Every UK small business owner knows the guilt of the unanswered review. Someone took two minutes to praise your cafe, your salon, or your plumbing work, and the reply you meant to write is still sitting in your head three weeks later. AI can close that gap, but done lazily it produces exactly the robotic thank-you messages that make customers trust you less. This guide covers how to set up AI review responses properly, what UK law says, and where a human still needs to hold the pen.
What are AI review responses and do they actually work?
AI review responses are replies to Google reviews drafted by a language model that has been given the review text, the customer's details, and facts about your business, then published either automatically or after human approval. They work when the model has enough context to be specific, and they fail when it is left to produce generic gratitude.
The mechanics are simple. A new review lands on your Google Business Profile, a trigger passes it to a model along with a prompt describing your business and tone, and a draft reply comes back in seconds. The difference between a system that helps and one that harms is entirely in that prompt and the review-handling rules around it. A model that knows the reviewer mentioned your Saturday brunch, your emergency call-out, or a specific staff member can write a reply that reads like the owner typed it. A model given nothing writes "Thank you for your kind words, we hope to see you again soon" and every reader recognises the template.
The stakes are worth taking seriously. BrightLocal's 2026 Local Consumer Review Survey, which polled just over 1,000 consumers, found that 97% read reviews for local businesses and 80% are more likely to use a business that responds to all of its reviews.
Why does replying to reviews matter for a UK small business?
Review replies are read by every future customer who scrolls your profile, not just the person who wrote the review. They signal that the business is run by someone paying attention, and consumers increasingly expect them quickly, with BrightLocal reporting that 81% expect a response within a week.
Think of your review page as a shop window you only half control. Customers write whatever they like in it, but your replies are the part you author, and they are on display permanently. According to BrightLocal, 42% of consumers say they are unlikely to use a business that ignores its reviews entirely, and expectations on speed are tightening, with a growing share now expecting a same-day response.
There is also a newer audience reading your replies: machines. The same BrightLocal survey found 45% of consumers now use ChatGPT style AI tools when choosing local businesses, sharply up on the year before, and 82% read AI-generated review summaries. Those systems summarise your whole review page, replies included. A profile full of calm, specific responses to complaints gives them better material than a wall of silence, which is the same reason review activity matters in the wider shift covered in AI search versus traditional SEO for UK businesses.
How do you set up AI review responses step by step?
The setup has four parts: verify your Google Business Profile, connect a trigger that captures new reviews, write a prompt containing your business facts and tone rules, and add an approval step for negative reviews. A basic version takes an afternoon with an automation tool, and a sturdier one is a small custom build.
First, verification. Google's own help documentation is clear that only verified businesses can reply to reviews, so that is the non-negotiable starting point.
Second, capture. You need new reviews to reach the model without you copying and pasting them. Options range from the Google Business Profile API for a custom build, to Zapier or Make watching for new reviews, to an off-the-shelf reputation platform. The custom route gives you the most control over tone, which is the whole game here.
Third, the prompt. This is where most systems are won or lost, and the section below on prompts covers what belongs in it.
Fourth, routing. Four and five star reviews can publish after a light check or automatically once you trust the output. Anything three stars or below goes to a human for approval, every time. Businesses in emotionally charged trades, like the ones covered in the guide for restaurants and cafes using AI, should be strictest about this, because food and service complaints get personal quickly.
What makes an AI reply sound like a robot, and how do you avoid it?
Replies sound robotic when they could apply to any customer of any business: identical openings, no reference to what the reviewer actually said, and corporate phrasing no owner would use aloud. The fix is forcing specificity, varying structure, and keeping replies short.
Consumers have read thousands of templated replies and can spot one instantly. BrightLocal found half of consumers are actively discouraged by generic, templated responses, which means a lazy automation is not neutral, it is a liability. The tells are consistent: every reply opens with "Thank you so much for your kind words", nothing in the reply proves anyone read the review, and the sign-off is "The Team" rather than a person.
The countermeasures are mechanical enough to encode in a prompt. Require every reply to reference one concrete detail from the review. Ban a list of stock phrases. Rotate opening structures so ten consecutive replies do not start identically. Cap length at two or three sentences, because real owners are busy and brevity reads as authentic. Sign off with a real first name.
How should AI handle negative reviews?
