A five-star review came in on Tuesday. Someone spent four minutes of their day saying your crew showed up on time and swept the floor before they left, and it's still sitting there unanswered come Friday because you keep meaning to get to it. Meanwhile the one-star from a guy whose mind was made up before you arrived is parked near the top of your profile, and your jaw tightens a little every time you open the app.
You know you should answer both. The warm ones deserve a thank-you that sounds like a person actually read them, and the angry one needs a reply before some stranger reads it cold and assumes the worst. But writing these is one more thing on a list that never ends, and the replies you fire off in a hurry tend to read like a form letter. "Thank you for your feedback, we value your business" fools nobody.
This is exactly the kind of small, repeating writing job where AI earns its keep, as long as you use it right. Not to hit send on autopilot. To hand you a decent first draft in ten seconds so the work becomes editing instead of staring at a blank box. Here's a simple system that keeps the good ones warm and stops the bad ones from making things worse.
Two speeds for two kinds of review
Reviews aren't all the same job, so don't treat them the same way. A happy customer's review is a gift, and all it needs is a warm, specific acknowledgment. An angry one is a small fire in a public place, and how you handle it says more to future customers than the complaint itself does.
Think of it as two speeds. Fast and warm for the five-star ones, where you can knock out a batch in a few minutes. Slow and careful for the one-star ones, where the smartest first move is usually to wait a day before you type a single word.
The five-star pass: warm, specific, done in a minute
Your reply to a happy customer has one job: sound like you remember them. Generic thanks beat silence, but barely. The trick is feeding the AI enough real detail that the draft has something specific to grab onto.
When you paste a review into a tool like ChatGPT or Claude (both free to start), give it:
- The actual review text, pasted in full so the draft can echo what they said
- The customer's first name, if you've got it
- One true detail about the job: what you did, the town, the season, anything real
- A note on your voice, like "friendly and plain, no corporate speak, two sentences"
Then you edit. Add one thing only you would know ("glad the new water heater got you through that cold snap in January") and cut anything that sounds stiff. Ten seconds of drafting, thirty seconds of editing, and it reads like you wrote it, because you basically did.
What you must never let the AI invent
Here's the catch with these tools: they'd rather sound warm than be accurate, so they'll happily make things up to fill the gaps. That's fine for a rough draft and dangerous if you don't read every word before it goes public. Some things you never let slide through:
- Discounts or offers you didn't authorize, like "come back for 10% off" that you now have to honor
- Details about the job the AI was never told, which it will invent to sound specific
- The customer's name, if you're not certain of it or the review is anonymous
- Awards, rankings, or "best in town" claims that aren't true
- Anything private, especially for a clinic or anyone under HIPAA, where even confirming someone was a patient can be a problem
Read the draft as if a competitor is going to screenshot it. If a sentence would embarrass you or box you in, cut it before it ever posts.
The one-star reply: calm, short, and never a fight
An angry review is different, and the biggest mistake is answering it while you're still angry. Remember who the reply is actually for. It isn't the upset customer, who's probably not coming back. It's the next person reading, deciding whether you're the kind of business that stays cool under pressure.
- Wait. Draft it, then sleep on it. Nothing good gets typed at 9 PM with your blood up.
- Keep it short. Three sentences. A wall of text reads as defensive.
- Acknowledge, don't argue. "I'm sorry your experience fell short" beats relitigating what happened in public.
- Take it offline. Give a name and a direct number so the conversation moves off the platform.
- Guard privacy. Never share job details, health information, or anything that confirms sensitive facts about the person.
Feed the AI the situation in a sentence or two and ask for "a short, calm, professional reply, no admission of fault, that invites them to call us." When you're steamed, a machine that isn't steamed is genuinely useful. It writes the level-headed version you can't quite reach yet, and then you make it sound like you.
When to close the app and just call
Some reviews don't need a better reply, they need a phone call. If you actually dropped the ball, or a regular you value is upset, or there's a real misunderstanding worth clearing up, pick up the phone before you post anything. A two-minute call fixes more than a perfect paragraph, and people who feel heard will sometimes update the review on their own.
If a review is flat-out fake, from someone who was never a customer or a mixed-up business name, don't waste a calm reply on it. Flag it with Google or the platform for removal, and if it lingers, post one short factual note ("we have no record of this visit and would like to help, please call us") so future readers see your side.
Worth doing this week
None of this takes long once you've done it a couple of times. Start small:
- Reply to your three most recent five-star reviews, each with an edited draft, not a copy-paste.
- Save a reusable prompt in your notes app so you're not rewriting instructions every time.
- Draft a reply to your worst review today, sleep on it, and send the calmer version tomorrow.
- Block fifteen minutes every Friday to clear the week's reviews in one pass.
Setting up the prompts and the weekly habit can feel like one more thing you'll never get to, and that's the sort of small system we help local owners put in place at Massachusetts AI Agency. Either way, the goal's the same: replies that sound like a person, sent while they still matter.