Plans and permissions:
- Add-on: AI agents, billed separately
- Edit critical instructions and profile-specific handoff instructions: Admin, Manager, or Team Lead role, or a custom role with Chat agent settings set to Can edit chat agent settings
- Edit a brand’s chat style: Admin role
- Edit topic-based handoff triggers: Admin role, or a custom role with Handoff settings set to Can edit handoff settings
- Create and edit workflows: Admin, Manager, or Team Lead role, or a custom role with Workflows set to Can create, edit, and delete workflows
Use the chat agent’s instruction settings to control how it sounds, how it answers, when it hands off, and what it does, so it resolves more conversations and behaves consistently. The chat agent doesn’t run on one big prompt: each setting is used at a different moment to make a different decision. Putting each piece of guidance in the right place matters more than adding more of it.
To find most of these settings, go to AI support > Chat agent and open the chat agent.
How the settings work together
Each setting answers one question:
| Setting | Where to find it | What it controls |
|---|---|---|
| Chat style | Configure > Brand studio, the brand’s Style tab, Chat style | How responses sound |
| Critical instructions | The chat agent’s Setup tab, Critical instructions for chat agent | How the chat agent answers when it replies |
| Topic-based handoff triggers | Configure > Handoff | Topics that always go to your team |
| Profile-specific handoff instructions | The chat agent’s Handoff tab, under Handoff criteria | How the chat agent decides when to hand off |
| Workflows | AI support > Workflows | What the chat agent does, step by step |
For each conversation, the chat agent works out whether to hand off, then whether a workflow should handle the request, then how to answer, and finally how the answer should sound.
Note: The chat agent uses its brand’s chat style only when Chat style on its Setup tab is set to Use customized brand style. See Brand studio.
Chat style
The chat style controls tone and delivery. It applies to every response, including responses inside workflows. Write it on the brand’s Style tab, in Chat style.
What to put in the chat style
- Tone, such as warm, neutral, formal, or concise
- How long or detailed responses should be
- How to show empathy
- Formatting preferences
- Phrases to use or avoid
For example: “Be calm, direct, and reassuring.” “Avoid over-apologizing.” “Prefer short paragraphs and bullets.”
What to leave out
Don’t use the chat style for handoff logic, business rules, process steps, or decision logic. If two people gave the same answer, the chat style is what makes them feel different.
Critical instructions
Critical instructions guide how the chat agent answers when it replies in conversation: how to respond, not what action to take. It follows them in every interaction, so keep them to what matters most. Write them on the chat agent’s Setup tab, in Critical instructions for chat agent.
What to put in critical instructions
- Whether to attempt an answer or ask a clarifying question
- How to handle ambiguous requests
- How confident or cautious to be
- Guardrails that apply to every conversation
For example: “Attempt a best-effort answer when confidence is high.” “Ask clarifying questions if multiple interpretations are possible.” “Avoid speculation when information is incomplete.”
What to leave out
Don’t use critical instructions for handoff conditions, workflow steps or logic, detailed policies, or tone (use the chat style).
How many to write
Aim for about 5 to 15. More isn’t better: unclear, overlapping, or very detailed instructions make responses less consistent.
Topic-based handoff triggers
Topic-based handoff triggers send conversations about a topic straight to your team. When a trigger matches, the chat agent hands off immediately. Set them up in Configure > Handoff, on the Topic-based handoff tab, and turn each one on for chat.
When to use them
- Legal threats
- Safety or emergency situations
- Highly regulated requests
- Customer groups who should always talk to a person
When not to use them
Don’t use them for general confusion, long conversations, or common topics that workflows can handle. Every trigger trades automation for certainty, so use them sparingly.
Profile-specific handoff instructions
Profile-specific handoff instructions guide the chat agent’s judgment about when to hand off. They influence the decision, but don’t force a handoff. Write them on the chat agent’s Handoff tab, under Handoff criteria. Think of them as how a strong support team member would decide when to escalate.
What to put in handoff instructions
- What “urgent” means in your business
- When to wait before handing off, so automation has a chance to resolve the issue
- How to read emotional language in your context
- Guidance such as “Prefer workflows when they can fully resolve the request” or “Hand off only after several unsuccessful attempts”
What to leave out
Don’t use them for step-by-step procedures, matching exact phrases, or policy text copied from internal documents.
Workflows
Workflows handle structured, repeatable actions. Set them up in AI support > Workflows. See Building workflows.
