Do Google Reviews Affect Whether AI Recommends Your Business?

Reviews are the one lever every UK business owner can see and move, so it is no surprise that the first question about AI search is whether they matter. They do. But the way they matter in AI answers appears to be different from the way they matter in the Map Pack, and the difference changes what you should do.

I have written before that review recency beats review volume for Google's local ranking. This post asks the follow-up: what happens when the thing choosing you is an AI assistant rather than a ranking algorithm? The evidence is thinner and more vendor-driven, so I will be clear about its limits as I go.

What does the evidence say about reviews and AI recommendations?

A 2026 study found that AI recommendations consistently favoured businesses with above-average sentiment, and that reviews acted as a filter rather than a ranking signal. The study is SOCi's 2026 Local Visibility Index, reported by Search Engine Land, covering nearly 350,000 locations across 2,751 multi-location brands.

SOCi's interpretation is that in traditional local search a business with an average or even middling rating can still rank on proximity and category relevance, whereas in AI-driven results those same locations were frequently excluded altogether, because AI systems prioritise confidence and risk reduction over breadth.

What does 'filter, not ranking factor' mean?

It means reviews decide whether you are eligible, not how high you place. In the Map Pack, more and better reviews push you up a ranked list. In an AI answer that names three businesses, the system may simply drop anybody who looks risky to recommend. A rating that sits below the others is a reason to be excluded, not a reason to be listed fourth.

The study's financial-services example makes the point. Brands with low profile accuracy, average ratings near 3.4 stars and review response rates below 5% were effectively invisible in AI recommendations. Weak fundamentals, in SOCi's words, translated directly into no AI visibility. Replace 'financial brands' with a London clinic or law firm and the logic is the same.

Why does response rate come up?

A reply is evidence that a real, attentive business is behind the profile. It is also a free, high-signal action. A profile where every review, good or bad, gets a considered answer looks maintained. A profile with hundreds of unanswered reviews looks abandoned, and an abandoned profile is a risk for a system that wants to recommend with confidence.

• Reply to every review, including the five-star ones, within a few days.
• Answer criticism calmly and factually, without defensiveness and without exposing private details.
• Mention the service and area naturally where it genuinely fits, since replies are text a system can read.
• Keep replies human. A pasted generic reply on every review reads as automation.

Does the content of reviews matter, not just the stars?

Plausibly yes, because review text is language an AI system can read. A review that says 'the plumber arrived within the hour and fixed the leak in our Brixton flat' contains the service, the speed and the area. That is the sort of detail that supports a recommendation, whereas 'great service' supports nothing in particular.

I want to be careful. I do not have evidence that assistants weight review text in a specific way, and I would not write a script for customers. You cannot instruct people what to say, and you should not try. What you can do is make it easy to leave a review at the right moment, when the job is fresh and specific, and ask an open question such as 'what did we help you with?' That produces detail honestly.

Which platform's reviews count?

It depends on the assistant, and the honest answer is that nobody outside these companies knows precisely. SOCi describes Gemini as grounded in Google Maps, which implies Google reviews matter there. Google's own guidance says prominence is based partly on how many reviews and ratings a business has, and that more reviews and positive ratings can help local ranking.

• Google reviews first, because they are attached to the profile that powers the Map Pack and, in SOCi's description, Gemini.
• One or two industry platforms your customers genuinely use, such as the sector directories and trade review sites that apply to you.
• Consistency across them. A strong profile in one place and a poor one elsewhere is a mixed signal.
• Do not spread yourself thin. Depth and recency on Google beats a thin presence on ten sites.

What does UK law say about getting reviews?

Fake reviews, hidden incentives and review gating are unlawful or banned for UK businesses. The Digital Markets, Competition and Consumers Act 2024 brought fake consumer reviews within the list of banned unfair commercial practices, and its consumer provisions came into force in April 2025. Check current guidance from the Competition and Markets Authority before you design a review process.

• Never write or buy reviews, including through agencies offering packages.
• Never gate. Screening customers so only happy ones are asked for public reviews breaches Google's policies and is a misleading practice.
• Disclose incentives if you reward feedback, and do not condition a reward on a positive review. Better still, do not offer one.
• Ask every customer, not a selected set.

These rules protect you as well as customers. A review profile built honestly survives a competitor's report, a platform sweep and an AI system's confidence checks. One that was manufactured does not, and the damage tends to arrive all at once.

How do I compare my profile with the field?

List the businesses that currently appear when you run your prompts, and compare them on rating, recency, volume and response behaviour. 'Above average' is relative to who else is in the answer, so the useful comparison is not a national figure. It is the five names an assistant gives for your service in your part of London.

Can I get unfair reviews removed?

Only if they break Google's policies, and a negative review you simply dislike does not qualify. Reviews that are fake, from someone with a conflict of interest, off-topic or otherwise policy-violating can be flagged through the profile, and you can keep a record of the reason you flagged each one. Removal is not guaranteed and can be slow.

• Flag what genuinely breaks the rules, with a brief factual reason.
• Reply to the rest, calmly and without disclosing private details.
• Do not pay for removal services, which can themselves breach platform rules.
• Dilute honestly. A steady flow of genuine recent reviews does more than any removal.

What if my rating is not great?

Fix the cause, then dilute the history with honest recent reviews. A 3.7 earned from a genuine service problem will not be cured by marketing. Address what customers are actually saying, and then make it easy for the happy majority, who rarely volunteer a review, to leave one.

1. Read the last fifty reviews and group the complaints. Fix the top one before asking for more reviews.
2. Reply to each negative review with what you have changed, where true.
3. Add a request to the end of every job, a one-tap link sent while the work is fresh.
4. Track the rolling average of the last ten and the last thirty reviews, not just the lifetime figure.
5. Be patient. Ratings move slowly, and a sustained flow matters more than a burst.

What would I do in a London business this month?

Treat reviews as qualification: get above the line, answer everything, and keep the flow running. I would first check the current rating against the three nearest competitors, since 'above average' is relative to the field. I would clear the backlog of unanswered reviews, set up the one-tap request, and re-run an AI visibility check in a quarter, as described in how to check your AI visibility.

The tidy summary is that for AI assistants a good reputation is the entry ticket, and the other work, accurate facts and clear, specific content, is what gets you named once you are inside. It is the same layering I describe in the post on getting recommended by ChatGPT.