What your Google reviews really mean for your salon's AI visibility

Why reviews carry more weight than they used to

For years the arithmetic was simple: more good stars on Google, more walk-ins. Someone read three reviews, got a feeling, and phoned. That still happens. What has changed is where the question gets asked first, because a growing share of it never reaches a results page at all - it goes straight into ChatGPT, Gemini or the AI summary sitting above Google's own results. Google says those generative features run on its core Search ranking and quality systems, so this is not a separate index with a separate trick to learn. It is the same open web, read by a machine instead of a person.

The scale of that is not speculative. Google's Gemini app passed a billion monthly active users in August 2026, and AI Overviews now reach over two billion people a month. Pew Research watched the real browsing of 900 US adults across nearly 69,000 Google searches: when an AI summary appeared, people clicked through to an actual website in 8 percent of visits, against 15 percent when no summary appeared - and clicked a link inside the summary itself in just 1 percent. Google disputes the methodology. Either way, for a lot of people the summary is now the entire visit, and whatever it says about your salon is all they will ever read.

What an engine can actually extract from a review

Say a client writes: finally a hairdresser who understands curly hair and did not talk me into cutting it all off. To you that is a compliment. To a retrieval system it is something more useful - a short, quotable passage containing the words curly hair and a specific claim about how you work. When someone later asks where they can get curls cut without losing the length, that sentence is exactly the kind of text an engine can lift and attribute. Five stars with no words attached gives it nothing to lift.

That is why the detail in a review, not the sentiment, is the part that does any work for your AI visibility. A blanket great, will be back is a perfectly good trust signal for a human and gives a machine nothing. Reviews that name the actual service - root touch-up, wedding updo, men's cut with a beard trim, Olaplex treatment - put your salon in the same sentence as the words people type into an engine. That is the vocabulary a question gets matched against, and it is vocabulary you cannot write yourself. It counts precisely because somebody else wrote it.

This is also the one lever with real evidence behind it. Ahrefs looked at roughly 75,000 brands and found that how often a brand is mentioned across the web correlates with how often AI engines cite it at about 0.664 - roughly three times the correlation for backlinks, at 0.218. Third-party mentions are the strongest signal anyone has managed to measure, and reviews are third-party mentions you can honestly ask for, in volume, every day you are open. So ask a happy regular to write a line or two about what they actually had done, in whatever language they searched in. It sounds trivial. It is the difference between a business the web barely mentions and one it mentions constantly.

Why the star average is the least useful number you have

Most owners watch one figure: the average. 4.8 beats 4.6, so 4.8 must be winning. That is the least informative way to read your own profile. None of the engines publish how they weigh anything - OpenAI, Perplexity and Anthropic all describe retrieving live web results for questions about businesses, and not one of them publishes a scoring formula, so anybody quoting you weighted signals is inferring, not reporting. What you can reason about is how much there is to retrieve. Nine short reviews from three years ago give a system almost no text and no evidence you are still trading. A couple of hundred detailed ones, some from last week, give it plenty of both.

Recency matters for a duller reason than ranking: these systems state wrong things about businesses with total confidence. Columbia's Tow Center ran 1,600 queries across eight AI search tools and more than 60 percent of the answers were incorrect, ranging from 37 percent for the best tool to 94 percent for the worst - and ChatGPT signalled any uncertainty in only 15 of 200 cases, never declining to answer. If the freshest thing written about you is two summers old, you have left the machine guessing about your prices, your hours and whether you still offer the service being asked about. A steady trickle of new reviews keeps that guessing narrow. A salon seeing dozens of people a week can produce it without effort - it just has to ask consistently.

Asking for reviews without making it weird

The biggest lever here is unglamorous: ask, at the right moment. That moment is the mirror, the second she sees the colour or the cut and is genuinely pleased. Willingness peaks there and falls away the whole journey home. One sentence at the till does most of the work: if you are happy with it, a short Google review helps us more than you would think - and please say what we actually did.

