By Scott Sumner | AI at Work | Published October 25, 2026

Two editorial speech bubbles contrast a generic THANK YOU stamp with a reply addressing the actual issue. Answer the person, not the keyword.

AI review replies can save writing time without making your business sound indifferent. Give the tool a narrower job: answer the customer’s actual feedback, using only facts you are comfortable publishing.

Answer: Use AI to draft short, relevant review replies, not promotional templates. Acknowledge the specific feedback, avoid private details and invented fixes, and have an authorized person check the response before it becomes a public statement from your business.

The customer has already chosen you once. A reply to their review should not read like another sales pitch.

Jump ahead

What a review reply is for

A public review reply has two audiences: the person who wrote the review and the next buyer reading it. Both can notice whether the business answered the concern or pasted a paragraph that could fit anyone.

Google’s review-reply guidance is unusually plain: keep it short and simple, be conversational rather than promotional, and avoid sending everyone the same thank-you. It recommends focusing on replies that can answer a question or provide a helpful update.

That is a useful brief for AI. It is not a request for a more elaborate version of “we value your feedback.”

My view: a reply earns its space when it shows that somebody understood what the customer said. Repeating the service name, city and slogan does not do that on its own.

Google’s local ranking guidance says reviews and positive ratings can help local ranking, and that helpful replies can help a business stand out. It does not establish a ranking benefit from inserting keywords into every response. Do not confuse customer reviews with the business’s replies to them.

This is useful work for AI Findability: make the public evidence of your business easier to understand and trust. It is not a guaranteed shortcut to a Google position or an AI recommendation.

A better reply to mixed feedback

Consider this fictional review of a small service business:

“The workshop was useful and the presenter explained things clearly. Getting the joining instructions took two follow-up emails, though.”

A weak AI reply might say:

“Thank you for choosing our premier training services! We are delighted you had an exceptional experience and look forward to exceeding your expectations again.”

It is cheerful. It also ignores the only part of the review the business needs to investigate.

A more useful draft would be:

“Thank you for the feedback on the workshop. I’m sorry you had to chase the joining instructions. Please contact our team through the support address on our website so we can look into what happened.”

That example assumes the business has a monitored support address and someone ready to investigate. If it does not, choose an actual working contact route before publishing.

Notice what the reply does not claim. It does not say the process has been fixed. It does not promise a refund. It does not disclose when the customer booked or what their colleague did. Those details are not needed to acknowledge the problem.

Give AI facts it can safely use

A common drafting mistake is giving the assistant a complaint and asking it to “make this sound reassuring.” Reassurance can turn into an unsupported sentence such as “we have retrained our team” or “this will never happen again.”

Separate the public review from the verified information approved for the reply. If an investigation is still open, say so in the brief. If no corrective action has been confirmed, the draft should not invent one.

Use an approved AI workspace and share only the minimum information needed. Public review text does not make your internal customer record public. There is usually no reason to paste payment details, private emails or a complete support history into a reply-writing task.

Google specifically advises businesses never to share a reviewer’s private information. For complex situations, it recommends asking the person to contact the business privately by phone or email.

Keep sensitive or disputed cases out of automatic publishing. A person responsible for the customer relationship needs to decide what the business can say. This is an operational recommendation, not a claim that Google requires a particular AI approval workflow.

A positive review needs less machinery. If the customer praises a clear explanation, thank them for that. Do not add an upsell because the assistant has room for another sentence.

A practical prompt for drafting review replies

Use this with fictional details first, then adapt it to your approved process:

Draft a short public reply to the review below. Acknowledge the specific feedback in ordinary language. Do not add a sales pitch, service keywords for SEO, private customer information, unverified explanations, refunds or claims that a problem is fixed. Use only the approved facts provided. If an important fact is missing, list a question for the reviewer of this draft separately rather than guessing. Give me one draft, followed by any claims that need confirmation. Do not publish anything.

Public review: [review text]
Approved facts for public use: [verified facts or “none”]
Working contact route, if needed: [approved route]
Preferred sign-off: [authorized name or initials, if appropriate]

The useful output is not just the polished paragraph. It is also the short list of claims that still need checking. A draft that correctly flags a missing contact route is more valuable than one that invents an email address.

You can give the tool examples of your tone, but do not let examples become a universal script. A customer who says “great session” and a customer who says “nobody answered my email” need different replies.

Check the customer issue before the grammar

Before publishing, read the review and reply together. Can you point to the sentence that addresses the actual feedback? Then verify every statement about what happened, what has changed and what the customer should do next.

Check that the contact route works and that the team receiving the message knows what to do. Otherwise you have turned a public concern into another unanswered request.

Google explains that approved replies appear publicly as the business, and the reviewer is notified. The person reading it is not seeing your prompt or your internal approval notes. They see a statement from you.

For an initial review of the process, take five recent replies and ask whether each addressed the feedback, contained any unsupported claims, or sounded promotional. Five is a manageable starting sample, not a scientifically validated threshold.

Do not measure this job only by response speed. A fast reply that misses the issue creates more work for the person who eventually has to resolve it.

Make the next reply worth reading

Start with one mixed review: praise plus a specific frustration. Ask AI for a draft that acknowledges both without inventing a resolution. Have the responsible person check it before publication.

If the review exposes a recurring question about your offer, fix the relevant service page too. Better explanations can reduce the need to repeat the same clarification, although they will not repair a service failure by themselves.

For help connecting your website, public business information and customer trust signals, talk to Sumner Digital. The goal is a clearer buying experience, not a larger pile of automated replies.

Sources

Sources checked October 25, 2026. The examples are fictional. The AI prompt and review process are Scott Sumner’s recommendations, not a new Google policy.