Google has reported removing or blocking policy-violating reviews at a scale of hundreds of millions per year, and Trustpilot publishes an annual transparency report accounting for the fake reviews it took down across its platform. Those are the two numbers most articles on this subject lead with, and both are almost useless to a business owner on their own.
They describe platform enforcement. What you need to know is different: how much of the review text buyers are reading about your category is fabricated, what it does to purchase decisions, what the legal exposure now is for anyone tempted to join in, and how to tell whether your own profile has been hit. That is what the eight numbers below are for.
A note on sourcing before we start. Figures in this field go stale faster than almost any other marketing statistic, because platforms change their detection and then report new numbers each year. Every figure here is either structural, meaning it does not change, or attributed to a source you can check with a date attached. Anything else is not worth repeating.
Number One: The Platform Removal Figures, and Why They Mislead
Large platforms publish removal counts, and they are real. They are also the most misread numbers in the category.
A removal count tells you about enforcement activity, not about prevalence. A platform that reports removing more fake reviews this year than last may have a worse problem or better detection, and the number alone cannot distinguish between those. Vendors quote these figures both ways depending on what they are selling, which should tell you how much interpretive weight they carry.
What the removal figures do establish is scale. The volume being removed is large enough that fake review production is an industry rather than a fringe behaviour, with brokers, account farms and pricing. Amazon has pursued fake review brokers through the courts, which is the kind of action a company takes against an organised supply chain rather than against scattered individuals.
Use these numbers as context, never as a statistic about your own exposure. For that you need to look at your own profile, which the later sections cover.
One rule worth adopting for any figure in this field. Before citing a removal or prevalence number, check that you can name the organization that published it and the year. If either is missing, leave it out. The same handful of percentages circulate through hundreds of articles in this category with no source chain behind them, and a prospect who checks one and finds it stale will discount everything else you said.
Number Two and Three: What Fake Reviews Do to Buyers
Here the research is more useful, because the effect sizes are stable across studies even when the specific percentages are not.
Two findings hold consistently. First, buyers are poor at identifying fake reviews when asked directly, performing only modestly better than chance in controlled tests. Second, buyers who come to suspect a profile is manipulated discount the entire profile rather than the suspect reviews, including the genuine ones.
That second effect is the commercially important one and it rarely gets discussed. The damage from a fake review problem is not that some reviews are false. It is that suspicion is contagious across a profile. A business with forty genuine reviews and ten obviously fabricated ones converts worse than the same business with forty reviews and none, because the ten poison the reading of the forty.
This is also why buying reviews has a worse expected value than the price suggests. The downside is not a fine. The downside is that your real customers’ reviews stop working.
Number Four: 82% Research Before Buying
Our own figures put the share of consumers who research a business online before purchasing at 82%, alongside 90% who report being more likely to buy after seeing positive coverage, a 34% lift in conversion rate where positive news is present, and 62% more revenue for companies with positive news.
The 82% is the one that connects to this subject. It establishes that the review-reading step is not optional for four buyers in five, which means whatever is on your profile is being read by almost everyone who considers you. A manipulated profile is not a cosmetic problem sitting in a corner of the internet. It is in the path of nearly every purchase decision.
The corollary is more encouraging. Because the research step is near-universal, improvements to it reach almost all of your potential buyers rather than a segment. Few marketing investments have that property.
There is a sequencing lesson in those four figures too. Coverage and reviews are not competing for the same budget, they are consecutive steps in the same decision. Coverage gets a buyer to the research stage. The review profile decides what happens when they arrive. A business spending on visibility while sitting on a suspect profile is paying to deliver qualified buyers to the page that talks them out of it, which is the most expensive possible way to run a reputation problem.
Number Five: The Velocity Tell
The most reliable indicator of review manipulation is not the content of individual reviews. It is the shape of the arrival curve, and you can read it on any public profile in about three minutes.
We call it the Velocity Tell. A genuine review profile accumulates at a rate roughly proportional to transaction volume, which for most businesses means a steady trickle with mild seasonal variation. Manipulated profiles produce a different signature: flat for months, then a cluster of positive reviews arriving within days of each other, then flat again.
