How to Use AI to Respond to Customer Reviews
The reply you write is not for the person who left the review. It’s for the hundreds of prospective customers who’ll read that exchange over the next two years and decide what sort of business you are. Once you accept that, the whole job changes — you stop defending yourself and start demonstrating competence.
AI helps with this more than most tasks, for one specific reason: it gives you a usable draft while you’re still annoyed, which is exactly when businesses do the damage.
This is the working process — a reply framework, a triage step, and a prompt for each type of review you’ll actually receive, including the ones that are unfair or fake.
Step 1: Learn the four-part reply structure
Every good reply does these four things, in this order
Almost every effective review reply follows the same shape. Learn it and you’ll spot immediately when an AI draft has gone wrong.
The four parts — and what each is for
| Acknowledge | Thank them and name their actual experience in one sentence. Not “sorry you feel that way” — that reads as dismissal to everyone watching. |
| Address | Respond to the specific point, briefly. One or two sentences. This is where businesses over-explain and lose. |
| Action | State what you’ve done or will do. Even “we’ve changed how we confirm appointments” is enough. Future customers care about this most. |
| Offline | Give a direct route to a person — a name and an email or number. Then stop writing. |
Under 100 words total. The discipline of brevity is doing real work: a long reply signals defensiveness, and a reader scanning your profile will judge the length before they read a word of it.
Step 2: Triage before you draft
Five types of review, five different replies
This is the step people skip, and it’s why so many AI-drafted replies land badly. A fair complaint, a factually wrong complaint and a fake review need genuinely different responses, and if you paste all three into the same prompt you’ll get the same apologetic tone for all of them.
Decide which you’re dealing with first: a fair complaint where something did go wrong; a mistaken complaint where the facts are wrong but the person is real and sincere; an unfair or hostile review where the expectation was never reasonable; a suspected fake from a competitor or someone who was never a customer; or a positive review, which most businesses waste.
Then use the matching prompt below.
Step 3: The prompts
When something genuinely went wrong
What actually happened, honestly, including what was our fault: [explain]
What we’ve done or will do about it: [be specific]
Direct contact for follow-up: [name and email/phone]
Write a public reply that:
– Thanks them and names their actual experience in one sentence — do NOT use “sorry you feel that way”
– Responds to the specific issue in no more than two sentences
– States what we’ve changed or done
– Offers the direct contact and then stops
– Is under 100 words
– Does not argue, over-explain, list events, or mention anything they did
Give me two versions: one warmer, one more matter-of-fact.
When the customer has the facts wrong
The temptation here is to correct them thoroughly. Resist it — a reader who sees a business winning an argument with a customer still forms a negative impression of the business.
What’s actually correct: [the facts]
What I can evidence if needed: [what records you hold]
Direct contact: [name and email/phone]
Write a public reply that:
– Acknowledges their frustration genuinely, without conceding the disputed facts
– Corrects only the single most important factual point, in one calm sentence, without saying they’re wrong
– Invites them to continue privately so we can look at it properly
– Is under 90 words
– Does not list evidence, does not defend at length, does not sound like a lawyer
Then separately, tell me what I should say in the private conversation.
When the review is unfair or hostile
Context from my side: [explain]
Write a public reply that stays completely calm and professional and is under 70 words. It should acknowledge the review exists, state our position once in a neutral sentence, offer a route to discuss it, and stop.
Important: assume other customers are reading this and judging me, not them. My goal is to look reasonable, not to win. Do not be passive-aggressive, do not use phrases like “we’re sorry you chose to” or “as explained at the time”, and do not hint that they’re being difficult.
A calm reply under an obviously unreasonable review is quietly persuasive in a way no rebuttal is. Readers can usually tell, and they notice who kept their composure.
When you think the review is fake
Why I think that: [reasons — no matching record, details don’t match anything we do, etc.]
Two things:
1. Write a short, neutral public reply (under 60 words) that notes we can’t find a matching record, invites them to contact us with order details so we can look into it, and does NOT accuse them of anything.
2. Tell me exactly what to include when I report this to the platform, and which platform policies it might breach.
Report it through the platform’s process with specifics, and reply neutrally in the meantime. Publicly accusing someone of faking a review is a risk you don’t need — the neutral “we can’t find a record of this” reply signals the same thing to readers without the exposure.
When it’s a four-star with a niggle
These are the most useful reviews you get and the most commonly ignored. Someone liked you and told you exactly what to fix.
Write a reply under 70 words that:
– Leads with genuine thanks for the specific thing they praised
– Acknowledges the criticism directly rather than glossing over it
– Says what we’re doing about it, if anything
– Doesn’t ask them to change their rating
Then tell me: is the criticism something I should actually fix, or is it a one-off? Ask me a question if you need more context to judge.
Positive reviews — batch these weekly
Write a short reply to each — under 40 words, all different from each other, each referencing something specific that reviewer actually said.
Rules: no template phrasing repeated across replies; don’t say “thanks for the 5 stars”; don’t add a sales pitch; sound like a person who remembers the job. If a review mentions a staff member by name, mention them too.
