The four-star with nothing written in it
Four stars. No comment. Nothing at all in the box where the words go.
Steph has looked at it three times this week. There is nothing to answer, nothing to fix, and nobody to argue with, and it still sits near the top of her Google reviews saying something about her salon she cannot read. Somebody has told her you can respond to Google reviews with AI now, which does not obviously help with a review that says nothing.
Steph runs a two-groomer salon in Sydney's Inner West. She is a worked example we follow through this series, and her numbers are the sort you would recognise rather than a real client's books.
Everybody has had one of these, and almost nobody replies, because there is apparently nothing to say. An angry review is at least legible: it tells you what it wants.
Here is the direct answer. You can use AI to respond to Google reviews, and the version that actually works is narrower than most people expect. One saved prompt. Two inputs every single time, the review and a written note about how you talk. And one rule you never break, which is that you read every reply before it goes live. Set up that way, a reply takes about 90 seconds instead of the seven minutes it takes to write one cold at 9pm.
The wordless four-star gets a reply too. Two lines in public, and the actual question asked privately, which is where it always belonged.
Why respond to Google reviews at all, when the reviewer has moved on
Here is the reframe that makes this worth any of your time.
A reply is not written for the person who left the review. They wrote it, they meant it, and they have gone. Hardly anyone comes back weeks later to read what you said.
It is written for the next person. Someone who has never met you, choosing between you and two other salons, scrolling with their thumb on the train. They are not reading the four-star to work out whether you are good at grooming. They are reading your replies to work out how you behave when something goes wrong, because that is the one thing they cannot get from your website, your feed, or your prices.
Which is why the wordless four-star deserves a reply, and why the real trap here is a page of thank-yous. A page with no replies reads as a business that is not watching. A page of replies that all open with "Thank you so much for your kind words" reads as a business that has automated its manners.
How to respond to Google reviews with AI, in one prompt
You do not need five prompts for five kinds of review. You need one good one, and you need to feed it properly.
A chatbot only knows what you put in front of it. That is its context, and it contains nothing about your salon unless you paste it in. So the prompt takes two things: the review, and your voice document, which is the written note about how you talk, what you charge, and what your policies are. Post 9 covers how to write that once, so we will not repeat it here.
Both ChatGPT and Gemini have a free tier that will handle this. Post 8 has what each one costs if you outgrow it.
Two rules do the heavy lifting there. "Ask me for it instead of guessing" is what stops a confident invention appearing on your public profile, and post 5 explains why it invents things at all. "Never state anything the reviewer did not say" is the privacy rule, and it has its own section below.
The five-star with no words
There is nothing to respond to, so do not pretend there is.
Two sentences. Warm, human, and about your salon rather than about them. "Cheers for the five stars, it means a lot to a two-person shop" does the whole job. What you must not write is "thanks for trusting us with Moose", because they never mentioned Moose, and now you have told a stranger who they are.
For the wordless four-star at the top of this article, the public reply is identical. The difference is a second step, and it does not happen in the review thread.
If the name matches exactly one client and you are certain, send this through your own channel. If you are guessing, let it go rather than tell a client you thought they reviewed you.
That last clause matters. You are asking for information, not for a better score.
The five-star with words, and the thank-you trap
This is the one everybody gets wrong, because it looks like the free one.
Somebody has written three sentences about how patient your groomer was with an anxious dog. If your reply is "Thank you so much for your kind words, we appreciate your support", you have answered a specific compliment with wallpaper. Do that 20 times in a row and the whole page reads as though nobody works there.
The move is to answer the specific thing, in one clause, without adding a single fact. They mentioned patience with a nervous dog, so your reply is about how you work with nervous dogs. Not about their dog. Not about their appointment. Yours.
That is also why the prompt bans repeating an opening word. A stranger reading top to bottom notices 30 identical first lines long before they notice good grammar.
The four-star that names a real problem
This is the most valuable review you will get all year, and it usually gets the worst reply.
Someone liked you enough to come back, and cared enough to write down the one thing that annoyed them: the wait at pick-up, the phone nobody answers on Mondays, the price that was different from the quote. That is a consulting report you did not pay for, and it is worth putting through whatever you already do with client feedback.
The reply has two jobs. Name the problem back, because that is what a stranger reads as evidence you were listening. Then say what changes, but only if it genuinely changes. "We've moved pick-up to a 15 minute window so nobody's standing in the doorway" is a good reply. "We'll take that on board" is a shrug in a nicer jumper.