AI should draft responses to negative reviews but never publish them unattended. The draft should acknowledge the specific complaint, avoid arguing, state one corrective fact if there is one, and move the conversation offline. A human approves every one before it goes live.
A negative review is the highest-stakes writing your business does in public, and it has an audience of every future customer. The AI's job is to remove the two failure modes humans fall into: replying angrily in the heat of the moment, or avoiding the reply for weeks because it feels confrontational. A model drafts a calm response within minutes of the review landing, and you approve or tweak it when you are calm too.
One detail from Google's documentation matters here: when you reply, the reviewer is notified and can edit their review afterwards. That is the mechanism by which a good response to a two star review sometimes becomes a four star update. Google also moderates replies against its content policies before they appear, usually within minutes though occasionally longer, so keep replies free of anything that could read as an attack or as disclosing private information.
Is it legal in the UK to use AI to answer reviews?
Yes. UK law regulates fake reviews, not automated replies. The Digital Markets, Competition and Consumers Act banned fake reviews from 6 April 2025, targeting reviews that are not based on genuine experience. A reply is openly the business speaking, so drafting it with AI is lawful provided it is truthful.
The legal line is worth understanding precisely, because it sits close to some tempting shortcuts. Analysis by the law firm CMS explains that the DMCC Act's ban, in force since 6 April 2025, covers any consumer review that purports to be, but is not, based on a person's genuine experience, whether positive or negative, and extends to commissioning fake reviews or publishing genuine ones in misleading ways, such as suppressing negatives. Breaches can attract fines of up to 10% of annual global turnover.
So the rules for an AI review system are straightforward. Never generate reviews of your own business. Never pay or incentivise anyone to post reviews without disclosure. Never use automation to bury or filter out genuine criticism. Replies, on the other hand, carry your business name on them by design. Automating their drafting is no different in law from using a template, as long as what they say is true. Do not let an AI invent an apology for a policy you do not have or promise a refund you will not honour.
Should every reply be fully automatic, or should a human approve them?
Run two lanes. Positive reviews can publish automatically once the system has proven itself over a few weeks of human review. Negative and mixed reviews always wait for human approval. This keeps the time saving while capping the downside of a bad automated reply.
Full automation is seductive because it takes the task to zero minutes, but the honest trade-off is that a language model will occasionally misread sarcasm, miss a legal complaint buried in a four star review, or thank someone for feedback that was actually a warning. The two-lane pattern captures most of the saving anyway. In a typical small business, the large majority of reviews are positive and safe to automate, while the handful of difficult ones each month take two minutes of approval rather than an evening of dread.
Start cautious. For the first few weeks, read every draft before it publishes and note the failures. Tighten the prompt after each one. Once a fortnight passes without an edit on positive reviews, let that lane run itself. This is the same crawl-then-automate approach that works across back-office tasks, as covered in how AI cuts admin hours for small businesses.
What should a good AI review response prompt include?
A working prompt contains five things: fixed facts about the business, a description of the owner's voice with example replies, hard rules on length and structure, a requirement to reference a specific detail from the review, and separate instructions for positive and negative reviews.
Treat the prompt as a staff handbook for one very fast employee. The facts section stops hallucination: opening hours, services actually offered, refund policy, the first names used for sign-offs. The voice section works best with three to five real replies you have written yourself and are proud of, because models imitate examples far better than adjectives. The rules section carries the anti-robot constraints: maximum three sentences, no stock phrases, one specific detail from the review, vary the opening, UK spelling.
Then split by sentiment. Positive review instructions focus on warmth and specificity. Negative review instructions add: acknowledge the specific complaint first, never argue or blame the customer, offer one factual correction at most, invite them to contact a named person, and never admit legal liability or promise compensation. Review the prompt monthly, because your services, staff, and prices change, and a reply thanking someone for visiting a location you closed in spring is a robot tell of its own.
How do you measure whether it is working?
Track four things: response rate, time to response, whether any negative reviewers upgraded their rating after a reply, and the trend in your overall rating and review volume. The first two should move within a week of switching on, the others over months.
Response rate is the bluntest and most visible metric, since any visitor can see what proportion of your reviews have replies. Getting it from sporadic to effectively complete is the immediate win. Time to response matters because expectations keep rising year on year, and an automated draft turns your response time from days into minutes plus approval.
