AI writing guidelines

Overview 

Scope 

Applies to: 

• AI-generated or AI-assisted content shown to customers or employees 
• AI summaries, suggested replies, autogenerated fields, and chat responses 
• Any experience where a model produces text or recommendations 

Voice and tone 

Genesys Agentic AI is professional yet conversational. Instructions may be technical, but user-facing language should feel human, clear, and helpful.  
Different types of messaging serve different purposes within our voice and tone. Informational text should highlight the value and motivate users by showing how AI is helpful and time-saving. In contrast, disclaimer text should make clear that users are responsible for reviewing AI output for accuracy, for liability reasons. Go to the Disclaimer & Labeling to learn more.  

• Speak in first person when needed (“I” / “we”) but avoid heavy personification. 
  Positioning: an objective companion, not a best friend.
• Use the present tense, short sentences, and everyday language.
• Prefer direct, supportive phrasing over overly enthusiastic or emotional language. 

Do 

• “I can’t verify that yet.”
• “Here’s what I need to continue.”
• “That didn’t work. Try this next.” 

Avoid 

• “I’m so excited to help you!!!”
• “Don’t worry, I’ll take care of everything.” (implies action/guarantee) 

Core principles 

 

Clear 

Say exactly what the user needs to do. 

• Put the action first.
• Use specific nouns and verbs.
• If something fails, explain why and what to do next. 

 

Concise 

Keep it short and direct. 

• One idea per sentence.
• Remove filler (“just,” “simply,” “actually”). 

 

Consistent

 Use the same terms for the same things. 

• Maintain stable naming across UI, docs, and agent responses.
• Avoid switching between near-synonyms (“ticket/case/request”) unless clarified. 

 

Conversational tone 

Friendly, human, and natural. 

• Write like a helpful colleague.
• Avoid robotic phrasing, but don’t over-humanize. 

 

Contextual guidance 

Tailor copy to the user task and channel. 

• Use the channel’s constraints (voice vs chat vs UI panel).
• Provide the smallest next step that moves the task forward. 

 

Inclusive and accessible copy 

Use plain language 

• Prefer common words: “start” over “initiate,” “help” over “assist.”
• Explain necessary technical terms briefly. 

Avoid jargon, idioms, or acronyms 

• If an acronym must be used, spell it out once.
• Avoid culture-specific phrases (“hit it out of the park,” “low-hanging fruit”). 

Write with screen readers in mind 

• Ensure labels are meaningful out of context.
• Avoid relying on color/position (“click the green button on the right”).
• Keep error messages tied to the field and include what to fix. 

Focus on key information 

• Provide only what’s necessary to complete the step.
• Use short lists for multi-step instructions. 

Be culturally sensitive 

• Avoid humor or slang in high-stress moments.
• Keep tone respectful and neutral. 

 
Key Terminology 

Actions terms 

What our AI does when it produces output for our customers.  

• Generate
• When AI writes it (i.e., summaries)
• Select
• When AI selects it (i.e., radio buttons)
• Use the word “Option” to identify the item that was selected. 
• Detect
• When AI detects it (i.e., anomaly detection)
• Predict
• When AI predicts it (i.e., values) 

Button label 

Buttons that trigger AI interactions:
 
Auto-Fill [item]
 
Examples:

• Auto-Fill form
• Auto-Fill summary

Auto-Predict [item]
Auto-Create [item]

Disclaimer Messaging  

By law, we are required to inform users whenever they are interacting with AI. More than this, disclaimer text helps maintain trust with our users and their end customers. It also ensures users stay informed, in control, and confident when interacting with AI. 

Disclaimer Guidelines 

1. Passive voice is used to avoid overly personifying AI 
2. Does not include any call to action (i.e., “review for accuracy”) 
3. Provide a “Learn more” hyperlink directing users to documentation  
   A. When there’s enough space, use the long version, “Learn more about how we use AI.”  
4. Include data source citations when applicable 

 
Disclaimer Messaging Templates  

Here are common variations to ensure messaging remains clear and consistent across components with different space constraints and contexts. Use consistent terminology listed in these guidelines and reach out to the UX Content team for any outlier use cases that require different verbiage. 

Shortest to longest variations: 

• The [item] was/were [action] by AI.  
• The [item] was/were [action] by AI. Learn more.  

 Examples: 

• The name was generated by AI
• The notes were generated by AI. Learn more. 
• An anomaly was detected by AI. 
• This option was selected by AI. 
• These values were predicted by AI.  

 Variations with insights: 

• The [item] was/were [action] by AI based on [source].
• The [item] was/were [action] by AI based on [source]. Learn more. 
• The [item] was/were [action] by AI using information from [source].
• The [item] was/were [action] by AI using information from [source]. Learn more.  

 Examples: 

• The summary was generated by AI based on the session’s transcripts. 
• These values were predicted by AI based on past performance. Learn more. 
• This summary was generated by AI using information from the conversation’s transcript. Learn more about how we use AI. 

 

Disclaimer: One-Time Messaging  

Proposal for a one-time Terms & Agreement messaging for users to have to click to acknowledge that they have read and understand that AI still requires some human interaction.  
 
More specifically, this message ensures: 

• Clarify limitations → Communicate what the AI can and can’t do
• Reinforce user responsibility → Remind users to review and validate information
• Reduce risk → Prevent misuse, overreliance, or incorrect decisions
• Support transparency → Make AI behavior more understandable and predictable
• Align with compliance → Meet legal, ethical, and responsible AI standards