Where the reviews come from
We use Apple’s public feed of App Store customer reviews. Anyone can read it, and it needs no login.
For each app, we ask for the newest reviews in one country’s store. Apple returns at most 500 of them, in 10 pages of 50. Some apps have fewer. We need at least 5 reviews to make a teardown.
We keep the summary and a few short quotes. We don’t keep a copy of every review.
How we read them
Synthiya sends the reviews to an AI model made by Anthropic. The model lists what users love, what they hate and what they ask for. For each point, it must give a short quote, word for word, from a real review. It also counts how many reviews bring up each point.
We tell the model to use only the reviews it was given. It must never make up features, numbers or quotes.
To keep things fast, we send up to about 70,000 characters of review text. We cut very long reviews to their first 350 characters.
If the AI is not available, we use a simpler method. It sorts reviews by their stars and shows real review titles and quotes.
What the numbers mean
- Satisfaction score
- A score from 0 to 100 for how happy recent reviewers are. 0 means almost everyone is unhappy. 100 means almost everyone is delighted. The AI sets it after it reads the reviews. With the simpler method, we work it out from the average stars, where 1 star is 0 and 5 stars is 100.
- Mentions
- About how many of the reviews we read bring up a point. The AI counts them, so treat them as close estimates.
- Rating mix and recent average
- We count these straight from the stars on the reviews we read. No AI is involved. The same goes for the share of 1 and 2 star reviews.
- Rating by version
- The average stars for each recent app version that has at least 3 reviews.
- App Store rating
- The rating Apple shows for the app. It covers far more ratings than the reviews we read, so it can differ a lot from our recent numbers.
- Quick answers
- We build these from the teardown data with fixed rules. They add no new claims.
How often we update
We save each teardown, so pages load fast. When a teardown is more than 3 days old and someone asks for it again, we fetch the newest reviews and build it again.
Each page shows the date it was last updated, and so does its Markdown version.
Limits and biases
A teardown is a quick, honest read of recent reviews. It is not a survey of all users. Keep these limits in mind:
- Recent reviews only
- A teardown shows how people feel now. One bad update can pull it down for a while.
- Loud voices
- People who write reviews are often very happy or very upset. Calm, content users write less.
- One country
- We read one country’s store at a time. Users in other countries may feel different.
- Review prompts
- Many apps ask happy users to rate them. This can lift the stars.
- Fake reviews
- We can’t tell if a review is fake or paid for. We rely on Apple’s own checks.
- AI mistakes
- The AI can merge two points or miss a small one. That’s why every point shows a real quote you can check.
How to cite a teardown
Link to the teardown page and give the date it was last updated. Each page has a “Cite this teardown” box you can copy.
AI tools can read a plain Markdown version of any teardown. Just add .md to the end of its link. You can also read short summaries of the top teardowns in our llms.txt file.
Spot a mistake?
Email us at hello@synthiya.ai. If you build the app and a teardown looks unfair, tell us. We’ll check it and refresh it.