Never offer anything in exchange for changing or removing a review. Cash, a discount, a freebie, a future credit — in the UK, incentivised reviews where the incentive isn’t disclosed are a banned practice under consumer protection law, with substantial penalties. Fix the customer’s problem because it’s the right thing to do, say so briefly in public, and let them decide independently what to do about the review. Many update it anyway.
Never suppress, hide or filter negative reviews. Publishing reviews in a misleading way — including hiding the bad ones or presenting an average that excludes them — is also caught. Our guide to tools for responding to negative reviews covers the legal position and current enforcement in detail.
Step 4: Mine your reviews for what they’re telling you
Your reviews already contain your operational to-do list
Once a quarter, run everything through this. It’s the highest-value use of AI in this whole process, because replying well protects your reputation while fixing the cause improves the business.
Tell me:
1. The three things customers consistently praise — I should be using this language in my marketing
2. The three problems that come up repeatedly
3. Any single operational change that would remove the most complaints
4. Any phrase customers use to describe what I do that I’m not using myself
5. Anything mentioned only once that I should still take seriously
Be direct. If a pattern suggests something uncomfortable about how we work, say it plainly.
Strip names before pasting. If you handle reviews at volume and want this running continuously rather than quarterly, that’s when review-management software starts earning its subscription.
Before you post any reply, check
- Under 100 words. If it’s longer, you’re defending.
- No customer details beyond what they revealed themselves.
- No money mentioned — no refunds, discounts or credits in public.
- No argument. Read it as a stranger: does the business sound reasonable?
- Every claim true — if you say you’ve changed something, you’ve changed it.
- A real person’s name and contact for the offline route.
- Read it aloud. Passive-aggression is much easier to hear than to see.
Frequently asked questions
Can I automate review replies completely?
Positive reviews, yes — a warm, varied thank-you generated and sent automatically is low-risk and better than the silence most businesses offer. Negative reviews, no, and this isn’t a soft recommendation. A fully automated reply to a serious complaint is how businesses produce the screenshots that circulate: cheerful boilerplate under an account of a genuine failure, or a template that reads a safety issue as a service niggle. Automation also can’t recognise when a review touches something with legal or regulatory weight, which is exactly when the wording matters most. The workable setup is AI drafting every reply, a human reading every negative one before it posts, and a firm rule that anything mentioning injury, discrimination, a regulator, a legal threat or a factual dispute goes to whoever runs the business rather than whoever is on shift.
Should I reply to positive reviews too?
Yes, and it’s the cheapest win in this whole guide because so few businesses bother. Three reasons. It signals to anyone browsing that you’re present and attentive, which is most of what a review profile is actually communicating. It makes the reviewer more likely to return and to recommend you, because being acknowledged matters. And practically, a profile where every review has a reply looks considerably more professional than one where only the complaints got answered — which, incidentally, is how it looks when you only reply to negatives. Batch them weekly using the prompt above, keep each under 40 words, and make sure they’re genuinely different from each other. Identical thank-yous repeated twenty times are worse than none, because the copy-paste is obvious and it undermines the impression you were trying to create.
How fast do I need to reply?
Within 24 to 48 hours for negatives, and weekly is fine for positives. Speed matters more for the reader than the reviewer: a prompt reply signals a business that pays attention, while a reply three months later — or none — tells someone browsing your profile that nobody’s home. There’s a practical benefit too, in that responding quickly sometimes catches the customer while the situation is still recoverable, and a resolved complaint occasionally becomes an updated review. What you shouldn’t do is reply within the first hour while you’re still annoyed, which is precisely where AI helps: draft it immediately, then leave it and read it again before posting. If a review lands at 10pm, draft it then and post it in the morning. The draft-now-send-later habit prevents most of the replies businesses regret.
What if the review is genuinely defamatory?
Take legal advice before doing anything, and be aware the bar is much higher than “unfair” or “untrue in one detail”. Genuine dissatisfaction, robust opinion and even harsh characterisation are generally not defamation, and pursuing a customer legally very often makes the situation dramatically worse and considerably more public — a complaint nobody would have noticed becomes a story. Practical sequence: reply calmly in public using the hostile-review prompt above, report it to the platform if it breaches their published rules on abuse, personal information, conflicts of interest or content unrelated to an actual customer experience, and gather your own records of what happened. Only then, if it’s genuinely damaging and clearly false as a matter of fact rather than opinion, speak to a solicitor. Do not let AI draft anything resembling a legal threat.
The bottom line
Write for the audience, not the reviewer — everyone reading your profile in the next two years is the actual audience, and they judge tone far more than facts. Use the four-part structure, keep it under 100 words, triage the review type before drafting because a fair complaint and a fake one need different replies, and always take disputes offline. AI’s real contribution is giving you a usable draft while you’re still cross, so draft immediately and send when calm. Then run your reviews through a pattern analysis quarterly, because replying well protects the business and fixing the cause improves it.
Legal information is general orientation only and reflects the UK position at the time of writing: incentivised and suppressed reviews are subject to consumer protection law with substantial penalties, and enforcement is active. Read the CMA’s own guidance and take proper advice before acting. Nothing here is legal advice.