Suggestions
The Suggestions tab manages the prompt suggestions displayed to users when they select this Custom Model. Suggestions help users understand what the model is capable of and provide a quick starting point for analysis.
Where Suggestions Come From
Suggestions in a Custom Model come from two sources:
1. Automatically Generated (primary source)
When the semantic model is created or updated via the Operations tab, background LLM agents analyze the full semantic layer — all connected data sources, their tables, columns, descriptions, and hierarchies — and automatically generate relevant example questions.
This means the majority of suggestions are produced without any manual input. The agents understand the data content and context and create questions that reflect what can be analyzed with this model.
The quality of auto-generated suggestions depends on:
- The completeness of descriptions at the data source level (table and column descriptions, synonyms)
- The coherence of the connected data sources
- The behavior of the underlying language model
Language models are creative and sometimes unpredictable. Not all generated suggestions will be semantically correct or useful for every use case. Some may be too generic, technically oriented, or not aligned with actual business questions. Review them before exposing them to end users.
2. Manually Added
Builders and Administrators can add their own suggestions at the Custom Model level. These complement or replace auto-generated ones and can be used to highlight the most important or frequently needed queries for this model's audience.
Managing Suggestions
Adding a Suggestion
Click the + icon to add a new suggestion. Each suggestion consists of:
- Title — a short label shown as a clickable chip in the session interface
- Content — the full prompt text submitted to the AI when the user clicks it
Editing and Deleting
Existing suggestions can be edited or deleted using the controls next to each entry.
Suggestion Visibility
Whether and which suggestions are shown to users is controlled by the Prompt Suggestion Visibility settings in the General tab:
| Setting | Effect |
|---|---|
| Show Model-Specific Suggestions | Shows suggestions added manually in this tab |
| Show Datasource Suggestions | Shows suggestions generated from connected data sources during extraction |
| Show Global Suggestions | Shows workspace-wide suggestions configured by the Administrator |
All three can be enabled simultaneously. Users see all active suggestion types combined.
Best Practices
- Review auto-generated suggestions after each semantic model creation or update — not all will be relevant.
- Lock or keep suggestions that accurately represent the model's key use cases.
- Add manual suggestions for questions that users ask most frequently but were not captured automatically.
- Keep suggestion text concise and phrased in natural language, as users would ask it.