Skip to main content

Informer Assistants Best Practices

Assistants make data analysis easier and more intuitive. To get the most out of them, follow a few best practices.

Presenting data​

Breaking data into smaller, distinct components helps the Assistant understand and query it effectively. For example, separating names into First Name and Last Name Fields lets you search by first name, last name, or full name. Combining them into one Field may require specific search formatting and complicate things.

Refining instructions​

Treat instruction-writing as a conversation:

  1. Ask the Assistant a query related to the behavior you want.
  2. Ask it to explain how it reached its answer (Debug mode shows the underlying steps).
  3. If an answer is wrong, tell the Assistant and continue until it produces the expected result.
  4. Ask the Assistant for instructions that would prevent the same mistake.
  5. Copy those into the Instructions section of the editor.

See Instructing Assistants.

Understanding AI Assistant limitations​

AI Assistants greatly enhance productivity, but it helps to understand their limits.

Why an Assistant might give an incorrect answer. AI models generate responses from patterns in data, not an inherent understanding of truth or logic. They may produce incorrect or misleading answers (especially for complex or niche queries), misinterpret ambiguous input, or lack real-time awareness. Even a confident-sounding answer can be wrong, particularly for numbers, calculations, names, and dates, which you should always double-check.

Answers are not always predictable. AI is probabilistic, not deterministic: the same question can yield slightly different answers, small changes in phrasing can change the response significantly, and the Assistant may "fill in the blanks" when context is missing.

Setting expectations. When an Assistant is not responding the way you want:

  • Be specific: Vague questions lead to vague answers.
  • Ask it to explain its answer: Understanding its reasoning shows where things went wrong.
  • Iterate and refine: Clarify, correct, and guide the Assistant.
  • Review and verify: Double-check factual or numerical output manually.
Also available through the REST API
Everything on this page is built on Informer's public REST API, the same one the product uses, so your scripts and integrations can do it too. Developer reference: Assistant API.