Back GuestGuy

Data & AI Usage

This page explains how GuestGuy uses artificial intelligence to power the guest concierge, and how data flows through the system.

How the AI Works

When a guest asks a question on a property's stay page, GuestGuy assembles a system prompt and sends it to Anthropic's Claude AI model along with the guest's message. That system prompt includes:

• Property name and host details
• All property guide content the host has entered — house rules, check-out instructions, parking information, trash instructions, appliance instructions, additional notes, and knowledge base content
• The WiFi network name. The WiFi password is not included in the system prompt — it is delivered separately when the guest taps the WiFi button in the chat interface.
• Check-in and check-out dates and the property's timezone
• The conversation history from the current session
• The guest's detected language
• Nearby places data, when the host has enabled local recommendations (see below)
• Aggregated FAQ insights derived from prior guest conversations at that property, when available and recent
• A rolling summary of earlier turns in the same session, when the conversation is long enough to require one
• Photo descriptions, for properties on a paid plan where the host has uploaded photos

The AI is instructed to act as a concierge for the specific property. Each AI response is logged to the host's activity history so hosts can review what guests are asking and identify gaps in their property guide.

AI also runs in two background roles beyond the per-turn guest response:

• Session summaries — after a guest session ends, the guest's messages are sent to Anthropic to extract any opinions they shared about the property or nearby spots. The result is stored as a host-facing summary.
• FAQ insights — periodically, a rolling 30-day window of conversation transcripts at each property is sent to Anthropic to identify common question patterns and gaps in the property guide. The resulting insights are used to improve the AI's future answers at that property.

How We Use Anthropic's AI

Each guest message and the assembled property context are sent to Anthropic's API to generate a response. This data flow is governed by Anthropic's Privacy Policy. Anthropic states that API inputs and outputs are not used to train their models by default.

Anthropic is also called for the two background roles described above. In both cases, data is sent as guest question text and AI response text drawn from conversation logs. Credential values — WiFi passwords, property access codes, door codes — are never stored in conversation logs and never reach Anthropic in any call. Turns where credentials were delivered are stored with placeholder text only (for example, "[WiFi credentials provided]"), and that placeholder is what appears in any data sent for secondary processing. Questions about credentials — such as "what's the wifi password?" — may appear in secondary calls as part of the question text, but the actual credential values do not.

What Data We Store

The following is stored for each conversation turn:

• Guest question — truncated to 500 characters
• AI response — truncated to 1,500 characters; credential turns are stored as placeholder text, never the actual values
• Session ID and turn number
• Question category — for example, WiFi, parking, check-in
• Escalation flag — whether the AI directed the guest to contact the host
• Device type — mobile, desktop, or tablet
• Detected language
• Interaction method — whether the guest typed a message or tapped a suggestion
• Token counts and response latency
• Fallback flag — whether the AI gave a confident answer or used a fallback response
• Host-facing insight — an optional short AI-generated note stored on the first turn of sessions that escalated or ran long
• Host resolution state — whether the host marked the conversation resolved, and how
• Guest feedback — optional thumbs-up or thumbs-down rating, if submitted

Other data stored by GuestGuy:

• Property information — as entered by the host
• Guest messages via Contact Host — stored only when a guest uses the Contact Host feature to message the host directly
• Property FAQ insights — aggregated question patterns and gaps derived from 30-day rolling conversation transcripts at each property
• Session summaries — per-session records including session duration, escalation state, and any AI-extracted opinion quotes from the guest
• Rolling conversation context — temporary summaries used to maintain continuity in long sessions; cleaned up when sessions become inactive
• Guest session tokens — a SHA-256 hash of the guest's browser token is stored server-side so returning guests can skip the PIN entry on repeat visits. The raw token lives only in the guest's browser and expires after 14 days. Hosts can invalidate all active sessions immediately by resetting their property's guest access PIN.
• Notification preferences — host account settings

Conversation logs are retained for up to 3 years, after which they are automatically deleted.

What Data We Do Not Store

Guest names or identities are not stored unless a guest volunteers their name in a message. Data in conversation logs is not linked to a specific individual.

WiFi passwords and property access codes are never stored in conversation logs and are not included in the system prompt sent to Anthropic. They are delivered to guests through a separate credential mechanism within the chat interface.

Device type is stored in conversation logs (listed above). We want to be clear about this because earlier versions of this page stated otherwise.

IP addresses are stored temporarily for rate limiting and security enforcement. Messages sent to the chat are rate-limited by IP address to protect against abuse; rate-limit records are pruned approximately every two hours. IP addresses associated with security events — such as repeated failed PIN attempts on a protected property — are logged separately; retention for those records is under review. In all cases, IP records are not linked to conversation history, not used to identify guests across properties, and not shared with hosts.

Local Recommendations and Google Maps

When a host enables local recommendations for their property, GuestGuy sends the property's street address to the Google Maps API to compute geographic coordinates and retrieve nearby places. The resulting coordinates are stored permanently with the property record. Nearby place information — including business names, street addresses, ratings, and maps links — is cached at the property level and included in the AI's system prompt when guests ask about local spots.

Token Usage & Billing

Each plan includes a daily message limit per property. If that limit is reached, guests see a polite message and are unable to send new questions until the limit resets at midnight UTC. Your usage dashboard shows token consumption and message counts per property.

AI Limitations

The AI concierge draws on what the host has entered in their property guide, cached Google Maps data when local recommendations are enabled, and aggregated FAQ insights derived from prior guest conversations at that property. It does not have access to real-time information or anything outside the assembled prompt. For sensitive or time-critical matters, guests are encouraged to contact the host directly.

Your Rights and Contact

Hosts can request deletion of their account and all associated property and conversation data by contacting support@guestguy.com. Deleting a host account permanently removes all associated property records and conversation logs.

Because guest conversations are not linked to individual identities, we cannot locate or delete conversations by guest identity. If a specific guest wants conversation records removed, the host at that property can facilitate this by contacting support on their behalf.

For any other questions about how your data is handled, reach us at support@guestguy.com.

Questions

If you have questions about how your data or your guests' data is handled, contact us at support@guestguy.com.