When to use them
- Processes with several steps
- Collecting information from the customer
- Verifying the customer’s identity
- Actions such as returns, cancellations, updates, or lookups
If the chat agent often explains how to do something instead of doing it, a workflow is probably missing, or an existing one is too narrow.
When not to use them
Don’t use workflows for simple explanations, one-off questions, or tone and style.
Write your settings with an LLM
You can use a large language model (LLM), such as ChatGPT or Claude, to draft your chat style, critical instructions, and handoff instructions. You don’t need prompt-writing experience. You answer a few questions about your business, paste your answers into the prompt below, and copy each part of the result into its setting.
Answer questions about your business
Write short answers to these questions. They describe how you want the chat agent to behave, not how it works.
- What is the chat agent responsible for? One to three sentences. For example: “Handle common how-to questions and order status, and hand off billing disputes.”
- Who is the main audience? For example: different types of customers, or enterprise admins.
- What should it handle end to end? Three to seven things it should fully resolve without a person, such as order status, account access, or basic billing questions.
- What should always go to a person? Hard boundaries, such as legal threats, safety concerns, refunds over a certain amount, or cancellations for VIP customers.
- How should responses sound? For example: “Warm, calm, and concise” or “Professional and neutral. Avoid emojis and slang.” This becomes your chat style.
- How should it answer? For example: “Attempt a best-effort answer before asking follow-ups” or “Avoid speculation if information is missing.” This becomes your critical instructions.
- How should handoffs work? Optional. For example: “Emotional language alone shouldn’t trigger a handoff” or “Hand off only after several unsuccessful attempts.” This becomes your profile-specific handoff instructions.
Run the prompt
Copy this prompt into your LLM, and replace each <<PASTE ANSWER>> with your answer to the matching question.
You are an expert in designing AI chat configurations for customer support.
Your task is to generate clean, minimal configuration content for an AI chat system.
Do not add extra rules or guess policies.
Here is information about our business and goals:
1) What this AI chat is used for:
<<PASTE ANSWER>>
2) Primary audience:
<<PASTE ANSWER>>
3) What the AI chat should handle end-to-end:
<<PASTE ANSWER>>
4) What should always escalate to a support team member:
<<PASTE ANSWER>>
5) Desired tone and style:
<<PASTE ANSWER>>
6) How the assistant should behave when answering:
<<PASTE ANSWER>>
7) Escalation philosophy (optional):
<<PASTE ANSWER>>
Using this information, generate the following sections:
1) Chat style
- Focus only on tone, verbosity, empathy, and phrasing.
- Do not include business logic or escalation rules.
2) Profile-specific handoff instructions (only if applicable)
- Focus on how to decide when to escalate vs continue handling automatically.
- Do not include step-by-step processes or exact phrase matching.
3) Critical instructions for chat agent
- Include only high-priority guidance for how the assistant should answer when responding conversationally.
- Do not include tone, escalation rules, or workflow steps.
Output the sections clearly labeled.
Do not include explanations outside the sections.Review the output
Before you add the output to Assembled, check that:
- Tone and phrasing are only in the chat style.
- Handoff guidance is only in the handoff instructions.
- The critical instructions are short, ideally fewer than 15 bullets.
- No instruction appears in more than one section.
If something looks wrong, change your answers and run the prompt again, rather than editing the output by hand.
Add the output to Assembled
Copy each section into its own setting. Don’t combine them into one field.
| Output section | Where it goes |
|---|---|
| Chat style | Configure > Brand studio, the brand’s Style tab, Chat style. Then set Chat style on the chat agent’s Setup tab to Use customized brand style. |
| Profile-specific handoff instructions | The chat agent’s Handoff tab, Profile-specific handoff instructions |
| Critical instructions for chat agent | The chat agent’s Setup tab, Critical instructions for chat agent |
Select Save changes on each page.
Get ideas for handoff triggers
If you’re not sure whether you need topic-based handoff triggers, ask the LLM: “Based on what I described, suggest a small set of optional handoff rules I should consider.” Add only the ones that are true hard stops.
Keep your setup simple
- Start with fewer, clearer instructions.
- Use workflows for anything that should happen the same way every time.
- Refine existing instructions rather than adding new ones.
- Keep each kind of guidance in its own setting: tone in the chat style, answering guidance in critical instructions, and handoff guidance in the handoff settings.
- Don’t add rules just in case, copy internal policy text word for word, or add an instruction to fix one bad response.
If a response isn’t what you expect, check the chat style first, then whether a workflow should handle the request, and then refine your instructions.