Using AI to draft review replies safely

Why review replies deserve care

For a home service company, the review profile is often the first impression a homeowner gets. Pew Research Center found that 82% of US adults at least sometimes read online ratings or reviews before buying something for the first time, and 40% always or almost always do; among adults aged 18 to 29, 53% always or almost always read them^[1]. Prospective customers read your replies too, especially to negative reviews, to judge how you treat people when something goes wrong.

Replying to every review takes time, which is why AI drafting is appealing. It can produce a reasonable first draft in seconds, vary the wording so replies do not all read the same, and summarize a long complaint so the owner sees the core issue quickly. Used carelessly, it produces generic replies, reveals details it should not and occasionally says something that makes a bad situation worse.

The rules every draft must follow

1. A person reads, edits and posts every reply. No automatic posting.
2. Reply within a few days, and within one business day for negative reviews.
3. Use only what the reviewer shared publicly. Never add their address, the price they paid, the date of the visit or anything from your records.
4. Name the job type in general terms, such as AC repair or gutter cleaning, which helps readers and keeps it relevant.
5. Thank the reviewer specifically, not with a stock phrase.
6. On negative reviews, acknowledge the experience, apologize for how it felt without arguing about facts in public, and give a named contact for follow-up.
7. Never offer a discount, refund or gift in exchange for changing or removing a review.
8. Never threaten legal action or hint at it in a reply.
9. Keep it short: two to four sentences for most reviews.

The legal backdrop is real. The FTC's final rule, announced August 14, 2024, bans the sale or purchase of fake reviews, including AI-generated fake reviews, bars businesses from using unfounded legal threats to suppress negative reviews and allows the FTC to seek civil penalties against knowing violators^[2]. FTC guidance also warns that asking customers to change or delete negative reviews could mislead readers, and that asking only customers you expect to be happy to leave reviews can be misleading if it skews the results^[3].

A prompt template for drafts

Paste the review text and nothing else from your records. If you want the reply to mention that the issue has been resolved, say so only after it has been and only in general terms, such as we are glad our team could return and sort it out.

Templates by review type

1. 01

   Five-star, specific

   Thanks so much, Priya. We are glad the new thermostat is making the upstairs comfortable, and we will pass your kind words on to Luis. We appreciate you choosing us.

2. 02

   Positive but brief

   Thank you, Tom. We are glad the gutter cleaning went smoothly and appreciate you taking the time to leave a review.

3. 03

   Mixed

   Thanks for the honest feedback, Sam. We are glad the repair worked, and we are sorry the arrival time ran later than promised. We are working on better updates when technicians are delayed.

4. 04

   Negative

   We are sorry to hear about your experience, Jordan. This is not the standard we aim for. Please contact our service manager, Erin, at [phone] so we can understand what happened and put it right.

5. 05

   Possibly not a customer

   We take every review seriously but cannot find a record that matches this description. Please contact Erin at [phone] so we can look into it.

6. 06

   Mentions a safety concern

   Thank you for raising this, Alex. Safety comes first, and our service manager will contact you today. If you notice a gas smell or a burning smell, please leave the home and call 911 first.

Use the templates as patterns, not copy. AI drafting is most useful when it adapts the pattern to the specific review, which is what keeps a page of replies from sounding automated.

Personalization without oversharing

Good replies sound like they came from someone who knows the job. That does not require private details. Reflect what the reviewer said, mention the technician by first name if the reviewer did, and refer to the type of work. Avoid confirming anything the reviewer did not post, such as which neighborhood they live in or what equipment they bought. For negative reviews, the details belong in the private follow-up, not in public.

Be honest about authorship in spirit. In Pew's 2025 survey, 76% of US adults said it was extremely or very important to be able to tell whether text was made by AI or by people^[4]. A reviewed, edited reply posted by your team is your company's reply. What crosses the line is presenting AI text as a personal story or experience that no one at the company actually had.

A weekly workflow

1. 01

   Collect new reviews daily

   Use your platform's alerts so nothing waits more than a day.

2. 02

   Triage

   Flag negative reviews, safety mentions and possible fake or competitor reviews for the owner or service manager.

3. 03

   Draft with AI

   Run the prompt for each review and save drafts in one place.

4. 04

   Edit and post

   A person checks facts, privacy and tone, then posts.

5. 05

   Follow up offline

   For negative reviews, the named contact calls within one business day and records the outcome.

6. 06

   Review monthly

   Look for repeated complaints in reviews and fix the cause in operations, not just in replies.

Mistakes that make replies backfire

• Identical replies under every review. Readers notice, and it signals that nobody actually read what customers wrote.
• Keyword stuffing, such as cramming best HVAC repair in [city] into every reply. It reads badly and helps nobody.
• Correcting the customer in public with details from the job record, even when your version is accurate.
• Replying while angry. Draft, wait an hour, then edit and post.
• Asking staff, friends or family to post reviews, or posting reviews written by AI. This is exactly what the FTC rule targets.
• Letting AI reply to a review that mentions an injury, property damage or a lawyer. Those go to the owner and, where needed, your insurer.

When in doubt, shorter is safer. A two-sentence reply that thanks the customer and names a contact is rarely wrong; a long one usually contains something you would rather not have said in public.

Example: a week of replies

Example: a fictional garage door company receives nine reviews in a week: seven positive, one mixed about a late arrival and one negative about a spring replacement that failed after a month. The office manager runs each through the prompt, edits two drafts that were too formal and posts all nine within two days. For the negative review, the reply names the service manager, who calls the customer, sends a technician under warranty and records the cause, a defective spring batch. The company does not ask the customer to update the review. A week later the customer adds an update on their own.

Reviews also feed your local search presence; see AI content for local SEO and our guide to handling complaints for the offline side.