---
title: Hub-wide Ground Rules
slug: editorial/hub-wide-ground-rules
docTags: 
createdAt: 2026-09-02T19:27:36.445Z
---

## Summary

**Ground rules** are house rules that must hold for every prompt in your Purple Hub — tone of voice, brand voice, spelling conventions, legal constraints, topics to avoid. Instead of repeating them in every single prompt, you maintain them once centrally, and Purple Prompts automatically adds them to every request it sends to the language model.

## Configuring the ground rules

Ground rules are edited in **'Purple Prompts' ❭ 'Settings'**, in the **Ground Rules** text area. Only administrators can open this settings page. The same page is where the API keys and the model for each job are configured, see [AI Models and API Keys](docId\:zTwrSgVsiEYd2-Lu5-fNo).



::Image[]{src="https://api.archbee.com/api/optimize/ygR6KtT9QI_R0u3uKTJsm/ITRs95_KXVZtRXyvHW4Mc_screenshot-2026-09-02-at-220018.png" size="70" isUploading="false" width="1450" height="882" darkWidth="1450" darkHeight="882" showCaption="false"}

The field is empty by default. As long as it is empty, every prompt behaves exactly as it did before — nothing is added.

Write the rules in whatever language fits your content most.

## Where the ground rules apply

The rules are added directly in front of the instructions of the individual prompt, for every feature that runs a prompt:

- prompts run from the Purple Prompts sidebar in the article view (see [Using Purple Prompts in the article view](docId\:GK6a6pboXGzNOblSOCw7k))
- the whole prompt sequence behind the **Finalize Article** button (see [Finalize Article](docId:2-bngePn8V16nhvnw7u5v))
- the AI metadata suggestions for images in Purple DAM (see [Uploading Media](docId\:XcDxtroOPSUfsGYaRVQ1m))
- prompts of the usage area **External Workflow**, which external automations fetch from the Hub

An individual prompt still wins where it needs to: the ground rules are handed to the model with the note that they apply unless the prompt's own instructions explicitly override them. So a prompt that has to produce, say, an all-lowercase value can still say so and will be followed.

## Keep them short

The ground rules are sent along with **every** request, so they add to the execution time and API cost of every single prompt, and a very long rule set can crowd out the instructions of the prompt itself. Keep them to the few rules that genuinely have to hold everywhere, and leave everything task-specific in the individual prompts.

## Example

:::BlockQuote
Always address the reader formally ("Sie"), never informally ("du").
Write product names exactly as they are spelled by the manufacturer.
Never use exclamation marks.
Do not mention competitors by name.
:::

For more on phrasing instructions that language models follow reliably, see [Prompt Writing Best Practices](docId\:XeiYYpZnVOcj5sC6AZUu9).