery/faculty filters if metadata supports them.”
  • “Keep the current configuration, but make News use date sorting and show date + type badges.”
  • “Review our public site and propose scopes for Programs, Events, and News based on navigation and URL patterns.”

Examples to avoid unless you intend a larger change:

  • “Replace all categories” — categories are usually managed on the Categories tab and in crawl configuration.
  • “Rename every field” — this can force a full reindex.
  • “Make search look like our homepage CSS” — styling is handled in Styling / style.css, not in search.conf.

When Reindexing Is Required

You usually need reprocessing when the assistant adds or materially changes indexed field definitions or extraction rules.

You usually do not need reindexing for:

  • adding or hiding scope tabs that filter on existing page_type values
  • changing result-card layout or badges that reference existing indexed fields
  • changing descriptor UI settings such as theme, Ask tab, or overlay template

If Apply queues background jobs, Airgentic will email the applying admin when those jobs complete.

You have unsaved changes