The hygienist is running late, the patient is fishing for keys, and the review card at the front desk is face-down under a stack of insurance forms. Across the street, the other practice has 400 Google reviews. This office has 14. The work isn't worse. Nobody asks at the right time, and the two replies they did post start with "Dear Valued Patient," which is how you know nobody in this zip code talks like that.
Reviews are how a stranger picks a dentist, an HVAC company, or a store with three locations on the same highway. They are Effectiveness: new customers find you. They are Quality: the person who just paid you should not get a ghost after the job. If your "AI review strategy" is fabricating five-stars or blasting the same paragraph at every two-star, you are burning both.
I will say this once, in plain English: we never write fake Google reviews. We never pay for them. We never put words in a customer's mouth and hit submit from the shop. If a vendor demo shows you a pile of auto-generated five-stars, walk out. That's not a workflow. That's a risk. What AI is actually good for: timing the ask, drafting a reply in your voice, routing the ugly ones to a human before they go live, and making sure location two doesn't sound like a different company than location one.
Why shops stall after one ask
Dental: the hygienist is supposed to mention reviews. Some days they do. Most days the chair is running late. HVAC: the tech is standing in a driveway with a dirty shirt and a card reader. Asking for a Google review while the customer is still looking at the invoice feels greedy. So they don't. Two days later nobody asks.
Multi-location retail: corporate (or you, wearing that hat) emailed a script in January. Store three's manager read it. Store one never did. Store two posted one reply that starts with "Dear Valued Customer" and you can smell the template from the parking lot. Law firms are in a narrower box. Some matters should never be on Google. A model does not get to decide that.
The pattern is the same. You asked once. You built no loop. The loop is the product. Filter it or don't bother. Effectiveness: does this create more honest public proof that a stranger will trust? If your shop is invisible on Maps because you have 14 reviews and the next dentist has 400, this is a money problem, not a vanity problem. Efficiency: does this return hours to the front desk? If the office manager is hunting through last week's appointments to remember who to text, that's unpaid archaeology. Quality: does the customer feel seen — including the unhappy one? A canned "We appreciate your feedback!" on a complaint about a no-show is how you look like you don't care.
If a tool fails all three, you don't buy it. A review-generation app that writes fiction fails Quality on purpose. Kill it.
The ask: timing is the whole trick
People leave reviews when the work is still warm and the outcome is clear. Not during the mess. Not six weeks later.
Dental: choose a neutral invitation rule, such as a fixed interval after every completed visit where contact is permitted. Apply it whether the visit was easy, delayed, or difficult. Do not ask from the chair or pressure someone to post. The office manager approves and sends the invitation. Consent and opt-out rules still apply; a treatment record does not belong in the drafting prompt.
HVAC: use the same completed-job rule across the queue. A tech can flag “smooth / delayed / callback likely” to route service recovery, but that flag never decides who receives a review invitation. Work on the complaint separately. Never make the public link conditional on a happy private survey.
Hypothetical (not a client, not a promise): if a crew runs four completed jobs a day and you only remember to ask on Fridays, you left a week of proof on the table. I'm not going to invent your conversion rate. You already know some people will tap a link and some won't. The failure is never sending the link. Count your completed jobs last week versus asks that actually left the building.
Multi-location retail: choose a neutral rule for real purchases, such as all eligible completed orders or the same documented sample across locations. Include dissatisfied customers under that rule. Use the same invitation language with the correct Maps link. A complaint changes the service-recovery route, never the invitation eligibility. A named manager approves and sends each ask.
Google prohibits incentives for reviews, discouraging negative reviews, and selectively asking for positive reviews. Do not offer a discount or gift. The FTC rule also prohibits fake reviews and incentives conditioned on review sentiment. The installation playbook explains the distinction and links to the official rules. For US business texts using local ten-digit numbers, confirm A2P 10DLC registration with your provider; other number types have separate requirements. Registration does not supply consent. Keep consent evidence and working STOP/HELP handling before a human sends.
The reply: draft, then a human hits send
This is where shops sound fake. The model will cheerfully write: "We're so sorry for your experience! We strive for excellence and would love the opportunity to make this right. Please call us at…" Every shop in your zip code has that paragraph. Customers can smell it.
Your replies need three things the model does not have until you give them. How you actually talk: short, named, specific. For a dental or medical public reply: "Thanks for the feedback. Please call our office manager at [number] so we can look into this." Never confirm a patient relationship or mention treatment, even if the reviewer did. What you never say: no medical claims from a dental office, no "we'll win your case" from a firm, no "we'll have a tech there in 20 minutes" from HVAC if you cannot. The facts of this review: which location, which job, what you can admit, what has to go to the owner.
