---
title: "Use @ mentions for better results | Grafana Labs"
description: "Practice @ mentions in Grafana Assistant so prompts target a specific data source."
---

> For a curated documentation index, see [llms.txt](/llms.txt). For the complete documentation index, see [llms-full.txt](/llms-full.txt).

# Use @ mentions for better results

At-mentions specify which data source, dashboard, or panel the request should use. Without a mention, the reply might use a different data source or request more detail. With a mention, you point the request at the right place so answers are more accurate.

Use this rule when you choose:

- **Mention** when you already know which data source (or dashboard/panel) should answer the question.
- **Omit** when you are exploring what is available and don’t have a target yet.
- **Retry with a mention** when the reply doesn’t use the source you meant, or stays too generic.

Here you send a vague prompt first, then the same kind of request with an at-mention so you can compare focus.

To compare a vague prompt with an at-mention, complete the following steps:

1. Open **Assistant** from the top toolbar (sparkle icon) if the panel is closed.
2. In the chat input, paste or type a vague prompt (no at-mention):
   
   > Show me something useful from my stack
   
   Or try:
   
   > What can you tell me about my data sources?
3. Click **Send**.
4. In a new message, type the at symbol (@) in the chat input.
   
   Pick a data source from the mention list (or type the name after the at symbol). Prefer one you already know from this path, for example a Prometheus or Loki source.
5. Finish the prompt so it uses that mention.
   
   Example shapes (replace the mention with yours):
   
   > Summarize what I can query from @prometheus-ds
   
   > Show recent activity from @loki-logs
   
   > List metrics available in @grafanacloud-usage
6. Click **Send**.
7. Compare the two replies for focus, not perfection:
   
   - The **mentioned** reply should stay scoped to the data source you picked (name it, include a query for it, or otherwise stay clearly limited to it).
   - The **vague** reply might cover several sources, use a different source, or request more detail.
   
   If the mentioned reply is clearer, keep using `@` when you know the target. If you only needed a tour of what exists, omitting `@` is fine.

You practiced turning a vague request into a targeted one with an at-mention, and when to mention versus explore without one.

For more examples, refer to [Use mentions for better results](/docs/grafana-cloud/machine-learning/assistant/get-started/grafana-cloud/#use-mentions-for-better-results).

In the next milestone, you’ll choose Learn mode in the Assistant panel.