5 ways to get better answers from your AI research agent
Tips for writing better prompts, structuring multi-step queries, and getting the most out of workspace reports.
1. Be specific about what you want
Instead of “show me the data,” try “show me the top 10 respondents by completion rate for the March 2026 survey, grouped by region.” The more specific your request, the more precise the agent’s SQL query will be.
Vague requests lead to broad queries that may return too much data or miss the insight you’re looking for. Treat the agent like a very capable research assistant who needs clear instructions.
2. Use multi-step conversations
Don’t try to pack everything into one message. Start broad, then drill down. “How many responses did we get this month?” followed by “Break that down by region” followed by “Compare the Central region numbers with last month.”
The agent maintains conversation context, so each follow-up message builds on previous answers. This iterative approach often leads to insights you wouldn’t have found with a single query.
3. Ask for reports, not just answers
When you need to share findings, ask the agent to write a report to your workspace. “Create a dashboard showing this month’s survey trends” produces a shareable HTML page. “Write a summary report comparing Q1 and Q2 results” produces a formatted document.
Workspace reports persist and are accessible via browser, making them far more useful than chat responses for stakeholder communication.
4. Reference your project context
Use /project to select the right project before asking questions. Use /status to confirm your current context. When working with multiple projects, explicitly mention which project’s data you want to query.
The agent’s answers are only as good as the context it has. Make sure you’re pointed at the right database before asking complex questions.
5. Iterate on workspace outputs
After the agent creates a dashboard or report, you can refine it through conversation. “Add a filter for date range.” “Change the chart colors to match our brand.” “Add a print-friendly version.” Each instruction updates the file in place.
Think of workspace outputs as living documents that evolve through conversation, not one-shot deliverables.
