Research Briefings
What an independent-hotel chatbot should never answer from memory
A guest asks whether they can arrive at 23:30. The guesthouse accepts some late arrivals, but its policy requires written confirmation.
A chatbot retrieves that policy and replies: “Of course. See you tonight.”
The information was available. The condition disappeared.
For an independent hotel or guesthouse, the question is what an assistant needs before answering: maintained property information, a current availability check, access to a particular reservation or a decision from someone responsible.
- Author
-
- Publisher
- Published
- 2026-08-09
- Updated
- 2026-11-07
Match the question to the evidence
Some questions can be answered directly from maintained information. Normal check-in hours, for example, should not require staff intervention when the applicable policy is clear.
Other questions concern something that changes, belongs to a particular guest or has not yet been decided.
| Guest asks | What the answer needs |
|---|---|
| “What time is normal check-in?” | The property’s maintained policy, including relevant exceptions. |
| “Is the Family Room available next Friday, and what does it cost?” | A current result for the dates, party and applicable rate from the designated booking system. |
| “Can you confirm my arrival after reception closes?” | The applicable condition and the recorded outcome for this request. |
| “Has my refund been processed?” | Verified permission to access the correct reservation and current payment information. |
| “I cannot get into the property.” | The property’s current urgent-contact procedure and an appropriate route to help. |
An assistant can explain what happens next while a check or decision remains outstanding. Its wording should make that outstanding step clear.
Check four things behind an answer
A useful answer needs more than a relevant passage. Before relying on it, establish:
- Source:
- which record, policy, system or person supports the statement?
- Freshness:
- is that information current enough for this question, and what could make it outdated?
- Scope:
- does it apply to this room, rate, date, reservation or request?
- Authority:
- can the source establish the claimed outcome, and is this user allowed to receive it?
A policy review date and an availability result have different meanings. A recently updated policy can explain how late arrivals are handled. It cannot establish that a particular arrival has been accepted.
NIST’s generative-AI profile identifies uncertainty, source evidence and the expected validity of information as information-integrity concerns. These four questions are Tamaga’s practical interpretation for hotel answers, rather than a standard scoring formula.
Ask what “checking availability” actually does
A conversational message does not establish that a live check occurred.
Cloudbeds makes this distinction explicit in its documentation. Its Guest Chat automation can display an availability-checking indicator while sending a configured response, without retrieving live inventory. Its separately documented Live Chat booking-engine connection can provide current availability and rates.
Chatbot Automations · Live Chat
HiJiffy’s Checkfront documentation describes another specific arrangement: retrieving room options and rates, then directing the guest to the booking page with details filled in.
HiJiffy–Checkfront integration
These examples give a hotel a useful question to ask its supplier: what happened behind this particular answer?
A configured reply, a current inventory result and a completed reservation each establish something different. Ask to see which operation occurred and how its result reached the guest.
Preserve the condition when wording the reply
Return to the illustrative late-arrival question. The source says:
Arrivals after 20:00 require written confirmation.
If no decision has been recorded for the guest, a useful reply could say:
Arriving at 23:30 requires written confirmation from the property. Your late-arrival arrangement is not confirmed yet. Please contact the team through the property’s stated channel to arrange it.
If the assistant can actually submit a request, it should explain that action accurately. A receipt can confirm that the request was received. Acceptance needs its own evidence.
The same discipline applies to connecting rooms, an extra bed or a particular room allocation. Preserve what is guaranteed, what depends on availability and what still requires agreement.
Keep conversation context separate from permission
Remembering the guest’s dates or unresolved question can prevent repetition. That context should remain available to the people or systems handling the enquiry, within the property’s access and retention rules.
It should not silently become a confirmed arrangement, a permanent guest preference or permission to disclose reservation details.
The distinction becomes more consequential when an assistant can take actions. Looking up availability does not require permission to cancel a booking or issue a refund. OWASP recommends limiting tools and permissions, enforcing authorization in the connected systems and requiring approval for high-impact actions.
Test the answers against your own property
Use the five questions above with your actual policies and a controlled demonstration. Ask the supplier to show the supporting source or result, not just the finished reply.
Include an unresolved request and a situation where the relevant system is unavailable. For reservation-specific questions, use an authorized test record.
Check whether the assistant:
- retains the condition that could change the guest’s decision;
- uses the correct room, dates, rate or reservation;
- states what remains unknown or pending;
- gives a usable next step when it cannot establish the answer.
Repeat the relevant examples after a policy, connection or configuration changes. A successful demonstration establishes what happened in those examples; it does not establish that every future answer will be reliable.
Evidence and scope
This briefing combines published technical guidance, vendor documentation and Tamaga’s interpretation of hotel information and decisions. It is not a comparative product test or evidence of measured guest or commercial outcomes.
The late-arrival exchange is illustrative. Vendor documentation describes particular features and connections; their availability and configuration for an individual property require checking. The briefing does not establish that Tamaga operates a guest-facing chatbot.
References
The original August 2026 bibliography is retained below. Its access dates identify that earlier research. External technical and transparency references were rechecked on 7 November 2026; this does not constitute a test of the products or a new review of the two Tamaga research papers. The transparency guidance remains a separate source from the operational question of who may establish an answer.
-
National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1). July 2024; NIST page updated April 2026. https://doi.org/10.6028/NIST.AI.600-1
-
Booking.com Demand API. Accommodations availability — v3.2 migration guide. Accessed 9 August 2026. https://developers.booking.com/demand/docs/migration-guide/v3.2/accommodations/availability
-
Booking.com Demand API. About the Messaging API. Accessed 9 August 2026. https://developers.booking.com/demand/docs/messaging/about-messaging
-
Cloudbeds. Configure Chatbot Automations. Accessed 9 August 2026. https://myfrontdesk.cloudbeds.com/hc/en-us/articles/8699961484571-Configure-Chatbot-Automations
-
Cloudbeds. Live Chat: Everything You Need to Know. Accessed 9 August 2026. https://myfrontdesk.cloudbeds.com/hc/en-us/articles/19706435203099-Live-Chat-Everything-You-Need-to-Know
-
HiJiffy. Integration between Checkfront and HiJiffy. Accessed 9 August 2026. https://www.hijiffy.com/integrations/checkfront
-
OWASP GenAI Security Project. LLM06:2025 Excessive Agency. Accessed 9 August 2026. https://genai.owasp.org/llmrisk/llm062025-excessive-agency/
-
OWASP GenAI Security Project. LLM02:2025 Sensitive Information Disclosure. Accessed 9 August 2026. https://genai.owasp.org/llmrisk/llm022025-sensitive-information-disclosure/
-
European Commission. Guidelines on transparency obligations for providers and deployers of AI systems. 20 July 2026. Transparency obligations apply from 2 August 2026. https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems
-
Tamaga. Small Independent and Family-Run Hotel Technology in 2026. Research working paper, August 2026.
-
Tamaga. One Hotel, Seven Versions. Public-source forensic edition, August 2026.
Corrections
7 November 2026: Revised to distinguish maintained information, current checks, private reservation access and recorded decisions. Removed the illustrative equation, proposed error metric and fixed demonstration duration; clarified the hypothetical exchange and vendor-evidence limits. The original publication date, citations and bibliography are retained.
Dan Metille
About the author
Dan Metille is the founder of Tamaga and works on travel knowledge infrastructure, semantic architecture and Schema.org-first Drupal for independent tourism.
Research and scope
Research synthesis first published 9 August 2026. Revised 7 November 2026. Sources, illustrative examples and limits are explained in the article.