← All articles

AI for replying to reviews: what it does well, and where a human is required

Replying to reviews is a job that's never finished and never a priority. So most owners arrive at the same question: can a machine write these? The answer is yes — but not all of them, and not without you. It's worth knowing exactly where the line falls, because errors on either side cost you: silence on one, a robotic profile on the other.

Where AI genuinely helps

The blank screen. The biggest cost of replying isn't the writing but everything before it — the moment you don't know how to start. A draft already sitting on screen turns a ten-minute task into a thirty-second one.

Languages you don't speak. A review in German or Italian without AI means a translator, or a reply in English, or silence. For businesses living off tourists that's the clearest benefit of all.

Consistency. Ten different people in a business write in ten different tones. A model given a defined tone holds the same voice on every review, which keeps the profile coherent.

Routine situations. A five-star with no text, a short compliment, the standard complaint about waiting — ninety percent of reviews fall into a handful of types answered by the same logic. That's precisely why we could write a library of templates at all; a machine does the same, only faster and fitted to the actual text.

Where AI gets it wrong

It invents details. A model that doesn't know your menu will write "we're delighted you enjoyed our famous dessert" for a place that serves no desserts. Every concrete detail in a draft has to pass your check.

It promises on your behalf. "We'll comp your dinner" is a sentence only someone authorised to honour it may write.

It doesn't recognise the serious ones. A review claiming food poisoning, discrimination, an insult or a threat isn't a "negative review", it's an incident. Those replies are written by a person, carefully, with legal advice where needed — never by a machine, and never in a hurry.

It sounds the same. If drafts get published without a single edit, after twenty reviews the profile reads as an automat: same constructions, same rhythm. Guests notice even when they can't say why.

When a human must write

Four situations where AI may help prepare, but only a person publishes:

  1. Health and safety — anything touching food, hygiene or injury.
  2. Claims about employees — accusations of rudeness, discrimination, dishonesty.
  3. Reviews you don't recognise — a suspected fake review calls for checking bookings and receipts, not a well-turned sentence.
  4. A regular or a known name — where a relationship exists, a template shows, and it stings.

How to do it properly

A practical setup that works in small businesses:

Honestly, about tools

Rewyer is exactly this kind of tool: it watches your profile, drafts a reply to every review in your tone and in the guest's language, and you approve with one tap. Which makes it only fair to state the other side — no tool, ours included, should publish unsupervised on reviews touching health, employees or suspicious claims. The value of automation is that ninety percent of routine replies stop eating your time, so that time is left for the ten percent where only your own words will do.

The test is simple: if a guest reading your profile can't tell whether a human or a machine wrote the reply, you're doing it well. If they can, fix the tone — not the volume.