The slower signals are the ones that pay. Watch for reviewers editing their rating upwards after a good reply, which Google's notification mechanic makes possible. Watch overall rating and monthly review volume, because a profile that visibly responds encourages more people to bother writing. None of this needs a dashboard on day one, a monthly fifteen minute check in your Business Profile stats is enough to know whether the system deserves more autonomy or a tighter leash.
Review responses are one of the rare automations where the AI version, configured properly, is not a cheaper imitation of the human version but a more consistent one. The owner's judgement still decides what the business will say to an unhappy customer. The machine makes sure the saying actually happens, every time, within the hour rather than within the month.
AI That Answers Your Google Reviews (Without Sounding Like a Robot) — FAQ
Can AI really write Google review responses that do not sound automated?
Yes, but only if it is given real context. A model that sees the review text, the customer's first name, what they bought, and a few facts about your business can write a reply that reads as human. A model given nothing produces the generic thank-you messages people have learned to skim past. BrightLocal's 2026 Local Consumer Review Survey found that 50% of consumers are put off by generic, templated responses, so a badly configured AI can do more harm than good. The practical fix is a prompt that requires the reply to reference one specific detail from the review, vary its opening line, keep to two or three sentences, and sign off with a real first name. With those constraints in place, most readers cannot tell the difference, and the replies go out in minutes rather than sitting in a queue for a week.
Is it legal in the UK to use AI to reply to Google reviews?
Replying with AI assistance is legal. What UK law now targets is fakery, not automation. Under the Digital Markets, Competition and Consumers Act, a ban on fake reviews came into force on 6 April 2025, covering any review that purports to be, but is not, based on a person's genuine experience, according to analysis by the law firm CMS. That means you must never use AI to write reviews of your own business, commission them, or suppress genuine negative feedback, and breaches can attract fines of up to 10% of annual global turnover. A reply, by contrast, is openly the business speaking, so automating it is no different in law from using a template or asking a member of staff to draft it. Keep replies truthful and you are on safe ground.
Should AI reply to negative reviews automatically?
No. Negative reviews are the one category where a human should always approve the reply before it is published. The reputational cost of a tone-deaf automated response to an angry customer far outweighs the minutes saved. The sensible pattern is a two-lane system: four and five star reviews get an AI drafted reply that publishes automatically or after a light skim, while one to three star reviews get an AI draft held for human sign-off. The AI still does the heavy lifting, pulling up the customer's history and drafting a calm, specific response, but a person decides whether it goes out. Remember too that Google notifies the reviewer when you reply and they can then edit their review, so a good response to a bad review is genuinely worth crafting.
How quickly should a business respond to Google reviews?
Faster than most UK small businesses currently manage. BrightLocal's 2026 survey found 81% of consumers expect a response within a week, and a growing share now expect one the same day. Expectations are moving in one direction. This is precisely where automation earns its keep, because the hardest part of review responses is not writing them, it is doing it consistently on a Tuesday afternoon when the phone is ringing. An AI system drafts the reply the moment the review lands, so the only delay is however long your approval step takes. Note that Google itself moderates replies before they appear, which usually happens within minutes but can occasionally take longer, so the reply timestamp is not entirely in your control.
Do review responses actually affect whether new customers choose you?
The evidence says yes. In BrightLocal's 2026 Local Consumer Review Survey of just over 1,000 consumers, 97% said they read reviews for local businesses, and 80% said they are more likely to use a business that responds to all of its reviews. Responses are also read by far more people than the original reviewer, because every future customer scrolling your profile sees how you handle both praise and complaints. There is a second-order effect as well: review pages are increasingly summarised by AI search tools, and the same survey found 45% of consumers now use ChatGPT style tools when choosing local businesses. Thoughtful, specific replies give those systems more favourable material to summarise than a wall of ignored complaints.
What tools do I need to set this up?
Three pieces. First, a verified Google Business Profile, since Google only lets verified owners reply to reviews. Second, a way to get new reviews out of Google and in front of a model, which can be the Business Profile API for a custom build, a Zapier or Make trigger for a lighter setup, or a reputation platform with AI built in. Third, the model itself with a well written prompt holding your business facts, tone rules, and reply constraints. A capable generalist model such as Claude works well because the task is mostly tone and judgement rather than specialist knowledge. The build is a few hours of work for someone comfortable with automation tools, or it can be handed to a studio that sets up this kind of system as part of a wider admin automation package.