Workflow: new review lands, tagged by location and star rating. Helper drafts a reply using your voice file and the never-list. One human edits and sends. Front desk for the easy ones. Owner or office manager for anything with a complaint, a clinical detail, a legal detail, or a threat. Save the edit. "We cut the word delighted. We named the tech. We offered a call, not a coupon." That lessons file is how next week's draft stops sounding like a hotel chain.
One-star and two-star reviews are Quality work. Do not hide. Do not argue medicine or the law in public. Acknowledge, take it offline, own the part that's yours. For dental or medical feedback, use a generic invitation to contact the office manager. Do not confirm a visit, wait, patient relationship, or treatment in public. A law-firm review that describes a matter is a human-only job, full stop.
Paste this. Fill the brackets. Never let it hit publish on its own:
Draft a Google review reply for our shop. Do not send. Do not invent facts.
Shop: [dental / HVAC / law / retail], location: [name]
Voice: short, named, specific. Never say: [never-list]
Review summary (de-identified; no patient, customer, or matter details):
Star rating:
What we know is true:
Draft:
- For dental/medical: use a generic acknowledgment and office-manager contact. Never confirm a patient relationship or treatment. For legal: flag the draft for the responsible attorney; disclose no matter details.
- For other businesses with 4–5 stars: thank them without inventing details, no "delighted."
- If 1–3 stars: acknowledge, take it offline, own only what we know. No medical claims, no legal conclusions, no promised arrival window we can't keep.
Flag anything you're guessing. A human will edit and post.
Multi-location: one voice, the right pin
If you have three retail stores or two HVAC shops, the internet thinks you are three companies. That's fine. The replies should still sound like one owner. Every review is mapped to the correct listing before anyone drafts. Shared never-list. Shared voice. Location name in the reply when it helps ("our Oak Street desk"), not a different personality per store. One person still owns the queue. If each manager replies "when they have a minute," you will get four tones and two unanswered weekends.
Don't copy a five-star reply from location A onto location B. Specific is what makes it human. "The red couch in the window" is Quality. "Great customer service!" is noise.
We will not generate customer reviews. We will not auto-post replies. Human sends. Always. A bad public sentence is expensive in dental, HVAC, law, and any retail brand you're trying to keep clean. We will not let the model invent a discount, a diagnosis, a legal position, or a promise of a same-day tech. We will not buy a tool whose pitch is "set it and forget it." Forgetting is how you got here.
If we were installing this as the first workflow in a shop, it would look like a month, not a campaign week. Week 1: write the rules. Neutral eligibility and timing. Consent and opt-outs. Who sends. What we never say. Week 2: build the ask. Job-complete trigger. Tag. Draft text. Human send. Correct Maps link. Week 3: build the reply queue. Week 4: count. How many asks actually left the building. How many reviews arrived (you will not control that number, and I will not pretend otherwise). How many replies posted inside 48 hours. How many drafts the office manager had to rewrite from scratch — if it's most of them, the voice file is wrong, not the staff.
You don't need a new review platform to start. Most of you already have Google, a texting tool, and a person at the desk. The missing piece is the definition: who gets asked, when, who drafts, who sends, what we never say.
For the practical steps, use the review requests and human-sent replies installation playbook.
FAQ
Can AI write Google reviews for my local business? No. We never write fake reviews, never pay for them, never submit from the shop. AI times the ask and drafts the reply. A human sends both.
When should we ask for a review? Use a neutral, documented rule after completed visits or jobs where contact is permitted. Apply it regardless of satisfaction. Handle complaints separately; do not exclude someone because they may leave a negative review. A named person approves and sends the ask.
How do we stop sounding like a template? Give the helper your voice, a never-list, and the facts of this review. Name people. Cut "delighted." Save every edit. If the office manager rewrites most drafts from scratch, the voice file is wrong.
If your team needs help using this process, we can consider it during a $1,500 AI Workflow Sprint. That is hands-on team practice with reusable workflows and instructions; a connected review system or larger rollout is scoped separately as Implementation. A free introductory conversation helps establish the right approach. It does not install the workflow or audit your listing.
Let’s talk about your business.
Put it to work
Install the steps, prompts, human approval rules, and scoreboard in Get Found in ChatGPT.