Google’s review policies are explicit about what gets taken down: spam, off-topic content, conflicts of interest, restricted content. Inaccuracy is not on the list. Neither is unfairness. A guest can write something that never happened, and as long as it breaks none of those rules, it stays on your profile and it counts toward your average.
That asymmetry is the whole problem, and it hits boutique properties harder than anyone else in lodging. A 400-room airport hotel absorbs a bad review the way an ocean absorbs a bucket. A twelve-room property does not have an ocean. It has a bucket.
The five surfaces where your reputation actually lives
Most operators think of reputation as one number on one platform. It is five, and they behave differently.
The first is your Google Business Profile, which is the surface most likely to be seen by a guest who already knows your name and is checking you out. The second is your OTA review streams on Booking.com and Expedia, which are gated to verified stays and therefore carry more weight with readers who know that. The third is TripAdvisor, which is not gated and which skews toward the strongly pleased and the strongly annoyed. The fourth is your own site, which you control and which most properties waste. The fifth is the layer nobody managed until recently: the summary an answer engine produces when someone asks which boutique hotel in your city is worth booking.
That fifth surface is assembled from the other four, and it does not show its work. When ChatGPT or Gemini describes your property as “charming but with reports of noise issues,” it has read a pattern across dozens of reviews and compressed it into a clause. No single review said that. The pattern did. You cannot respond to a clause, which means the only way to change it is to change the pattern underneath it.
Call this the Five-Surface Audit, and run it as one exercise rather than five. Open each surface, read the most recent twenty entries, and write down every recurring noun. Not the sentiment, the nouns. Wifi. Shower pressure. Street noise. Breakfast. Parking. The nouns are what compress into the summary.
Why one bad review costs a boutique hotel more
The math is not subtle and most operators have never run it. If you have sixty reviews at 4.7 and you take a one, your average drops to roughly 4.64. If you have six hundred reviews at 4.7 and you take a one, it drops to 4.69. Same review, one tenth the damage.
This is the small-denominator problem, and it produces a counterintuitive priority. For an independent property under about two hundred reviews, the highest-value reputation work is not responding to criticism. It is increasing the denominator. Every additional legitimate review you generate makes the next bad one cost less, permanently.
Most properties do this backward. They put enormous effort into the twelve people who complained and no effort into the two hundred who left happy and silent. The happy and silent guests are the asset. They are also the easiest group in your entire business to move, because they already feel warmly toward you and are simply waiting to be asked.
There is a second effect that compounds the first. Rating is not only a number a guest reads, it is an input into how the platforms rank you. Google, Booking.com and TripAdvisor all weight some combination of rating, review volume and review recency in their default sort, and a property with forty reviews is penalised on two of those three axes regardless of how good the forty are. Volume is not vanity. It is a ranking input that you can move without spending anything.
Recency deserves its own note, because it is the one operators forget. A property with two hundred reviews, none of them from the past four months, reads as stale to a browsing guest and often ranks as stale to the platform. The guest reading your profile in January is looking at the dates, working out whether the ownership has changed, whether the refurbishment they read about happened, whether anyone still cares. A steady trickle is worth more than an occasional flood, which means the asking has to be part of your operation rather than a campaign you run when the rating dips.
What can you actually get removed?
Less than you hope, and more than you think.
Reviews that name a specific staff member in a way that reads as harassment are removable. Reviews describing an incident at a different property, which happens more often than you would guess with similarly named hotels, are removable. Reviews posted by someone with an obvious conflict of interest, including a competitor, are removable if you can demonstrate the connection. Reviews containing profanity or discriminatory language are removable. Reviews that are actually about a third party, such as a food delivery driver or a rideshare, are frequently off-topic and removable.
What is not removable: a guest who says your room was small when your room is in fact small. A guest who was angry about a policy you correctly enforced. A guest whose stay you cannot locate in your system, because the platforms do not require you to be able to.
The practical move is to stop treating removal as the main event. File the flags that are legitimate, expect roughly a third of them to succeed, and put the rest of your energy into the two things that actually move the number: response quality and review volume.
Responses are written for the next reader
Here is the reframe that changes how most operators write. Your response is not addressed to the person who complained. They have already checked out, already vented, and will almost never read it. Your response is addressed to the person reading the review six weeks from now while deciding whether to book.
That person is running one test: does this property seem like it would handle a problem well? Everything about your response either passes that test or fails it.
Defensiveness fails it. Correcting the guest on a factual detail fails it, even when you are right, because the reader cannot verify who is right and defaults to assuming the business is spinning. Boilerplate fails it, because the reader has seen twenty of them and knows a template when they read one. What passes is specificity plus a named change. “You are right that the second-floor rooms catch noise from the bar next door on weekends. We have since added secondary glazing to 204 and 206 and we flag the remaining rooms at booking.” That is a response that makes the next reader more likely to book, not less.
The corollary: never respond within an hour. Responses written while you are still annoyed read like it. Sit on it overnight and write the version aimed at the reader.
Build the review you want people to leave
Most properties ask for reviews with a single generic prompt and get generic reviews back. A generic review is worth almost nothing to the fifth surface, because it contains no nouns for a model to extract.
Ask better and you get better. If your differentiator is the breakfast, your request should mention breakfast. If it is the rooftop, mention the rooftop. Not as a script to be parroted, but as a prompt that reminds the guest what they liked. People write about what they are reminded of, and what they write becomes the raw material for every summary an answer engine generates about you for the next two years.
Timing matters more than most operators believe. The request that lands during the departure drive converts better than the one that lands three days later, because the feeling is still present. The request that comes from the person who actually served them converts better than the one from a no-reply address, because it is a favor for a human rather than a task for a company.
And stop asking everyone. Asking a guest who had a bad stay to leave a review is a decision to manufacture a one-star. Ask the guests your front desk knows were happy. That is not gaming anything. That is choosing who to invite.
Where you send them matters as much as who you ask. Most properties point every request at Google, which is reasonable until you look at where your bookings actually come from. If a third of your business arrives through Booking.com, a property with four hundred Google reviews and thirty on Booking is strong in the place guests check you out and weak in the place they find you. Rotate the destination by channel. Guests who booked direct go to Google. Guests who booked through an OTA are already inside that platform’s review flow, so your effort is better spent making sure they complete it.
One practice to avoid entirely, no matter who suggests it. Do not filter your requests by asking guests how their stay was first and only routing the happy ones to a public platform. Review gating of that kind violates Google’s policies, has been the subject of regulatory action against businesses in several markets, and produces a profile that reads as suspiciously clean to anyone paying attention. Choosing which guests to ask, based on what your team observed, is judgement. Building a funnel that intercepts unhappy guests before they can post is something else.
The sixty-day repair sequence
If your rating is somewhere it should not be, here is the order of operations that works.
Days one through seven, run the Five-Surface Audit and write down the recurring nouns. Days eight through fourteen, fix the cheapest one. There is almost always a complaint category that costs under a thousand dollars to eliminate, and eliminating it stops the bleeding at the source. Days fifteen through thirty, respond to every unanswered review from the past year, oldest first, writing each one for the next reader. Days thirty-one through sixty, run a volume push: every departing guest who your team flags as happy gets a personal request naming something specific about their stay.
Sixty days of that moves a number that felt fixed. It also changes what the answer engines say about you, on a lag of roughly a quarter, because the pattern underneath the clause has changed.
Start with the nouns. Open your last twenty Google reviews this afternoon and write down every physical thing a guest mentioned. The list will be shorter than you expect, and one item on it is costing you more than the other four combined.