Almost every Trustpilot statistic you have read in a marketing article is either out of date, stripped of the context that made it true, or both. The platform publishes real figures and then watches them get recycled for three years by people who never checked whether they still hold.
So this is not a stat dump. It is the nine numbers that actually change decisions, which of them are stable enough to build on, and which ones you should refuse to quote without a fresh source. Some of them are structural facts about how the platform works, which matter more than any headline figure, because they explain why two businesses with the same average rating get completely different results from it.
The Numbers That Do Not Move
Start with the facts that will still be true next year, because these are the ones worth memorizing.
Trustpilot was founded in 2007 in Denmark by Peter Holten Mühlmann and is headquartered in Copenhagen. It listed on the London Stock Exchange in March 2021, which matters for a reason most people miss: a listed company publishes transparency reporting on a schedule and can be held to it, which makes its own numbers more citable than those of a private review platform.
The TrustScore runs from 1.0 to 5.0. The labels attached to score bands run from Bad through Poor, Average, Great and Excellent, and those labels are what a buyer actually reads. The practical consequence is that the distance between 3.9 and 4.0 is worth more than the distance between 4.5 and 4.6, because one crosses a labelling threshold and the other does not.
Trustpilot is an open platform. Anyone with an email address can review any business without proving they bought anything. That one design decision explains most of the complaints about the platform, most of its fake review problem, and most of its usefulness, since it also means a business cannot quietly suppress reviews by controlling who gets invited.
The Recency Cliff
Here is the mechanic that costs businesses the most money and gets written about the least.
The TrustScore is a weighted average, not a flat mean, and recent reviews carry more weight than old ones. Run that forward and a conclusion falls out that most owners find counterintuitive: a profile with a great score and no recent activity is on a slow decline. Nothing bad has to happen. The good reviews simply age out of their own influence.
We call this the Recency Cliff, and it produces a specific failure pattern. A business runs a review push, collects two hundred positive reviews in six weeks, hits 4.7, declares victory and stops. Nine months later the score reads 4.3 and the owner is looking for the negative reviews that caused it. There are none. The collection just stopped.
The operational number that follows from this is the only review metric worth putting on a dashboard: reviews collected per month, held steady. Not total reviews. Not average rating. The rate.
The Statistics You Should Refuse to Quote Unprompted
Trustpilot publishes an annual transparency report covering review volume and the fake reviews it removed. Those figures are real and worth reading. They are also the most abused numbers in reputation marketing, because they get quoted years after publication with no date attached, and the volumes change substantially year to year as the platform grows and as its detection improves.
The rule we apply internally: any platform volume or removal figure gets quoted only from the current report, and the sentence carries the report year in it. If you cannot name the year, do not use the number. This sounds pedantic until a prospect checks one of your figures, finds it three years stale, and quietly discounts everything else you told them.
The same caution applies to the broad consumer behaviour statistics that circulate in this space. Figures about what percentage of buyers read reviews are usually drawn from small commissioned surveys with a vendor’s name on them, and they drift upward every year because each new survey is designed by someone who wants a bigger number than the last one.
There is a simple test for whether a review statistic is worth citing. Can you name the organization that collected it, the year, and roughly how many people they asked? If any of those three is missing, the number is decoration. Most of the figures in circulation fail all three, which is why you see the same handful of percentages in hundreds of articles with no source chain behind any of them.
This matters beyond pedantry because buyers and funders increasingly check. A prospect who finds one stale figure in your deck starts auditing the rest, and the cost of that is much higher than the benefit of having had an impressive number on slide four.
What Our Own Data Says About Review Weight
We have published more than 2,000 articles and served over 80 clients, and the pattern across that work is consistent enough to state plainly.
Our own figures put the lift from positive news coverage at 34% on conversion rate and 62% on revenue for companies with positive news present, with 90% of consumers reporting they are more likely to purchase after seeing positive coverage, and 82% researching a business online before buying. That last number is the one that connects to Trustpilot. The research step is where reviews do their work, and it happens for four buyers in five.
What we see in practice is that reviews and earned coverage are not competing signals, they are sequential ones. Coverage gets a buyer to the research step. Reviews decide what happens there. A company with strong press and a 3.4 TrustScore is spending money to deliver qualified buyers to a page that talks them out of it.
