Inbox AI Replies: Rule Best Practices and Examples

Review best practices and example rules for Inbox AI Replies so you can improve reply quality, reduce escalation risk, and match your team’s communication policies.

1. Best practices for writing rules

One rule = one topic

Each rule should address exactly one scenario or question type. Don't combine unrelated instructions.
 

Use the General tab for settings across all listings

The General tab has two default settings: Tone of voice (with a custom text field for additional instructions) and Response length. Language preferences, formatting, and sign-off templates should go in the Tone of voice custom field — not in standalone rules.


 

Be explicit about what not to do

The AI benefits from negative instructions. If there's a behavior you want to avoid, state it clearly.
 

Split conditional logic into separate rules

If a response depends on a condition (e.g., calendar availability), create one rule per branch rather than one rule with if/else logic.
 

Use tags for listing groups

Instead of writing listing names or cities into rule text, use tags to target rules to specific listing groups. Tags update dynamically — if you add/remove a listing from a tag, the rules automatically adjust.


 

Draft for complex, Auto-Reply for simple

Trusting AI with your guest communications can feel like a big step. That’s why you’re always in control of which topics can be automated—and you can start safely in Draft mode.

• Use Draft mode for complex, sensitive, or nuanced inquiries.
  Draft mode is a great place to start: AI suggests a reply, and your team reviews and sends it. This helps you build confidence and fine-tune tone and accuracy before enabling any automation.
• Use Auto-Reply for simple, predictable questions.
  Turn on Auto-Reply only for topics you’re comfortable automating (for example: check-in instructions, Wi‑Fi details, parking, or other common FAQs). You can enable or disable topics anytime.

When to enable Auto-Reply

Only set Auto-Reply on rules where:

• The response is predictable and well-defined
• There’s a clear 1-to-1 match between the guest question and the expected answer
• You’ve tested the rule and are confident in the output

Add a safety layer with delivery settings

For extra reassurance, adjust your delivery settings to let your team answer first. You can also configure AI to provide support only when needed—such as after-hours—so guests still get timely help without fully automating every conversation.

Test before enabling a rule

Use the Test AI feature to simulate guest conversations and verify that rules produce the expected output before enabling Auto-Reply.


 

2. Anti-patterns to avoid

"Kitchen sink" rules

Don't create one rule that covers tone, language, greetings, sign-off, and specific scenarios all at once. This will match every conversation and cause unpredictable behavior.
 

Tone rules as standalone rules

Tone, language, and formatting instructions should live in the General tab, not as separate rules. When they exist as rules, they match nearly every message and interfere with Draft/Auto-Reply logic.
 

Single rule for conditional scenarios

Don't write: "If available, say X. If not available, say Y." in one rule. Split into two rules — one per condition. This gives the AI clearer, unambiguous instructions.
 

Duplicate instructions across rules

If two rules both say "respond in Portuguese," the AI may get confused or produce inconsistent results. Each instruction should appear in exactly one place.
 

Using rules for proactive messages

Rules are reactive (respond to guest messages). Don't try to use rules for scheduled/proactive communications — use Automations instead.
 

Assuming rules apply retroactively

Updating a rule does not change previously generated draft responses. Only new incoming messages will use the updated rule.
 

Expecting complex cross-listing logic

Currently, the AI's context is limited to one reservation and its listing. Rules that require checking other listings' calendars or cross-referencing multiple properties won't work yet.
 

3. Rule examples

We recommend writing rules using explicit labels to help the AI understand the intent of each part of your instruction. Using labels like When:, Condition:, Reply with:, and Except: directly in the rule text gives the AI clear structure to work with — compared to a wall of unstructured text, labeled rules produce more consistent and accurate responses.
You can use these labels directly in the free-text rule editor today. Not all fields are required — use only the ones relevant to your rule.
 

Note on the General tab: The examples below marked as General tab are not rules — they are instructions that go in the Tone of voice custom field or Answer length setting and apply to all messages.

