Writing Better AI Queries
Estimated reading time: 12 minutes
The quality of your AI analysis depends heavily on how you phrase your questions. This guide covers practical techniques for getting accurate, useful results.
Principle 1: Be Specific
Vague questions produce vague answers. Compare:
| Vague (Avoid) | Specific (Better) |
|---|---|
| Show me sales | Show me total revenue by product category for Q4 2024 |
| How is marketing doing? | What was the cost per acquisition for each marketing channel last month? |
| Tell me about customers | What is the average order value by customer segment for the past 6 months? |
Principle 2: Include Time Ranges
Without a time range, the AI either analyzes all data (which may be huge) or guesses a reasonable window. Be explicit:
Good: "Revenue by month for January through June 2025"
Good: "Compare Q1 2025 to Q1 2024"
Good: "Customer signups in the last 90 days"
Okay: "Recent revenue trends" (AI picks a reasonable window)
Bad: "Revenue" (no time context at all)Principle 3: Name the Metrics You Want
If you know the column names in your data, use them. This reduces ambiguity:
Good: "Sum of order_total grouped by product_category"
Good: "Average response_time_hours by ticket_type"
Okay: "Total sales by product" (AI will guess which column is "sales")Pro Tip
Not sure what your columns are called? Ask the AI: "What columns are available in my data?" It will list the fields from your connected sources.
Principle 4: One Question at a Time
Multi-part questions sometimes get incomplete answers. Break them up:
Instead of:
"Show me revenue by product and also the top customers and
their purchase frequency and the month-over-month trend"
Do this:
1. "Show me revenue by product for Q4 2024"
2. "Who are the top 10 customers by total spend?"
3. "What is the purchase frequency for those top customers?"
4. "Show me the month-over-month revenue trend"Principle 5: Request Specific Outputs
Tell the AI what format you want:
"Show me a bar chart of revenue by region"
"Give me a table of all customers with more than 5 orders"
"Calculate the percentage breakdown, not absolute numbers"
"Sort by revenue descending and show the top 10"Principle 6: Use Follow-Ups to Drill Down
Start broad, then narrow with follow-up questions. The AI maintains context within a conversation:
1. "What were total sales by region last quarter?"
2. "Focus on the top region. Break it down by product."
3. "Which product had the biggest increase compared to the previous quarter?"
4. "What might explain that increase?"Principle 7: Ask for Explanations
The AI can do more than just pull numbers. Ask it to interpret:
"Why did revenue drop in March?"
"Are there any unusual patterns in this data?"
"What factors correlate with high customer lifetime value?"
"What would you recommend based on these trends?"Note: explanations are AI-generated interpretations, not definitive answers. Always apply your own business knowledge.
Common Patterns That Work Well
Trend Analysis
"Show me [metric] over time for [period]"
"What is the trend in daily active users for the past 3 months?"Comparison
"Compare [A] vs [B] by [metric]"
"How does this quarter compare to the same quarter last year?"Ranking
"What are the top/bottom N [items] by [metric]?"
"Which products have the lowest margins?"Composition
"What percentage of revenue comes from each channel?"
"Break down expenses by category"Anomaly Detection
"Are there any outliers in the data?"
"Which days had unusually high/low traffic?"What to Do When Results Are Wrong
- Check the data table in the response. If the raw numbers look wrong, the AI may have misinterpreted your column names.
- Rephrase the question with more specific column names or clearer aggregation instructions.
- Verify against a known value. If you know last month's total revenue, ask the AI for it and compare.
- Reduce scope. If analyzing multiple sources, try one source at a time to isolate the issue.
Related
Ready to try clariBI?
Start your free 14-day trial. No credit card required.