Resources

Products

Customize the agent persona and tone of voice

Modified on: Thu, 14 May, 2026 at 12:09 AM

TABLE OF CONTENTS

The Instructions page is where you shape how your AI agent presents itself and communicates with employees. Two settings work together to define this: Define your business context establishes the agent's persona and the world it operates in, and Set custom instructions defines the voice, tone, and behavioral rules the agent follows in every conversation.

Getting these two settings right means employees receive responses that feel consistent, helpful, and appropriate for your organization, rather than generic or robotic.

This article explains what each setting does, how to fill it in effectively, and provides IT support examples you can adapt for your own context.

Where to find Instructions

Instructions are configured within your agent's Build section. To get there:

  1. Open AI Agent Studio and select the agent you want to configure.

  2. In the left navigation, expand Build and click Instructions.

Define your business context

The Define your business context field is where you describe the environment the agent operates in: what your organization does, what kinds of requests the agent handles, and who it is talking to. This is the agent's persona foundation.

The field accepts up to 300 characters. Keep it focused and factual. The agent reads this context every time it responds, so the clearer and more specific this description is, the more accurately the agent will interpret employee requests and tailor its replies.


What to include

A strong business context description covers three things:

  • Who the agent supports. Name the audience. For example: employees across the organization, new joiners, or the IT support team.

  • What it helps with. List the main request types the agent handles. Be specific rather than broad.

  • Why speed and accuracy matter. A sentence about the nature of the service, such as high request volumes or the need to reduce manual effort, helps the agent prioritize resolution over conversation.


Example: IT support context


Employees across the organization rely on internal IT teams for support with password resets, VPN connectivity, software access requests, laptop provisioning, email issues, and troubleshooting workplace applications used for day-to-day operations.


Support teams typically handle a high volume of repetitive requests across multiple systems, making it important to automate common IT workflows, reduce resolution time, and provide employees with faster, always-available assistance.


This example tells the agent it is operating in an internal IT support context, gives it a clear list of the request types it will encounter, and signals that speed and automation are priorities. That context shapes how the agent interprets ambiguous requests and how it frames its responses.


Use the AI assist button

Below the business context field there is a sparkle icon (AI assist) with a dropdown arrow. Click it to let the AI generate a business context description based on your agent's existing configuration, such as the workflows and knowledge sources already connected.

Use AI assist as a starting draft rather than a final answer. Review the generated text and edit it to reflect the specifics of your organization before saving.


Tip: The character limit is 300. If your first draft is too long, cut anything that describes how the agent should behave — that belongs in custom instructions, not here. Business context should only describe the situation, not the rules.


Write the business context

  1. On the Instructions page, click inside the Define your business context field.

  2. Type a description of your organization's service context, or click the AI assist sparkle button to generate a draft.

  3. Review and edit the text. Check that it names the audience, lists the main request types, and reflects the service's nature.

  4. The character counter below the field shows your usage against the 300-character limit.


Set custom instructions

The Set custom instructions field is where you define the agent's voice and tone, and set the behavioral rules it follows across every interaction. While business context tells the agent what it is working with, custom instructions tell it how to communicate and how to behave.

These instructions apply globally. Every response the agent generates, across all workflows and all channels, is shaped by what you write here.


Voice and tone

Voice is the consistent personality of the agent. Tone is how that personality adjusts to context. An IT support agent might have a voice that is knowledgeable and reassuring, but its tone when handling an urgent system outage should be calm and direct, while its tone for a routine password reset can be warm and quick.

When writing your custom instructions, define both:

  • Voice: the qualities that stay the same in every response, such as being professional, approachable, and clear.

  • Tone adjustments: how the agent modulates based on the situation, such as being more empathetic when an employee is frustrated, or more concise when responding to a straightforward request.


What to cover in your custom instructions

Strong custom instructions address four areas:


1. Communication style

Tell the agent how to phrase its responses. Specify whether it should use plain language, avoid technical jargon, keep responses short, or always confirm before taking action.


Use natural, conversational language without jargon.

Keep responses short and focused on the next step the employee needs to take.

Avoid robotic or overly formal phrasing.


Write this: "I can help you with that right away."

Not this: "Your query has been received and will be processed."


2. Empathy and handling frustration

Define how the agent should respond when an employee is clearly frustrated, has been waiting a long time, or has a recurring issue.


Always acknowledge the inconvenience before moving to the resolution.

If an employee mentions they have already tried something, do not suggest it again.

When an issue cannot be resolved immediately, reassure the employee that it is being escalated and provide a reference number.


3. Escalation behavior

Tell the agent when and how to hand off to a human agent, and what to say when it does.


If the issue cannot be resolved within two interactions, offer to raise a ticket and connect the employee with the IT support team.

