How is an app store rating calculated?

This is how nextspec scores apps. I didn’t test them. I read what critics and users already wrote, add it up with fixed rules, and show every source. This page is the whole method. The numbers below come from the same code that scores the pages, so they cannot drift apart.

1. What we read

• Critics: reviews from named publications, each with the date it was published.
• Stores: ratings from the App Store, Google Play, G2 and Capterra.
• Badges: four facts about what an app asks of you, each read from a public source (see section 7).

Every number has a source link and a date it was read.

2. Critic score

• Each scored review is turned into a score out of 100. A score out of 10 is multiplied by 10. A score out of 5 is multiplied by 20.
• Every publication counts once. If a publication has more than one review, only the one published most recently is used.
• An app needs at least 3 critic scores. With fewer, its critic cell reads “too few reviews”.
• Unscored reviews are quoted, not scored.

3. User score

Store ratings are weighted by how many reviews each one has. Then the average is pulled toward the category average. The fewer reviews an app has, the harder the pull. This stops 12 happy reviews from beating 80,000.

(n × average + 200 × C) ÷ (n + 200)

• n is the number of ratings.
• average is the mean star rating out of 5.
• C is the category mean: the average store rating of every app on that category’s pages, weighted by review count. It is recomputed whenever the data changes.
• The result is divided by 5 and multiplied by 100.

A worked example

This is an illustration. I use C = 4.2 here as a made-up category mean. Real pages use their own.

• 12 reviews at 5.0. (12 × 5.0 + 200 × 4.2) ÷ (12 + 200) = 4.25 stars. That is a user score of 85. The perfect 5.0 is pulled most of the way back to the category.
• 80,000 reviews at 4.6. (80,000 × 4.6 + 200 × 4.2) ÷ (80,000 + 200) = 4.60 stars. That is a user score of 92. With this many reviews, the pull barely moves it.

So the app with the lower star average ends up ahead. A big crowd is stronger evidence than a small one.

4. Consensus and ties

• The consensus score is the mean of the critic score and the user score. It sets the rank.
• An app with too few critic reviews is ranked on its user score alone. The table marks its consensus with *.
• An app with critic scores but no fresh store ratings is ranked on its critic score alone. The table marks its consensus with †.
• An app with neither score sits unranked at the bottom of the table.
• The rank uses the exact scores, before any rounding. The table shows them rounded to whole numbers, so two apps can show the same number and still be ranked apart. The consensus is worked out from the exact scores, so it can differ by a point from the average of the rounded numbers shown.
• A tie is decided only when the exact scores are equal. Then the app with more store ratings goes first. If they still tie, they go in alphabetical order.

5. Split verdict

When the exact critic score and the exact user score are 15 points or more apart, the app gets a “split verdict” flag, and the page explains the gap.

6. Freshness

A critic review published more than 365 days before the page’s read date is left out. It is the review’s publication date that counts, not the day I read it. A store rating counts as current on the day I read it. Each page shows “Scores read on” with the date.

7. The badges

Each app carries four facts. They are dated and sourced like the scores. They never change the rank.

• No login: from the store listing and the developer’s documentation.
• Pay once: from the store’s in-app purchase list. One-time purchase, not a subscription.
• No data shared: from the Apple privacy label and the Google Play Data safety section.
• Works offline: from the developer’s own statement. It is marked “developer says” because I did not test it.

If the store and the developer do not say, the badge reads ‘not stated’. I do not guess.

8. What we do not do

• We do not test apps. We read what others published.
• We take no paid placement. Nobody can pay to move up.
• We do not use rating stars markup on our pages. Our scores are our own sums, not a review of one product.

9. PDFlow disclosure

I make PDFlow. It never appears in a ranking; where it fits, it appears in a labelled box below the table.

10. Corrections

If a number is wrong, write to corrections@nextspec.tech. Send the page and the number. I fix it and add a line to that page’s changelog, with the date.