Response Style: customize your AI agent's voice
Citai Team
April 10, 2026 · 10 min read
The problem: generic tone doesn’t represent your brand
An unconfigured LLM writes generically. But your company has a voice: formal or casual, technical or approachable, brief or detailed.
What you can configure
Custom signature
HTML signature auto-appended to every generated response. Include name, title, phone, website.
Preferred tone
Default tone for responses: Professional, Friendly, Formal, or Concise. Changeable per response.
Custom instructions
Free-text rules injected into the LLM system prompt: “Always mention free shipping over $50”, “Never promise specific delivery dates”, “Use formal address”.
Format and length
Choose HTML or plain text, max length (short/medium/long), auto-greeting and auto-farewell toggles.
Where it applies
Style applies to quick-mode responses and thread-generated responses. Templates have their own fixed format and are not affected.
Tone calibration by industry
Each industry has different communication expectations. Here is the calibration guide:
Banking and financial services
- Recommended tone: Formal
- Key instructions: “Always use formal address”, “Never promise specific returns”, “Include legal disclaimers when mentioning financial products”
- Length: Medium to long — clients expect complete, detailed responses
- Example: “Dear Mr./Ms. {{client_name}}, regarding your inquiry about our investment products…”
E-commerce
- Recommended tone: Friendly or Professional
- Key instructions: “Always include order number”, “Offer alternatives when a product is unavailable”, “Use emojis sparingly”
- Length: Short to medium — quick, actionable responses
- Example: “Hi {{client_name}}! We checked your order #{{order_number}} and here’s the update…”
SaaS / Technology
- Recommended tone: Professional
- Key instructions: “Include links to relevant documentation”, “Use technical terms but explain briefly”, “Offer a clear next step”
- Length: Medium — balance technical detail with clarity
- Example: “{{client_name}}, thanks for reporting this issue. Our engineering team is investigating…”
Healthcare
- Recommended tone: Formal with empathy
- Key instructions: “Never give specific medical advice”, “Always recommend consulting a professional”, “Empathetic and calm tone”
- Length: Medium — informative but careful
Education
- Recommended tone: Friendly
- Key instructions: “Use accessible language”, “Include deadlines clearly”, “Offer additional resources”
- Length: Short to medium
How to detect and adapt tone automatically
Citai can semi-automatically adapt to the received email’s tone:
- Pre-analysis detects sentiment: If the client’s email is negative/urgent, Citai suggests Formal tone automatically
- Thread history: If the conversation started casual and escalated to a complaint, tone adjusts
- Conditional instructions: You can write rules like “If the client mentions ‘urgent’ or ‘complaint’, use Formal tone with apology”
Brand voice guidelines integration
How to document your brand voice
Use the custom instructions field as a mini brand guide:
- Words to USE: List words that represent your brand (e.g., “innovative”, “reliable”, “approachable”)
- Words to AVOID: List forbidden words (e.g., “cheap” -> use “affordable”, “problem” -> use “situation”)
- Preferred structure: Define how to open and close emails (e.g., “Always start by thanking the client”)
- Format rules: “Use bullet points for lists of 3+ items”, “Bold for key information”
Example of complete brand voice instructions
“We’re an edtech company. General tone: professional but approachable. Use first names. Always start by thanking the inquiry. Never say ‘we can’t’ — say ‘we suggest as an alternative’. Include link to docs.our-platform.com when relevant. Always sign off with ‘Need anything else? We’re here.’ before the signature.”
Measuring tone consistency across agents
The inconsistency problem
Without a defined style, the AI agent generates different responses each time. Even with a configured style, it’s important to verify:
Consistency indicators
- Instruction adherence: Do generated responses follow custom rules? Review 10 random responses per week
- Length variation: If you set “short” but some responses have 8 paragraphs, adjust instructions
- Prohibited vocabulary usage: Search email history for words you asked to avoid
- Correct format: Is the signature included? Is the configured greeting used?
Monthly audit process
- Export 20 random responses from history
- Verify against brand instructions
- Identify recurring deviations
- Adjust instructions to correct
- Document changes in an internal changelog
Tone variation examples
The same response in all 4 available tones:
Query: “What is the delivery time?”
Professional:
“Dear {{client_name}}, standard delivery time is 5-7 business days. For express shipping, you can select the option at checkout for an additional $10. We’re at your disposal.”
Friendly:
“Hi {{client_name}}! Deliveries take 5-7 business days. Need it faster? We have express shipping for just $10 extra. Anything else I can help with?”
Formal:
“Mr./Ms. {{client_name}}: In response to your inquiry, we inform you that the current delivery period is five (5) to seven (7) business days from order confirmation. For expedited delivery service, an additional charge of USD 10.00 applies. Sincerely.”
Concise:
“Delivery: 5-7 business days. Express: +$10. Anything else?”
Sidebar > Email Assistant > Response Style
Try Citai for free
Create your intelligent knowledge base in minutes. No credit card required.
Create free accountRelated articles
Email Templates: consistent replies in seconds
Create reusable templates with dynamic variables, multilingual support, and predefined seed templates to start instantly.
Read article → ProductEmail Assistant: reply to customer emails with AI in seconds
New Citai module that generates professional email replies using your knowledge base. Pre-analysis, 4 tones, embedded images and full history.
Read article → ProductEmail Threads: manage complete conversations with AI context
Organize emails into threads with automatic analysis, smart alerts, in-thread search, and response generation with full conversation context.
Read article →