The reverse case is less painful but still costly. A company with an excellent profile and no coverage has a strong close rate on a small number of buyers who found it some other way. Reviews convert demand. They do not create it. Teams that treat a review push as a growth strategy tend to discover this about four months in, when the rate of new reviews flattens out because the number of transactions did not change.
Our catalog data gives a rough sense of what the coverage half of that costs. The median placement sells at $1,500, the largest single band runs from $1,001 to $2,500 with 523 outlets in it, and 174 publications carry a Domain Authority above 80. A buyer comparing that against the cost of a review collection tool is comparing two line items that do different jobs, and the honest answer is that a business with neither should usually fix reviews first, because it is cheaper and because coverage amplifies whatever the research step currently does.
Volume, Recency and Response
Three numbers describe a healthy profile, and you can read all three in under a minute.
Volume: total reviews, read as a sample size rather than an achievement. Under 25 the score is noise. Past 100 it stabilizes. Past 500 it becomes genuinely hard to move, which is protection once you are high and a trap once you are low.
The asymmetry inside that last point is worth sitting with. A business at 4.6 with 900 reviews can absorb a bad month without a buyer noticing. A business at 3.4 with 900 reviews cannot recover inside a year, because the weight of history is working against every new five star review. Which means the moment to build volume is when things are going well, and the moment most businesses actually start is after a complaint. That inversion is the whole reason reputation work has to be scheduled rather than triggered.
Recency: reviews in the last 90 days as a share of total. Under about 5% the Recency Cliff is already working against you.
Response: the share of one and two star reviews that have a public reply. This is the number buyers read most closely and businesses track least. A negative review with a specific, unapologetic, factual reply reads better than no negative reviews at all, because it proves a human being is on the other end.
What AI Assistants Do With Your Score
This is the part that has changed since 2024 and the reason review strategy stopped being a conversion tactic.
When a buyer asks an AI assistant which supplier in your category to use, the model is reading review aggregates, roundup articles and editorial coverage, then summarizing. It does not reproduce your score. It characterizes it. The difference between a model saying a company has strong reviews and a model saying reviews are mixed is the difference between making a shortlist and not existing, and it turns on structural facts about your profile rather than on a decimal place.
Three things push a profile toward the favourable characterization: enough volume to be worth mentioning, recent activity so the data looks current, and visible responses so the business reads as accountable. Those are the same three numbers above, which is convenient, and they are all within your control, which is the point.
One extra factor applies only to the AI layer. Models read text, so the wording of your reviews matters in a way it never did for a star average. A hundred reviews saying great service give a model almost nothing to work with. Twenty reviews that name the specific thing you did well, by product and by situation, give it language to reuse when a buyer asks a specific question. The practical move is to change what you ask for: a review request that says tell us what you were trying to do and whether it worked produces far more usable text than one that says rate your experience.
The One Statistic Worth Tracking Yourself
Ignore the industry figures for a quarter and measure one thing: how many of your completed transactions produce a review.
Most businesses have never calculated this. The ones that do are routinely shocked, because the honest answer is often under 2%, which means the review profile buyers judge them on is built from a tiny and self selected slice of their customers. Push that rate to 10% through a request built into delivery, and the profile stops being a complaints board and starts being a representative sample. Nothing else you can do to a Trustpilot profile matters as much.
Work out the number this week. Pull total completed orders or closed jobs for the last ninety days from whatever system holds them, pull reviews received in the same window, divide. The arithmetic takes ten minutes and the result will tell you more about your reputation position than any benchmark figure you could find.
Then fix the request. The single biggest change you can make is moving it from a follow up email sent days later to a prompt that arrives at the moment the customer is most satisfied, which is almost never the moment the invoice clears. For a service business that is the handover call. For a product business it is three days after delivery, not three hours. For anything with onboarding it is the first time the customer gets a result, and your own product data knows when that is.
Trustpilot’s own mechanics reward this. Steady collection beats bursts, recent reviews carry more weight, and a profile with a consistent trickle reads to both buyers and models as a business that is currently operating well rather than one that ran a campaign in 2024.
That rate is your number. Every other statistic in this article is context for it.