Then take the friction out. A QR code at the till, or taped beside the mirror, pointing straight at the review form converts far better than good intentions do; plenty of people mean to and have forgotten by the time they park the car. What you must not do is buy it or stage it. No discount for stars, no printed paragraph to copy out, no reviews written by staff. Google polices this, and its own guidance on AI features makes the parallel point about websites: content produced to work on machines rather than to inform people is treated as abuse. Faked reviews are the quickest way to lose a profile you spent years earning.

One thing worth doing deliberately: ask for reviews on the services you want to be known for. If balayage is what you want the phone ringing about, ask the balayage clients specifically. You are not only lifting a star count - you are choosing which words accumulate next to your name, and those words are what an engine matches a question against.

Your replies are text about your salon too

Most salons either ignore their reviews or answer with thanks. That is a wasted line of text. Your reply sits on the same page as the review, in the same crawlable block, and it is the only part of the exchange you get to write. Answer with something like: glad the copper tone suits you, and Sarah always enjoys a proper balayage - and you have put the service, the shade and the stylist into the record in your own words, hanging off somebody else's testimony rather than your own marketing copy.

Criticism is the same job. One bad review does not sink you with people or with machines; what gets read is whether anyone was home. A calm, specific reply - here is what went wrong, here is what we did about it - is a public record of a business that answers. Silence, or a sarcastic comeback, is also a public record. Five minutes a week, and it is the cheapest reputation work available to you.

What reviews cannot do

Straight with you: reviews are not a switch that puts you at the top of every AI answer. Two findings should temper expectations. Profound analysed around 730,000 ChatGPT conversations and found only about 18 percent trigger a web search at all - the rest are answered from what the model already absorbed. Among the ones that do cite sources, there were roughly six unique citations. So many questions never look anything up, and when they do, the slots are few. Reviews sit alongside your Google Business Profile, the directories, your own site and any local coverage. Anyone selling you AI visibility as purely a review-count problem is selling you a story.

The systems also keep moving. Citation sources are volatile: ChatGPT's share of citations coming from Reddit reportedly fell from around 60 percent to around 10 percent inside about a month in 2025, after a change on the platform side. So do not build on any single engine's current habits, not least because Gemini or Perplexity choose their sources differently anyway. The useful part is that a salon with a large, recent, specific review record reads well under all of them. You are not tuning for one algorithm; you are building the thing every one of them has to read. That work does not go stale when a model updates.

Three habits that fit in a working week

Adding AI visibility to a diary that is already full sounds like another job. Reduce it to three habits instead. One: ask every genuinely happy client, at the mirror, for a review that says what you had done. Two: block ten minutes a week to answer everything new. Three: keep the Google Business Profile literally correct - services, hours, a few real photographs - because clean, machine-readable facts are what stop an engine inventing your closing time. Google has been clear that markup is not a ranking factor; it is there so the facts come out right.

Those three routines separate a salon an engine knows by accident from one it can describe with confidence. No agency required, nothing technical to learn. What it takes is consistency, plus the habit of remembering that a review now does two jobs: it persuades the person reading it, and it feeds the machine answering for everyone who never gets that far.

Where this is heading

The shift is measurable rather than speculative. SparkToro and Similarweb found that 68 percent of US Google searches in early 2026 ended with no click to any website, up from 60 percent two years earlier - about 276 clicks per thousand searches reaching the open web, down from 374. And Ahrefs found only 6 to 8 percent of the URLs ChatGPT cites also sit in Google's top ten for the same query, so ranking well is no longer proof you are being recommended. The conversation that decides your next new client increasingly happens somewhere you cannot watch. You only see the outcome: someone in your chair saying she came across you online somehow.

So start small, and start this week. Ask the next obviously delighted client, on purpose. Answer the three oldest reviews sitting there unanswered. Open your Business Profile and check that every service you actually sell is listed. Each of those makes you a little more legible to a machine that is already answering questions about hairdressers in your area, whether you take part or not.