# AI Reports best practices > Best practices for getting accurate, useful and cost-effective AI reports in NIZU Cloud. (Apps: AI Reports) - Page: https://dashboard.nizu.io/knowledge_base/view/161-ai-reports-best-practices - Part of: [Knowledge base](https://dashboard.nizu.io/knowledge_base/llms.txt) on NIZU — https://dashboard.nizu.io/llms.txt - Updated: 2026-11-13T15:10:04Z ## AI Reports best practices ### Write better requests - State the **decision** you want help with, e.g. “which clients should we chase for payment first?” - Ask for **one theme per report** rather than mixing revenue, projects and tasks in a single request. - Mention the **comparison** you care about (month over month, by client, by team member). ### Trust and accuracy - Remember that **NIZU computes every figure**; the AI only explains them. The numbers in metric cards, tables and charts come straight from your data. - Use the expandable data view and the metric cards to **sanity-check** a report before sharing it. - If a narrative claim ever seems off, re-run the report — and rate it *Bad* with a note so you have a record. ### Privacy and data minimisation - AI Reports sends **aggregated figures, not raw records**. Keep it that way: prefer summary sections over requests that would need row-level detail. - Use the **maximum rows** setting to cap how much data any source aggregates. - Each workspace should use **its own API key** so usage and billing stay under your control. ### Cost control - Templates are tuned to be concise — start there. - Generation cost scales with the amount of data and the model you pick. Choose a balanced model for routine reports and a stronger model only when needed. - Re-run reports when something has changed, not on every page visit. ### Security - Grant the **Settings** permission only to people who should manage API keys. - Reports are rendered from structured data, so no AI-authored code is executed — but still review who can **Share** reports.