General tab settings (not rules)

These belong in the General tab > Tone of voice custom field. They apply to every AI-generated message and should never be created as standalone rules.
Tone of voice:
 

- Match the guest's tone and formality: if they are kind and slightly informal,
  reflect that warmth without becoming too casual.
- Keep replies warm, professional, concise, and courteous.
- Use "we" instead of "I" throughout the message.
- For positive or neutral messages, thank the guest for reaching out.
- For multiple questions, use concise paragraphs. Do not use bullet points.
- Always format messages as: greeting line, blank line, message body, blank line,
  then sign-off block:
  [Team Name]
  [Phone Number]
  [Website URL]

3.1 Language selection

Pattern: set a default language and define how to detect the guest's preferred language.
 

When:       The guest sends a message in any language.
Reply with: Always greet the guest by name. Detect the language of the guest's
            most recent message based on the primary content (the main question
            or statement), ignoring greetings and sign-offs. Always reply in the
            same language the guest used. If the guest switches languages 
            mid-conversation, follow their latest language choice.
Except:     If the language is mixed or unclear, default to [default language].

3.2 Sign-off

Pattern: define a consistent closing for all outgoing messages, with optional variations based on reservation status.
 

When:       Any outgoing message to a guest.
Reply with: At the end of the message, mention that if they have any other
            questions, they should let us know and we'd be happy to help.
            If the guest hasn't checked in yet, also add that we look forward
            to welcoming them. Always end with:
            "[Closing phrase],
            [Agent Name] / [Team Name]"

3.3 Conditional responses

Pattern: when the answer depends on a condition (e.g., availability, policy), split into one rule per branch. Never combine into a single rule.
Rule A — Condition not met:
 

When:       Guest asks about [service or request].
Condition:  [Service or request] is not available.
Reply with: Apologize and explain it's not possible. Offer an alternative if
            one exists (e.g., a nearby service, a different date, a workaround).
Except:     If the guest is confirming a previously agreed arrangement, simply
            acknowledge it.

Rule B — Condition met:
 

When:       Guest asks about [service or request].
Condition:  [Service or request] is available.
Reply with: Confirm availability and provide the relevant details (pricing,
            options, next steps). Ask any clarifying questions needed to
            proceed.
Except:     If the guest is confirming a previously agreed arrangement, simply
            acknowledge it.

3.4 Guest acknowledgement

Pattern: handle simple "thanks" or "okay" messages without over-replying.
 

When:       Guest sends a simple acknowledgement (e.g., "Thanks", "Okay",
            "Sounds good", "Got it").
Condition:  Our previous message indicated a pending action on our end (such as
            checking with a team, waiting for an update, or processing a
            request).
Reply with: A brief, final acknowledgement. Do not repeat the pending action.
Except:     If the guest is asking a follow-up question disguised as an
            acknowledgement (e.g., "OK, but when will you know?"), treat it as
            a new question instead.

3.5 Charge or fee inquiry

Pattern: when a guest asks about an unexpected charge.
Rule A — Guest asks about the charge:
 

When:       Guest asks about an unexpected charge or extra amount.
Condition:  The charge corresponds to [fee type, e.g., security deposit,
            cleaning fee, tourist tax].
Reply with: Explain what the charge is for. Reassure the guest it is standard
            procedure and reference where it was mentioned (listing description,
            booking confirmation, etc.). If applicable, explain refund terms.

3.6 Troubleshooting

Pattern: for property issues, guide the guest through self-service steps first. Always use separate rules.
Rule A — Initial troubleshooting:
 

When:       Guest reports a problem with [system, e.g., electricity, hot water,
            Wi-Fi, appliance].
Reply with: Apologize for the inconvenience. Ask clarifying questions to
            understand the scope of the issue. Provide step-by-step
            troubleshooting instructions the guest can try themselves.
Except:     If the guest already mentions they tried the troubleshooting steps,
            skip to Rule B.

Rule B — Failed troubleshooting:
 

When:       Guest reports the issue is not resolved after trying the suggested
            steps.
Condition:  A previous message already included the troubleshooting instructions.
Reply with: Apologize for the continued inconvenience and let them know you are
            contacting the property owner or maintenance team immediately.