How we score apps
We read what real users actually wrote, balance it so the angriest voices don't drown out the truth, and cite every claim back to the review it came from. Here is exactly how.
The data
Every verdict is built from Apple's own public customer reviews, nothing bought or scraped from behind a paywall. For each app we pull the most recent reviews across four English-language storefronts (US, UK, Canada, Australia), remove duplicates, and translate any that are not in English. That is roughly 900 to 1,000 real reviews per app. We use several regions on purpose: a complaint that shows up in the US, the UK and Australia is far stronger evidence than one market having a bad week.
Why 120 reviews
From those ~900, we hand the model a balanced sample of about 120. Not the newest 120 (those skew angry), and not a random 120 either, we sample evenly across negative, positive and neutral reviews so both sides are represented. 120 is enough to capture every recurring theme, the top complaints and the real praise all surface within the first hundred or so, past that it is mostly repetition, while staying fast and cheap to generate. The full ~900 still drive the star distribution and the headline signals; the 120 are the ones the verdict actually quotes and links to.
How we keep the verdict from being skewed
People mostly review apps when they are angry, and they review the latest version. Left unchecked, that makes every app look like it is on fire. Four things guard against it:
- Balanced sampling. Happy and unhappy reviewers get equal weight, not just the loudest.
- Recent vs lifetime. We always reconcile the recent sample against the app's lifetime store rating and describe the trajectory ("well-loved for years, but the recent update angered users over X") rather than treating this month's mood as the whole story.
- Multi-region. A gripe has to recur across markets to count as consensus.
- Free-vs-paid awareness. When one app has a real free tier and a rival is paid-only, we flag that the free app's negativity is often free-tier friction, not worse quality.
Every claim is cited
There are no unsourced opinions. Every pro, con, theme and comparison point links to the specific reviews it came from, click any [ref] and read the exact quote. If the evidence is thin or one-sided, we say so instead of pretending. We never invent features, prices or facts that are not in the reviews.
The score
The 0 to 100 is a synthesis of user sentiment, balanced across recent and lifetime signals. It is a considered summary, not a precise measurement. Read it as "how happy are people, really," not as a lab number.
What we cannot tell you
The honest limits, because a methodology that hides them is not one:
- Apple only exposes about the 500 most-recent reviews per storefront, so we see a recency window, not the full history.
- Reviews capture what people complain about (speed, bugs, price) better than why they choose an app. Nobody writes a five-star review praising a VPN's no-logs policy, even when that is the real reason they use it.
- We cannot always tell a genuine free tier from a paid-only subscription from Apple's data alone. The reviews usually reveal it, and the verdict reflects that.
- Pages are generated automatically and grounded in real reviews, but not hand-audited one by one.
Freshness
Each page shows when it was last built and refreshes on a schedule, so recent complaints stay current. We also record each app's key signals over time, which lets us show when sentiment is genuinely shifting, not just where it stands today.