Apply prompting fundamentals

Well-structured prompts help Grafana Assistant deliver precise answers. This article covers the core prompting patterns you can apply across every workflow, plus guidance on when to start fresh conversations.

Before you begin

Active chat: Open the Assistant sidebar.
Objective: Know what outcome you want, such as dashboards, investigations, explanations, or navigation.
Context references: Identify any data sources, dashboards, or memories you plan to mention.

Structure prompts for clarity

Lead with the outcome. Verbs like Create, Explain, or List tell the Assistant what you expect before you mention any details. Follow with a timeframe and scope so the model knows which service, environment, or customer impact to consider.

Include examples or sample values whenever you expect a specific format or label, and break complex requests into smaller prompts because sequential instructions produce better, more reliable answers than a single long paragraph.

Use mentions and context wisely

Mention data sources, dashboards, or panels with @ to bind the prompt to the exact resource you want the Assistant to analyze.

Reference previous answers explicitly; phrases like Using the latency panel you created earlier signal that you expect the model to build on earlier work.

When you move to a different system or investigation, start a new conversation so the context stays focused and token usage remains low.

Iterate and correct

Read each response with a critical eye, and correct inaccuracies directly; simple guidance such as Use namespace=production instead keeps the conversation on track.

Ask for validation, alternatives, or summaries when you want confirmation that the Assistant understood the task.

If a response drifts from the goal, restate the requirement before moving to the next prompt so the updated context carries forward.

Combine prompts with workflows

Apply these prompting patterns inside the workflow hubs to guide incident response, dashboard creation, or query tuning.

When a prompt works well, add it to your team playbook so others can reuse it without experimentation.

Plan follow-up prompts that transition smoothly between hubs. For example, move from data analysis to dashboard management by referencing the query you just validated.