AI Guidelines

AI Guidelines control how AI Assistant communicates and which communication rules apply to a particular Zendesk context. Open Sources → AI Guidelines. Procedures and guidelines have different jobs:

• Procedures define facts, decisions, and operational steps.
• AI Guidelines define response behavior such as tone, terminology, formatting, language-specific rules, and customer-facing communication policy.

​

Response versus Translate guidelines

Assistant guidelines can be assigned to two use cases: Translate-specific guidelines are available for AI Assistant. The public AI Chatbot does not use Assistant-only Translate guidelines.

​

Why separate Translate guidance?

Translation often needs rules that are inappropriate for normal generation. For example, a team may require a specific French term for subscription, preserve order numbers and product SKUs exactly, or prohibit translation of registered product names. Those instructions should affect Translate, but they should not be treated as general response policy for every ticket.

​

Target a guideline

Guidelines can be targeted using dimensions including:

• channel;
• tags;
• Zendesk groups;
• Zendesk organizations;
• language; and
• match mode.

The current channel values include Email, Chat, Social, and Phone.

​

Match all versus match any

Use the targeting mode to decide whether all configured dimensions must match or whether a match in an allowed dimension is sufficient. Be deliberate here. A beautifully written policy attached to the wrong group is still the wrong policy, just with better punctuation.

​

Language targeting

Language targeting is especially important for Translate guidelines. A Translate guideline can apply only when the translation’s target language matches the configured language. This lets you maintain Spanish terminology rules for Spanish output, French legal wording for French output, or locale-specific style instructions without polluting other translations.

​

Write effective Response guidelines

Good response guidance is specific and observable. Define things such as:

• whether replies should be concise or detailed;
• whether to lead with the resolution;
• phrases or claims RAUM must avoid;
• formatting rules;
• terminology requirements; and
• when information must not be promised unless a procedure confirms eligibility.

Avoid instructions such as Be helpful and sound like our brand. Emotionally satisfying, operationally close to nothing.

​

Brand Voice versus AI Guidelines

AI Guidelines are authored rules. Brand Voice is mined from phrasing your agents repeatedly use in real public replies. Use Guidelines for policy and intentional style. Use Brand Voice to preserve recurring language patterns that actually exist in your team. Learn about Brand Voice →

​

Review changes

After editing a high-impact guideline:

1. use Copilot Mode on representative tickets;
2. test every targeted channel or language;
3. use the Translate button for Translate-specific rules;
4. check Ticket Activity to see which guidelines were used; and
5. run Simulation when the response is coupled to procedures or actions.