How we get the numbers
Where every revenue figure and every teardown on Appscope comes from, and how far to trust it.
Revenue labels
The founder showed the number themselves, for example a revenue dashboard on camera or in an interview.
A public statement or press report, such as an acquisition announcement. For the largest apps (Netflix, Spotify, ChatGPT and others) it is a company filing or an app-intelligence estimate from Sensor Tower or Appfigures; some of these cover the whole company or segment, not only the iPhone app.
No published figure, and the app is too far outside the apps with known revenue for an honest estimate (for example Facebook, Instagram and Microsoft Word, whose owners report them only combined with other products).
Our estimate. The tilde means ballpark. Hover a figure on the leaderboard to see the full range.
How estimates work
Every day at 06:00 UTC we read public App Store data for every app we track. No AI is involved in revenue.
- Ratings. Total ratings across 10 App Store countries (US, UK, Canada, Australia, Germany, France, Spain, Italy, Brazil, India), divided by the months since release. For the ~1,350 apps added from Apple's top-grossing charts, the 10-country total is refreshed weekly and the US count daily.
- Price. The subscription prices in the app's in-app purchase list, turned into a monthly price (weekly × 4.33, yearly ÷ 12).
- Fit. Ratings per month × monthly price, compared against apps whose revenue is public. The relationship is fitted again every day as new known figures come in.
- Range. Each estimate comes with a low and high figure. The mid point is what you see.
How accurate it is
We test by hiding one app's real revenue, estimating it from the others, and repeating for every app. These figures were measured on October 16, 2026.
Where it struggles: very large apps (it tends to estimate too low), apps that earn mostly outside the App Store, and apps with very few ratings.
How teardowns work
- Reviews. Apple's public review feed: up to the 500 most recent reviews in each of the US, UK, Canada and Australia.
- Grouping. An AI model (Claude) groups the reviews into what users hate, what they ask to be added and what they love, and suggests fixes with build prompts and a tech stack to build your version with. The stack is our recommendation, not what the original app uses.
- Checks. We count only reviews the model cites that exist and have the right star rating. A quote is shown only if it appears word for word in the review it cites; otherwise it's dropped.
- Freshness. A teardown is rebuilt at most once every 30 days, when someone opens it.
Percentages are shares of the reviews we read, not of all users. Written reviews lean negative, because people with a problem are more likely to write one. Fixes and prompts are suggestions, not facts. Early teardowns were grouped with keyword rules instead of AI; each teardown says which method was used.