Be specific about what you want
Instead of vague requests, describe the exact outcome:| Less effective | More effective |
|---|---|
| ”Make a chart" | "Create a bar chart of monthly revenue by product category" |
| "Fix this" | "This query fails with a column not found error on user_name — can you check the correct column name?" |
| "Show me data" | "Show me the top 10 customers by total spend in the last 90 days” |
Reference columns and tables by name
The AI searches your catalog, but it’s faster and more accurate when you reference specific names:“Join theorderstable withcustomersoncustomer_idand show total order amount by customer name”
Use the right mode
Switch to the mode that matches your task:- SQL mode when you’re focused on writing or debugging queries.
- Dashboard mode when you’re building or rearranging boards and charts.
- Ask mode when you just want to explore or understand data without making changes.
- Auto mode when your task spans multiple areas.
Enrich your catalog
The AI uses your catalog’s table and column descriptions for semantic search. If your table is calledtrx but you ask about “transactions,” the AI relies on descriptions to bridge that gap.
Catalog descriptions can be generated automatically — see Catalog — or you can add them manually.
Let the AI validate
The AI can run queries to check its own work. If you’re unsure about a suggestion, ask:“Run the query first to make sure it works”
Upload images for context
Paste screenshots of:- Error messages for debugging.
- Schema diagrams or ERDs for reference.
- Charts you want to reproduce.
Iterative refinement
Start with a simple request and refine:- “Show me monthly revenue” — AI creates the query.
- “Add a breakdown by region” — AI updates it.
- “Now make it a stacked bar chart on the Sales board” — AI creates the chart.