Review management

Last updated: 2026-08-20

Replying to a review is the easy part. Nobody needs software for that.

The hard part is the question behind it: is what it says true? And if it is, has it been dealt with, or does it happen again next week?

An AI writes the reply and does not publish it

When a review arrives, a draft reply is ready before anyone opens the list. So your manager reads drafts rather than pressing buttons.

Nothing is published without a named approval. That is not a setting you can switch off, and not a statement of intent: in the code there is exactly one way to set an approval, it requires a user, and without that field publishing refuses to run.

The reason is not caution but classification. Since 2 August 2026 the transparency obligations of the EU AI Act apply. A system in which a person approves every public statement is a different thing from one that writes and sends by itself. A path in the code around that approval would not break a rule, it would remove that classification.

Where a platform accepts no reply over an interface, the text is prepared for copying. That it was pasted in is something a person records. The system never infers it from somebody having clicked "copy".

A public reply never contains a guest's name

That is the rule most reply templates break, and they break it while trying to be warm: "Dear Ms Berger, we are sorry that your room on the third floor..." is a public confirmation that this person stayed with you.

Regulators read it strictly. The Austrian Federal Administrative Court upheld a fine for it in 2025; the Thuringian data protection commissioner formally warned two hotels.

So this is not checked by a model but by a fixed test standing in front of publication. A prompt can be argued out of a rule; a check cannot. It also applies when the review could not be tied to a stay at all, which is exactly the case where a name-based blocklist would have nothing to compare against.

Star rating does not decide what opens a ticket

"The hotel was great, only the air conditioning did not work" is a four-star review and a maintenance job.

"Terrible, overpriced, never again" is one star and contains nothing anyone could repair.

The second gets a reply. The first gets a reply and a ticket, because the air conditioning is still broken on Friday if nobody writes it down. The guest who reported it has long since left. This is about the next one.

The cause is looked for in your own operation

Before anything is answered, the system gathers what is known about that stay: the room, the cleaning history of that specific room, the person responsible in that period, open and closed tickets relating to it.

An assessment comes out of that, and two rules about it are not negotiable.

Missing evidence is not evidence of a failure. If no cleaning record is found for that room, it means we know nothing. It does not mean it was not cleaned.

The assessment is explainable. It comes from the facts gathered and not from a model. If your head of housekeeping disagrees with it, the path to it can be walked step by step instead of arguing about a piece of software's opinion.

Whether the fix held

This is the part that exists nowhere else.

An inspection record answers: what did we find, and what did we do? It cannot answer what matters a month later: did it hold?

A supervisor can write "found it, fixed it" in perfect good faith and be wrong. The record itself will never say so. The only one who says so is the next guest.

So the system looks for exactly that shape: the same room, the same category, and a complaint published after the sign-off. All three conditions must hold. When they do, the case is no longer a new complaint but a repeat, and the room gets a standing check in housekeeping.

A review then becomes an inspection in the day plan rather than a mood reading.

Limits

Google needs its own access, per property. Reading Google reviews and replying there requires authorisation for your own business profile, which has to be applied for and which requires the profile to be verified. That is a process between you and Google that we help with but cannot shorten. Anyone promising you an instant Google connection is describing a different process.

Not every platform can be polled. Where there is an interface, we fetch the reviews through it. Where there is none but notifications arrive by email, we read those emails. Where there is neither, somebody enters them. We tell you in advance which of the three applies to which platform, instead of calling all three "integration".

Tripadvisor is read-only, and on the free tier limited to the five most recent reviews. That is their limit, not ours.

Without a link to the stay, a review stays an opinion. If it cannot be established who was in which room and when, there is no cleaning history to look at and no repeat detection. The reply works, the search for a cause does not.

The difference

Most tools in this area are faster at composing a reply.

A faster reply changes nothing about the cause. The water pressure on the third floor, the room next to the lift, the Sunday handover: those are the things that keep appearing in reviews and that stay invisible between individual replies.

Hunzi is also the system that builds the cleaning plan. So a recurring complaint can become a fixed check in that specific room, rather than one more friendly reply.