AI Voice Agents in Hospitality: Who's Liable When the Bot Promises Something?

The Rate on the Phone Isn't Always the Rate at Check-In

A guest calls your hospitality business at 2am. An AI voice agent answers, quotes a room rate, confirms a booking, and sends a confirmation. The guest arrives on Friday, and the rate at check-in doesn't match what was quoted on the phone. Who is responsible for that discrepancy?

The answer, under New Zealand law, is straightforward: your business is. When an AI voice agent communicates a price, a booking detail, or a policy on behalf of your hospitality business, that communication is treated as a representation made by your business, not by the software vendor, and not by the AI itself. The Fair Trading Act 1986 prohibits misleading or deceptive conduct and false representations about price, and it does not carve out an exception for who, or what, made the representation. If a human front desk staffer misquotes a rate, the business wears it. The same principle applies when the voice on the phone is synthetic.

This isn't a hypothetical risk confined to some future AI regulation. It's the current legal position, applied to systems that are already answering calls for hospitality businesses today.

Where the Exposure Actually Sits

For a hotel, motel, or restaurant running an AI voice agent, the exposure isn't abstract. It concentrates in a small number of specific moments:

Rate quotes. A caller asks what a room costs for a given date range. If the AI quotes a rate that differs from what's actually available or what gets charged, that's a discrepancy your business has to resolve, usually in the guest's favour.

Booking confirmations. The AI confirms a reservation, a room type, or an arrival time. If that confirmation doesn't match what's actually in your property management system, the guest has a legitimate complaint, and a receipt trail is the only way to establish what was actually said.

Cancellation and refund terms. A caller asks about your cancellation policy. If the AI states the terms incorrectly, whether by outdated information or a genuine error, your business is on the hook for whatever it said, not for what your actual policy is.

Special requests. A guest asks for a specific room feature, an accessibility accommodation, or an early check-in, and the AI confirms it. If that confirmation never makes it into your PMS, the guest arrives expecting something that was promised and never delivered.

None of these require a chatbot to invent a fictional policy, the way Air Canada's chatbot invented a bereavement discount that didn't exist and was later held liable for by Canada's Civil Resolution Tribunal, with the “the chatbot is a separate legal entity” defence rejected outright. Most hospitality AI liability doesn't come from a dramatic invented policy. It comes from small, ordinary mismatches between what was said on a call and what's actually true in your systems, multiplied across every call the AI handles.

Disclosure Is Not Optional

There's a second layer to this that hospitality operators specifically need to get right: callers have to be told they're speaking with an AI. Under NZ Privacy Act 2020 obligations around recording and transcription, a caller must be informed at the start of the call that they are speaking with an AI system and that the call is being recorded and transcribed. This isn't a nice-to-have disclosure. It's a compliance requirement, and it needs to be built into the call flow from the first second, not added as an afterthought.

Getting the disclosure right doesn't reduce your liability for what the AI says during the call. It's a separate obligation that sits alongside the representation risk, not a substitute for managing it.

The Receipted AI™ Answer for Hospitality

Receipted AI™ is designed to close this gap by treating every AI action the way your business already treats a financial transaction: with a receipt. When an AI handles a hospitality call under this model, every material action it takes produces an immutable, timestamped record: the rate quoted, the booking reference created in your PMS, the confirmation sent and to whom, the cancellation terms stated, and the special request logged.

In practice, that looks like this: a guest calls after hours and books a room. The AI quotes a rate, confirms the room type, creates the booking in your PMS, and sends a confirmation. Every one of those steps produces a receipt: the exact rate quoted, the PMS booking reference, the confirmation timestamp. If the guest arrives and disputes the rate, the receipt resolves it in under a minute, not with a guess about what the AI probably said, but with the actual record of what it said.

This is a different thing from a call transcript or a quality score, which is what “auditable AI” usually means across the industry. A transcript tells you what was said in the conversation. A receipt tells you what the conversation caused: the specific booking, the specific rate, the specific confirmation, tied to the specific system record it produced.