Apple processed 1.3 billion ratings and reviews across the App Store last year and blocked around 195 million of them as fraudulent. That is close to one submission in seven, removed before any reader saw it. The figure sits inside Apple’s annual fraud prevention report alongside 1.1 billion rejected account creations and 193,000 terminated developer accounts. It is one of the more detailed public disclosures available on the volume of fraudulent activity that moves through app ratings.
A Star Rating Is a Statistic With No Methodology
Any number published as evidence is expected to declare its sample. Who was asked, how many of them there were, when the question was put, and what share of those contacted responded. A star rating declares none of it.
It arrives as a single figure to one decimal place, with no indication of sample size, collection method or response rate. The sample is whoever felt strongly enough to submit a review, which skews toward strongly positive and strongly negative responses with less representation in between. Gaming and casino apps sit among the categories most affected by this: both attract users who engage intensely and leave reviews at the extremes, and both have documented histories of coordinated fake review activity across app stores.
Developers have started raising this publicly. Jeff Johnson of Lapcat Software argued at the end of July that the App Store inherited its rating system from a platform that sold ninety-nine cent songs and never adapted it for software, where a one-star review is as likely to be a feature request as an assessment of whether the product functions.
The Device Variable Goes Unrecorded in Blended Ratings
Mobile passed half of global web traffic in April at 52.8% against desktop’s 45.61%, and the device breakdown behind those figures puts desktop conversion between 3.5% and 4.0% while phones manage 1.8% to 2.5%. The same brand, the same catalogue and the same prices produce different outcomes depending on the device.
Bounce rates follow a similar pattern, running at 54.3% on phones against 42.8% on computers. Cart abandonment reaches between 79% and 85% on mobile where desktop sits closer to 68%. In iGaming specifically, a failed deposit flow on mobile is a direct lost transaction that registers in revenue data the same session. A blended rating averages those two populations into one figure and presents it to a reader who belongs to exactly one of them.
Splitting by Device Costs Nothing to Publish
This requires no new data collection. Every platform records which device each session came from, and the split is available in existing analytics whether it is published or not.
Some sectors do publish it. Retail separated app ratings from web ratings years ago, and streaming services have reported device-level performance for longer than that.
Gambling and gaming platforms have separated device performance in their rankings because deposit and withdrawal flows are where the device gap is most significant and where a failed transaction registers in the data the same day. That is the reasoning behind talkSPORT’s mobile rankings, which sort the market into different casinos made for different devices including mobile rather than publishing a single table. The criteria shift with the form factor: touch target sizing, one-handed reach, the number of taps between opening an app and completing a payment, and whether the native app and the mobile site perform equivalently.
Published Criteria Beat a Bigger Sample
The instinct when a rating looks unreliable is to gather more responses. Volume reduces random error. It does not address a sample that was skewed before collection started.
Apple’s figures illustrate this. Filtering out one submission in seven still leaves a rating drawn entirely from people who chose to respond. Survey methodology treats an unreported response rate as a limitation, since the people who declined to respond tend to differ from those who did. Consumer ratings publish no response rate at all.
Disclosure addresses the problem. A rating that states its criteria, its weighting and its device split can be evaluated and challenged. Casino review publications that list their assessment methodology, test withdrawal speeds independently and separate mobile from desktop scores are applying a standard the blended App Store rating does not meet. Publications with structured fact-checking processes apply source and methodology requirements to the figures they publish. The star rating has not been subject to equivalent scrutiny.
Every Reader Belongs to One Population
A reader checking a rating is using a specific device on a specific connection. They belong to one of the two groups that were averaged together to produce the figure, and the figure does not indicate which.
Publishing the device split requires one additional column. Publishing the methodology requires a paragraph. Both make the rating more useful and more verifiable than the blended figure it would otherwise replace.