Do not argue the fourth star. Do not ask for the fifth. And do not promise a change you have not made yet, because the reply outlives your intentions.
If the same problem turns up in a second review, it has stopped being feedback and become a fact about your business.
The one-star that is fair
Sometimes they are right. The dog went home with a nick, the booking got lost, somebody was short with them on the phone.
Short reply. Three moves, then stop. Acknowledge the specific thing, say plainly that it is not the standard you hold yourself to, and give exactly one route to a real person, an email or a phone number. Nothing else.
The hardest instruction is this one: do not put the story in the reply. Every extra sentence hands the next reader more of the incident, and you are the one publishing it. A defence reads as a defence however carefully you word it, and this is where handling the conversation well matters more than the wording.
One boundary worth naming. If the review is about an injury, a health outcome, or a behavioural incident, that is not a review-reply problem. It is an incident, and post 20 covers where AI must never go in pet care. Write that reply by hand, after you have dealt with the actual thing.
The one-star that is not fair, or not even your client
Two different problems, and they need different nerve.
Unfair, but a real client. Reply once, briefly, and state your side only in terms that are true of your business rather than of them. "Our cancellation window is 24 hours and it's on the booking page" is a fact about your salon. "You cancelled at 8am after three no-shows" is a fact about them, pulled out of your file, and published by you. Never the second one, however satisfying the draft was.
Not your client at all. Mistaken business, mistaken identity, a review describing a service you have never offered. Reply once in a neutral line that does not accept the premise, along the lines of "We can't find any record matching this, and we'd genuinely like to. Please email us." Then leave it.
Whether Google ever removes it is not in your hands and it is not a plan. The reply is the part you control, and how your Google Business Profile is set up is the part you control after that.
The second prompt, for when tone is everything
The one-star is the review you should not be writing at 10pm on the day it lands. This prompt is for that evening: three temperatures to choose between, when you cannot trust your own.
That last rule is the one to keep. An assumption list at the bottom is how you catch the invented detail before it becomes a public statement.
The three rules with no exceptions
Never publish a reply you have not read. This is the second heading of the never list, and it does not bend for a busy Saturday:
And never let it:
Post 4 has the list in full and the reasoning behind every line. An AI-written reply posted unread onto a complaint is exactly how a small problem becomes a screenshot, and the screenshot is forever.
Never reveal in a public reply anything the reviewer did not. No confirming they were a client. No naming their dog. No discussing what happened to it. No correcting their account with details out of the file. This is the mistake that is hardest to walk back once it is public, and it is a privacy problem as well as a bad look.
Stated carefully, because it is worth getting right. Under the Privacy Act's own definitions an animal is not a person, so a grooming note about a dog is not health information the way a human's medical record is. What it becomes, the moment it sits against a named owner, is information about that owner.
Whether the Act binds your business at all is a turnover question post 4 works through, and exemption is not the same as no exposure: a separate statutory tort for serious invasions of privacy commenced on 10 June 2025 and reaches businesses the Act does not. The regulator's position on AI tools is a recommendation, not a prohibition, so nobody is telling you a chatbot is unlawful. They are telling you to be careful with other people's information.
Never write a fake review, and never ask AI to write one. Not for yourself, not for a mate's business, not as a "placeholder until some real ones arrive". Publishing a fabricated review is misleading or deceptive conduct under section 18 of the Australian Consumer Law, and the ACCC's position is that claims about your services must be accurate, truthful, and based on reasonable grounds, with no allowance for whether you meant to mislead (ACCC). That is the whole paragraph. There is no version of it that is fine.
What this actually reclaims for you
Say Steph has 31 reviews and 24 of them have never been answered. Cold, from scratch, at the end of a 40-dog week, a reply takes her about seven minutes of staring at it. That backlog is nearly three hours in one lump, which is exactly why it is still a backlog.
With the voice document written once and the prompt saved, each reply becomes a paste, a read, and one edit. Call it 90 seconds. The backlog is now a bit over half an hour in a single sitting, and the six or so new reviews a month are nine minutes, done between two dogs.
Those figures are a worked example, not a measurement, and none of this is a promise about your rating or your bookings. What you get is a profile where every review has a human answer under it, written in your words, published by someone who read it first.
The replies stay yours to write. Asking for the review in the first place is the part that can happen with one tap at the end of the appointment, so if that is the half you are missing, you can start your free trial.