You can use AI to reply to Google reviews, and it does save time, but only if you give it rules and read every reply before you post it. A bare request like “write a friendly reply” gets you the same cheerful paragraph every time, and sometimes a fact the customer never said. Below is the prompt we use, what happened when we tested it on our own Google reviews, and the checks a person still has to make.
Should You Use AI to Reply to Google Reviews?
Yes, for the first draft. No, for the final say. Replies matter because other customers read them: Google posts your reply publicly under the review, as your business, and tells the reviewer you answered. That’s exactly why a person should approve each one.
AI is good at the slow part: finding something specific to say, keeping it short, and staying calm when a review stings. It’s bad at knowing what really happened. It doesn’t know the job ran late because a part was back-ordered, or that this customer has been with you for 10 years. You do.
One line not to cross: use AI to write replies, never reviews. The FTC’s 2024 rule on fake reviews names “AI-generated fake reviews” outright, along with buying reviews and posting reviews from insiders. Google’s own policies also forbid offering discounts or freebies in exchange for reviews.
What Did We Find When We Tested It on Our Own Reviews?
Our own replies were less personal than we thought, and a basic prompt made the problem worse. We have eight Google reviews, all five stars, all answered. Before writing this post, we looked at our replies the way a stranger would:
- All eight start with “Thank you”. Read together, they sound like one form letter.
- Two include a pitch for our services. Google’s tips say to be “conversational, not promotional”, since the reviewer is already a customer.
Then we ran the same eight reviews through two prompts:
| Basic prompt (“write a friendly reply, mention something specific”) | The prompt below | |
|---|---|---|
| Opens with “Thank you so much” | 8 of 8 | 0 of 8 |
| Uses the word “thrilled” | 7 of 8 | 0 of 8 |
| Adds something the review never said | 2 (a ranking gain, and more reviews) | 1 claim on the first try, cut by a person |
| Pitches our services | 0 | 0 |
| Average length | About 33 words | About 31 words |
The basic prompt’s worst mistake was small but real. One review said the customer was impressed with “the results we’ve seen”, and the draft thanked them for their website “climbing in the search results”. Nobody said that. Posted under your name, an invented detail like that reads as a claim you made.
The Prompt We Use
Copy this, fill in the brackets once, and save it. We tested it in Claude. It’s plain text with no special features, so you can use it in whichever AI assistant you already have.
You're helping me reply to a Google review for [BUSINESS NAME],
a [TYPE OF BUSINESS] in [TOWN].
About us, in a sentence or two: [WHAT YOU DO AND HOW YOU TALK,
e.g. "family-run, plain-spoken, we answer our own phones"].
Rules for the reply:
1. 20 to 60 words. Shorter is fine.
2. Don't start with "Thank you". Thank them somewhere, but open
with their name or the thing they mentioned.
3. Mention one specific thing from the review, in new words.
Don't repeat the review back to them.
4. Use only facts that are in the review or in "About us".
Never add results, numbers or details they didn't mention.
5. No offers, discounts, links or sales pitch.
6. Never confirm or reveal anything about their visit, order or
account that isn't already in their review.
7. If the review is 3 stars or lower: say you're sorry about the
specific problem, don't argue, and invite them to contact
[NAME] at [PHONE OR EMAIL]. Don't admit fault for anything the
review doesn't make clear. Mark anything I need to check
with [CHECK].
8. Sign off as [YOUR NAME], [YOUR ROLE].
Give me the reply, then one shorter version.
The review ([NUMBER] stars):
"[PASTE THE REVIEW]"
Rule 4 is the one that matters most. Rules 2 and 3 are what stop your replies from sounding like a form letter.
What Does a Good AI Reply Look Like?
Short, specific, and signed by a real person. Here’s one from our test, replying to a review that said everything a client needed was “all in one place”:
“All in one place” is exactly what we aim for, so that made our day. Thank you. Rohit, Symphony Core
(We’ve left out the reviewer’s name here. The real reply opens with it.)
And here’s the prompt on a negative review. This one is made up for the example, for a plumbing company:
Review (2 stars): “Tech showed up two hours late and didn’t call ahead. Work was fine but I took the morning off for nothing.”
Draft: Mike, you’re right to be frustrated. Showing up two hours late without a call cost you a morning, and that’s on us. The work matters, but so does your time. I’d like to make this right, so please call me at [PHONE] or email [EMAIL]. [CHECK: confirm the visit date and what happened before posting.] Dana, Owner
The [CHECK] marker is the point. Before that reply goes up, someone confirms the tech really was late. If the customer actually got the time wrong, you’d change “that’s on us” before posting. For tougher cases, such as a 1-star rating with no comment, we’ve written a separate guide to replying to a 1-star review with no comment.
What Should a Person Check Before Posting?
Four things, and they take about a minute per review.
- Is every fact true? Read each sentence and ask, “Did the customer say this, or do we know it for sure?” This is where AI goes wrong most often.
- Does it sound like you? If you’d never say “made our day”, change it. The reply should sound like the person who signs it.
- Is anything private? Don’t confirm that someone was a patient, a client or a customer if they didn’t say so. Google tells businesses never to share a reviewer’s private information.
- Is it a new reply? Compare it with your last few. If they all open or close the same way, rewrite the opening.
Then post it from your Business Profile. Google checks replies against its content policies first. That usually takes a few minutes but can occasionally take much longer, and you can edit or delete a reply later if you need to.
How Long Does This Take Each Week?
For most local businesses, 10 minutes or less. A business getting five reviews a week spends a few seconds pasting each review, a minute reading the draft, and a minute fixing it. That’s the real saving: you’re editing a draft instead of facing a blank box. It’s also what makes replying to every review realistic, not just the ones that stand out.
When reviews arrive faster than that, the next step is a system that flags each new review the day it lands and asks for a reply. The drafting can be AI. The approval should still be a person.
The Part Nobody Tells You
AI makes it easy to reply to every review. It doesn’t make every reply good. We found that out by testing our own replies, which had drifted into the same “Thank you so much” opening without anyone noticing. The prompt fixed the pattern, but a person still caught the one sentence that didn’t belong.
If you’d like help setting up review requests and replies so they run every week without you remembering, book a demo and we’ll show you how it works for a business like yours.


