Quick answer: yes — as a drafting assistant, with you as the editor. An AI assistant can turn a painful review into a calm first draft quickly, and that alone is useful on a bad day. But a reply is a public statement made in your business's name, the reviewer is told when you post it, and the AI has no idea what actually happened. So let it draft, then run six checks before anything goes live: facts verified, no unconfirmed admissions of fault, no private details echoed, tone matched to severity, a real next step offered, and the rules of the platform — and of consumer law — respected.
What a review reply actually is
On a Google Business Profile, a reply is not a private message. It is posted publicly under the customer's review, the reviewer is notified when you reply, and customers can still update their review after reading it. Google also checks replies against its content policies before they appear. Google's own guidance on managing reviews covers the mechanics.
Replies also get read. Research by Which?, cited by the Competition and Markets Authority (CMA) in March 2026, found that 89% of UK adults use reviews when researching a product or service. Whoever is choosing between you and the business down the road can see how you handle the awkward ones — and a calm, specific reply to a two-star review often says more about a business than the review itself.
Where AI genuinely helps
- First drafts when emotions are high. Staring at an unfair review while angry is how defensive, regrettable replies get written. An AI assistant is not upset, so it will produce a level-headed draft you can correct rather than a rant you have to retract.
- Tone and length edits. "Make this shorter, less defensive, no jargon" is the kind of edit assistants do well, saving you the worst part of the job.
- Bilingual replies. If your business serves customers in English and Portuguese, the replies should be in the language of each review. An assistant can help draft and translate them; someone who knows the language should still read the final version, because sarcasm and idiom are exactly where machine translation stumbles.
- Themes across many reviews. With a few dozen reviews, an assistant can summarise recurring complaints ("parking", "waiting times on Saturdays") far faster than reading them all in one sitting. Only paste reviews you are comfortable putting into that tool, for the same reasons an AI usage policy exists.
One thing no assistant can do: know your side of the story. Whichever tool you use — ChatGPT, Claude, Gemini or another — it knows nothing about your business beyond what you paste into it, and it will not tell you who was right.
Where a human must stay in charge
- Factual disputes. "They overcharged me" may be a till error or a misread menu. The AI cannot know, so the reply must wait until someone in the business has checked.
- Complaints with legal weight. Accusations of discrimination, safety issues or demands for refunds deserve a considered response, often privately first. This is not the place for a quick draft.
- Reviews that name staff members. A public reply naming an employee back at the reviewer drags a private personnel matter into public. Handle it internally; keep the reply about the customer's experience.
- Suspected fake reviews. Do not debate a review you believe is fake under the review. Report it through the platform — Google lets you report reviews that violate its policies — and say your piece, if at all, in one calm sentence.
A constructed example
A fictional café receives a two-star review: "Waited 25 minutes, staff were rude, and they overcharged me for a flat white."
Asked for a reply, an assistant will typically produce something like:
"We are so sorry to hear that our staff were rude and that you were overcharged. That is not the standard we hold ourselves to..."
That is two admissions of fault nobody has checked. If the till log shows the correct price was charged, the reply has just confirmed an inaccurate story — publicly, in the business's own words.
After checking, the café writes (constructed for illustration):
"Thank you for taking the time to review us, and sorry about the wait that Saturday — weekends stretch us thin. On the price: the flat white is £3.60 on our menu board and on every till receipt, so we would genuinely like to see your receipt and look into it — please get in touch. We have also shared your comments about the service with the team."
No unconfirmed admission; one verifiable fact stated plainly, without calling the reviewer a liar; a route to resolution; and the service complaint acknowledged without pretending anyone knows exactly what happened at the table.
Six checks before you post an AI-drafted reply
- Facts. Every factual statement in the reply was verified inside the business — not inferred from the review itself.
- Fault. No admission of fault you have not confirmed. (If you did get it wrong, a plain apology and a fix is usually the right answer — the check is about deciding deliberately, not about never apologising.)
- Privacy. The reply does not repeat personal details from the review beyond what is needed, does not name staff or other customers, and whatever you pasted into the AI tool had names and details removed first.
- Tone. It matches the severity of the review. Copy-pasting near-identical replies to every review is noticeable, and readers discount them.
- A real next step. The reply moves the issue to somewhere it can actually be resolved — a phone call, an email, a conversation — rather than only performing for onlookers.
- The rules. Platform content policies apply to replies too. And since April 2025, obtaining or posting fake reviews, and paid-for reviews that are not clearly marked as incentivised, are "banned practices" under the UK's Digital Markets, Competition and Consumers Act 2024, per the CMA's announcement. In the EU, rules in force since May 2022 under Directive (EU) 2019/2161 require traders publishing consumer reviews to be transparent about whether and how they check reviews come from real customers, and prohibit submitting or commissioning fake reviews (Directive (EU) 2019/2161). The practical rule that keeps you clear of all of it: never make discounts, refunds or freebies part of review handling. (This is a practical summary of published rules, not legal advice.)
What AI is bad at, here
It does not know what happened, and assistants are tuned to please — which sometimes means drafting the soothing version rather than the accurate one. It misreads sarcasm, and translated replies can lose the politeness you intended. It cannot reliably tell a genuine grievance from a fake review. And because the reviewer is notified the moment your reply posts, edits after the fact help, but the first version is the one most people will see.
When you do not need AI
If a handful of reviews come in each month, write the replies yourself. It takes a couple of minutes each, and a reply in the owner's own words, with a specific detail only the business would know, beats any draft. AI earns its place when volume makes replies a chore — or when one particular review has wound you up enough that a level-headed first draft is worth having.
Where a second pair of eyes helps
Reading about checking replies is one thing; doing it on your own reviews, where you are the one who is annoyed, is another. That is what a Personal AI Session is for: one 60-minute, online, one-to-one session on a prepared, anonymised task — for example, drafting and checking replies to a few of your real reviews with names and details removed — in English or Portuguese. It costs £75 per session (89 € in Portugal), includes a short written recap within two working days and one clarification email thread — up to 15 minutes in total — within seven days of the recap, and requesting a session creates no booking or payment obligation.
Prefer to talk first? Book a free 30-minute call — it is optional, and it reserves nothing.
Related reading: AI customer service for small business, does your small business need an AI usage policy? and how to check AI spreadsheet output before you trust it.
Last updated: 31 October 2026