How to analyze App Store and Google Play reviews

1. Write the question before you read. "Why did the rating drop after 4.2?", "what should we build next?" and "where are a rival's users unhappy?" need different windows and different apps. The question picks the dates and the stores; the reviews pick the answer.
2. Collect from every store and country that matters. The App Store shows each country its own reviews, and its public pages reach about the newest 500 per country. Google Play holds far more (about 30,000 for a large app) in one pool with a language filter. A US only read of a global app can miss the country where the drop happened.
3. Read enough for the shares to hold. A share from 50 reviews is a guess. The table below shows how the likely range narrows as the sample grows. Spread the sample over the window too: the newest 300 reviews of a busy app can all come from one bad weekend.
4. Use one fixed set of themes, and give each review one main theme. Crashes, login, pricing, ads, missing features, design. Decide the list once, then sort every review into it, with an other row for the rest. Count the main theme for shares; count every theme a review mentions separately, and never mix the two numbers in one sentence.
5. Keep an unsure pile, and show it. "Doesn't work" is a complaint with no theme. Forcing it into crashes inflates crashes. A visible unsure row tells the reader how much of the count is judgment.
6. Read stars apart from words, then split by version and set it against a rival. Five star reviews carry complaints and one star reviews carry praise. A theme that jumps in one version is a regression; a theme that is high for you and low for a rival is the gap to close.

One row per review, and these columns:

• Store, country, date and version, so you can filter by release and market later.
• Stars, kept apart from what the words say.
• Kind: complaint, praise, request, or none.
• Main theme from your fixed list, plus a second column for any other theme it mentions.
• Leaving: yes when the reviewer says they quit, uninstalled or cancelled.
• Sure: no when you had to guess. Those rows are your unsure pile.
• A short quote of the sentence that decided the theme, so anyone can check your call.

A COUNTIF per theme over the sure complaints gives the shares. It works well up to a few hundred reviews. Past that, the sorting takes days, and two people sorting the same review disagree often enough that the count drifts.

The report's title asks about price. The reviews answered in a different order: pricing and fees is 14% of sorted complaints, behind missing feature (23%) and content quality (22%). That gap is the reason to count before you decide what the story is.

• Sorting by most helpful. The store's helpful order favors old, long reviews. Read by date, or spread over the window.
• Reading only the one star reviews. They tell you what hurts, and nothing about what keeps the people who stay. Praise is half of the roadmap.
• Reading one country. A login bug with one bank, a price change in one currency or a translation can live in a single storefront.
• Counting a wave as a trend. A burst of reviews about one event (a price change, a news story, an outage) can dominate a month. Check the volume against a normal week first; the review bomb guide shows the rule.
• Quoting without the share. One vivid review proves that a complaint exists. Only the share says whether it is common.