When escalating, tell the employee what will happen next: "I have created a ticket for you. A support engineer will follow up within 4 business hours."

Never leave an employee without a next step, even if the next step is a ticket reference number.


4. Boundaries and data handling

Set clear rules about what the agent should and should not do, especially around sensitive information.


Do not share internal system details, backend configuration, or diagnostic outputs with employees.

Always confirm the employee's identity before processing account changes.

Do not store or repeat sensitive information such as passwords or personal identification numbers in the conversation.


Full custom instructions example: IT support agent

The following is a complete example of custom instructions for an IT support agent. Use it as a reference when writing your own.


Voice and tone

Maintain a friendly yet professional tone: confident, helpful, and concise.

Use natural, conversational language without jargon.

Avoid sounding robotic or overly formal.

Always show empathy and aim for resolution with clarity and speed.


Communication style

Confirm the employee's issue in one sentence before starting the resolution steps.

Use numbered steps when guiding an employee through a process.

Ask one question at a time. Do not present multiple questions in a single message.


Empathy and frustration

If an employee expresses frustration, acknowledge it before proceeding: "I understand this has been disruptive. Let me sort this out for you now."

If the same issue has occurred before, note it and prioritize escalation.


Escalation

If an issue is not resolved after two attempts, offer to raise a ticket automatically.

When escalating, always provide the ticket number and the expected response time.

Never end a conversation with an unresolved issue and no follow-up path.


Boundaries

Do not share internal system details or backend error logs.

Verify the employee's identity before making any account changes.

Do not repeat sensitive data in the conversation.


Voice and tone: quick reference

The table below shows the difference between instructions that produce natural responses and those that produce robotic ones.


Write this

Not this

I can help you reset your password right now. Can you confirm your employee ID?

Your password reset request has been received. Please provide your employee identification number to proceed.

I understand this is frustrating. Let me check what is happening with your VPN access.

Your issue has been noted. An investigation will be initiated.

It looks like your account is locked. I will unlock it now and send you a confirmation.

Account lock detected. Unlock process will be executed upon verification.

I have raised a ticket for you. A support engineer will follow up within 4 business hours.

Ticket created. Resolution timeline subject to SLA parameters.


Save your instructions

Instructions are saved automatically as you type. There is no separate save button on the Instructions page. Changes take effect immediately and apply to all new conversations the agent starts after the update.


Note: Instructions apply to all workflows and channels the agent is deployed on. If you want different behavior for a specific workflow, use the custom response node within that workflow to override the global instructions for that step.


Best practices

  • Be specific, not general. "Be helpful" gives the agent no useful direction. "Confirm the employee's issue before starting resolution steps" is actionable and produces a measurable difference in responses.

  • Write instructions the way you would brief a new team member. If it would make sense to say it to a person joining your support team on their first day, it belongs in custom instructions.

  • Separate context from behavior. Business context describes the situation. Custom instructions describe the rules. Keep them distinct to avoid confusion.

  • Test after every change. Use Preview AI agent to run through a few common IT requests after updating instructions. Check that the tone matches what you intended before the change goes live.

  • Revisit instructions after deployments. As you add new workflows or deploy to new channels, check that the existing instructions still produce the right behavior across all scenarios.


Frequently asked questions

Do custom instructions override workflow-level behavior?

Custom instructions set the global defaults. Individual workflows can include custom response nodes that override the global tone for a specific step, but the overall voice and behavioral rules in custom instructions remain in effect everywhere else.

Is there a character limit for custom instructions?

The business context field has a 300-character limit. The custom instructions field does not display a character limit in the product, but keeping instructions focused and concise produces better and more consistent agent behavior than very long or complex rule sets.

Can I use the AI assist button for custom instructions too?

The AI assist button is available in the business context field. For custom instructions, you write the rules yourself. Use the example in this article as a starting template and adapt it to your organization's requirements.

How quickly do instruction changes take effect?

Changes to instructions apply immediately to new conversations. Conversations already in progress are not affected by mid-session changes.

Should instructions be written in first person or second person?

Write instructions in the imperative form, as direct rules: "Use plain language." "Confirm identity before making changes." This is the clearest format for the agent to follow. Avoid writing instructions as descriptions of what the agent might do.


What's next

After configuring your agent's persona, voice, and tone, you are ready to:

  • Add knowledge sources so the agent can answer questions accurately alongside its configured persona.

  • Connect workflows so the agent can take action, not just respond.

  • Preview and test the agent to confirm that its tone and behavior match your expectations before going live.

  • Deploy the agent on the Support Portal, Microsoft Teams, or Slack.


Need help? Contact your Customer Success Manager or Freshservice Support at support@freshservice.com for help with configuring agent instructions in Freddy AI Agent Studio.