Adjusting the Tone of Voice for the AI Assistant

This article will explain the tone of voice configuration for the AI Assistant in the Admin Center.

What is the tone of voice configuration?

Setting up a tone of voice in the AI assistant allows you to specify the brand tone of voice and off-brand language, and then apply that tone of voice to all or selected AI prompt templates. This helps users to use appropriate corporate language and to protect the brand. 

Why does the tone of voice matter?

In instructing AI models, the tone of voice matters to companies, because the wording of the output can significantly differ depending on context given in a prompt. Tone of Voice reflects your organization's character in written and spoken communication. Setting the tone of voice centrally ensures that the GenAI outputs stay consistent with your brand and preferred style of writing.

How to adjust the tone of voice

After the initial set up, owners/admins can adjust the tone of voice anytime:

1. Click on the Tone of voice button.
2. Make the desired changes to the tone of voice.
   • Specific language descriptors can be chosen to create the desired tone of voice.
   • Tone of voice can also be updated as custom text to match existing brand messaging.
3. Click Save.
4. Click Publish changes and Confirm to publish the changes made to the tone of voice.

The tone of voice will be updated everywhere it is in use. To be applied to actions, the tone of voice should be used within the prompt template at the desired placed using {{ToneOfVoice}}. This allows for the creation of custom AI prompt templates leveraging a specific tone of voice that may need to be different from the one set up in the Admin Center or omitting a tone of voice altogether. 

Tone of voice best practices

Many organizations have long, detailed tone of voice guidelines as part of their brand identity. While these can be useful for humans to reference, LLMs generally need much more concise instructions. The following tips should be kept in mind when writing the tone of voice for your AI Assistant.

1. Start with the purpose of the tone. 
   LLMs respond well when you specify why the tone exists, and it can base decisions off of this brief purpose statement.
   Example: "Write in a tone that feels human, builds trust, and helps readers understand complex topics quickly."
2. Summarize long guidelines into 3-6 core behaviors.
   When summarizing long tone of voice guidelines, try to avoid compressing everything equally. Instead, extract the most important behaviors that actually change the writing. This also can include your brand's distinct or signature details. For example:
   • Sentence structure (long, short, varied?).
   • Level of formality.
   • Attitude.
   • Use of jargon.
   • Use of humor.
   • Perspective (first person, second person?).
   • What to avoid. 
3. Use observable language.
   LLMs struggle with vague marketing terms. Instead, use actionable instructions. Especially when summarizing tone of voice guidelines, ensure high-level values are rephrased into executable writing behaviors. Avoid abstract ideas like "authentic", "bold", "delightful", etc. 
4. Tell the LLM what to prioritize
   LLMs can sometimes focus in on one element, so if your brand has a hierarchy of what to prioritize, specify this. 
   Example: "Clarity comes first, friendliness is second, and humor can be used lightly unless explaining something complex."
5. Avoid overloading with too many rules
   Each rule given to a LLM dilutes the others. This means that LLMs are much more likely to follow a few very clear lines of instruction than they are to follow a 2+ page tone of voice statement.