1,784 Companies Would Jump From Below 3 Stars to Above 4 — Just by Hiding One-Star Reviews

In March 2026 the UK Competition and Markets Authority opened consumer-protection investigations into five businesses over online reviews, with case updates expected in September 2026. Three of them go to the heart of how a star rating is built: the Autotrader and Feefo case concerns one-star reviews that were allegedly moderated and never published, and so never counted towards the star rating; the Just Eat case concerns ratings that may have been inflated; and the Pasta Evangelists case concerns discounts offered in exchange for five-star reviews without disclosure.

The CMA investigations are ongoing. Nothing has been decided and no company has been found to have broken the law. This article does not comment on any of those cases or on the conduct of any named business.

Our interest is in the question those cases raise rather than the cases themselves, and it is a question nobody has answered with data: if one-star reviews quietly stop being published, how much does a rating actually move? We can measure that, because we hold the full star distribution for tens of thousands of companies. So we recomputed every rating with the one-star reviews removed and compared it with the real one.

Removing one-star reviews adds 0.57 stars

The size of the shift matters because of where it lands. Four stars is the threshold consumers most commonly filter on, and 5,630 companies — 23.1% of the set — would cross from below four stars to at or above four stars once their one-star reviews stopped counting. More striking still, 1,784 companies would move from below three stars to above four stars: from a rating a shopper would avoid to one they would trust, with no change in what customers actually experienced.

The effect is uneven by industry

The other way a rating gets distorted: the missing middle

Suppressing the bottom of the distribution is only one mechanism. The other is the incentivised-review pattern the Pasta Evangelists case describes, and it leaves a different fingerprint. A genuine distribution has a middle: some customers are quietly satisfied, some are mildly disappointed, and they show up as twos and threes. When reviews are solicited in exchange for something, those middle bars thin out and the shape collapses into a spike at five.

6,312 companies — 25.9% of the set — show what we call a missing middle: at least 85% five-star reviews with 2% or less in the two- and three-star band. That is a measurable shape, not a verdict.

A missing middle is a pattern, not an accusation and not evidence of wrongdoing. A company with genuinely delighted customers can produce exactly that shape honestly. The point is that the shape regulators look for is measurable — and that a star rating alone cannot tell the two cases apart.

Both mechanisms move the same number in the same direction

Suppressing one-star reviews pushes the average up. Soliciting five-star reviews pushes the average up. A consumer sees only the result: one number, with no way to tell which distribution produced it. This is the argument we have made for a long time — a star average compresses away the information that actually predicts whether customers leave, which is what the low end and the middle of the distribution are telling you. The current regulatory moment is a good occasion to show the size of that distortion rather than assert it: 0.57 stars, 5,630 companies over the four-star line, and a quarter of the market already shaped like a spike.

Methodology: the set is 24,411 companies that each have at least fifty reviews. Ratings were recomputed arithmetically from the stored star distribution for each company rather than re-scraped, by removing the one-star bucket and re-averaging the remainder. The comparison is arithmetic and describes what the published number would look like under that condition; it is not a claim about any specific company's conduct, and no company in the set is alleged to have suppressed or solicited reviews.