Three secondary signals usually accompany it. The clustered reviews tend to be short, because fabricated text is expensive to write well. They tend to use the full business name in a way real customers do not, because the person writing them was given a brief. And the reviewer accounts tend to have thin histories, often a single review or a handful across unrelated categories in different cities.
Run this on your own profile first, then on your three closest competitors. Most owners find the exercise clarifying in an uncomfortable way, because the competitor whose rating they could never explain usually has a visible cluster.
Two cautions on reading the tell. A genuine burst does happen: a business that runs its first ever review campaign produces a cluster that looks identical to a purchased one from the outside, which is one reason steady collection beats campaigns. And a negative attack has its own version of the signature, usually a handful of one-star reviews within a day or two, often with no text at all, from accounts with no other activity. That pattern is the one worth reporting, because it is the easiest for a platform to confirm.
Number Six and Seven: The Legal Position Changed
This is the part that genuinely changed in the last two years and that a lot of businesses have not registered.
In the United States, the Federal Trade Commission finalised a rule on consumer reviews and testimonials in 2024, effective later that year, which among other things addresses buying and selling fake reviews, review suppression and insider reviews that are not disclosed. The significant feature is that it opens the door to civil penalties, which changes the calculation from a reputational risk into a financial one.
In the United Kingdom, the Digital Markets, Competition and Consumers Act 2024 made commissioning or publishing fake reviews a banned practice, with the relevant consumer protection provisions commencing in 2025 and enforcement powers sitting with the Competition and Markets Authority.
Two practical consequences follow. The first is obvious: do not buy reviews, and if an agency offers them as part of a reputation package, that is a disqualifying answer rather than a negotiating point. The second is less obvious and catches honest businesses. Incentivising only happy customers to review, or filtering review requests based on expected sentiment, sits in the same regulatory territory as suppression. A review request should go to everyone or to a genuinely random sample, and it should not be conditional on how the interaction went.
What AI Assistants Do With a Manipulated Profile
A new consequence has arrived in the last two years and it is worse than the conversion effect.
When a buyer asks an AI assistant for a recommendation in your category, the model is reading and summarizing review text rather than reporting a star average. That changes which manipulation signals matter. A cluster of short, generic five-star reviews gives a model almost no usable language, so it describes such a business vaguely or skips it in favour of one it can say something specific about.
The result is counterintuitive and worth sitting with. A business with a 4.9 built from thin fabricated text can lose a recommendation to a business with a 4.3 built from detailed genuine reviews, because the second one gave the model something to work with. Review manipulation optimises for a metric that the fastest-growing discovery channel does not read.
It also means the defence and the growth strategy are the same action. Asking customers what they were trying to do and whether it worked produces text that is useful to models, resistant to being drowned out, and impossible to fake cheaply at volume. A review that says the installer arrived early and rerouted a vent to avoid a load-bearing beam cannot be produced by someone working from a brief.
Number Eight: Your Own Review Rate
The last number is the only one that is actually about your business, and almost nobody has calculated it.
Take completed transactions over the last ninety days, take reviews received in the same window, divide. For most businesses the answer lands under 2%, which means the profile buyers are judging you on was written by a tiny, self-selected minority of your customers. That is the vulnerability that makes fake reviews work, both for the people writing them about themselves and for anyone writing them about you.
A profile built from 2% of customers can be moved by a handful of fabricated entries. A profile built from 10% cannot, because the genuine volume swamps anything an attacker can afford to produce. Raising that rate is the only durable defence against review manipulation, and it is entirely within your control.
The mechanics are unglamorous. Ask every customer, at the moment they got the result rather than at the moment you invoiced them, with a link that takes two taps. Build it into the delivery process so it happens without anyone remembering to do it. Then watch the rate monthly as a number on a dashboard, because it will drift down the instant nobody is watching.
Everything else in this article is context. Platform enforcement will do what it does, the law has moved in your favour, and your competitors’ clusters will eventually get caught. What decides whether any of it reaches your revenue is how much genuine review volume you have when it happens.