What users hate about any app — one API call, ranked pain points, verbatim evidence
"Read your competitors' one-star reviews" is the most repeated product advice on the internet. It's also terrible advice, mechanically.
When I was deciding whether to build yet another habit tracker, I actually did it: four competitor apps, a week of tab-switching between store pages, a spreadsheet. The findings were gold — 28% of Loop Habit Tracker's complaints trace to a single Android OEM bug (battery whitelisting silently killing reminders), and the most-shared pain across every app (20–24%) was sync/reminder reliability, yet not one competitor marketed "your streak will not disappear."
The process was miserable. The stores don't give you reviews over a clean API (Apple killed the official reviews RSS in October — we verified it returns empty entries). You scroll, you copy-paste, you deduplicate, you eyeball for themes. Then you do it again next month, because reviews drift.
I built the tool I wanted instead. It's called App Pain API.
What it is
App Pain API fetches recent app-store reviews for any app, keeps the unhappy ones, clusters them with an LLM, and returns ranked pain points over one REST call — or one MCP tool call, straight from your AI client.
curl https://hawkmill.com/v1/apps/android/com.zhiliaoapp.musically/painpoints
No key needed for the first look. This is the real response shape (TikTok, actual values from our docs):
{
"painPoints": [
{
"title": "Intrusive Ads and Ad Frequency",
"summary": "Users are frustrated by the high frequency of ads, especially mid-video and unskippable formats.",
"sharePercent": 16,
"severity": 4,
"evidence": ["ads every two videos", "so many ads in every single video I watch it's so annoying"],
"nReviews": 16
},
{
"title": "Shadowbanning and Algorithm Suppression",
"sharePercent": 12,
"severity": 4,
"evidence": ["my views keep dropping and I'm shadow banned"],
"nReviews": 12
}
],
"overview": "Monetization pressure (ads) and opaque algorithm changes dominate complaints.",
"analyzedReviews": 300,
"negativeReviews": 187
}
Three things matter in that response:
- sharePercent is not the model's guess. It's computed deterministically on the server (largest-remainder rounding over actual cluster membership). The LLM never gets the final word on numbers.
- Every pain point ships verbatim quotes. Treat the titles as an index into the evidence; verify before you act on anything.
- Unified across stores. iOS, Android, macOS, WordPress plugins and Chrome extensions all return the same schema, so you can compare a plugin against its mobile sibling.
What it covers
- Five platforms, fully: iOS App Store, Google Play, Mac App Store, WordPress.org plugins, Chrome Web Store. (Steam works for metadata, ratings and search; review mining is coming soon.)
- Reviews: 50/page up to 10 pages, unified fields,
sort=recent|helpful. - Metadata: ratings, histograms, icons, developer, price.
- Versions: per-version review stats — see the rating crash after a bad release. Spotify's 9.1.94 averaged 3.41 across 437 reviews right after shipping, versus 4.05 on the previous build. That's a regression story you can watch from the outside.
- Search: one query across stores.
- Webhooks: subscribe to an app and get digests of new 1–2★ reviews and new/rising pain points pushed to you (Slack and Discord formatted natively). Paid feature, from the $9 tier.
- Over a hundred popular apps pre-analyzed, free to browse at hawkmill.com/pains — no account, no key.
What you can do with it
1. Validate a product idea before building it. That's the habit-tracker story above: one call per competitor instead of a week of scrolling. If the #1 pain in a category is something nobody advertises against, that's a positioning gift. (It's also how you find anti-opportunities — some categories are miserable for reasons no indie can fix.)
2. Run a competitor teardown in an afternoon. Search the space, pull pain points and reviews for each rival, read the quotes. Why do people churn from the market leader? What does the $9.99 app do that the free one won't? The evidence quotes keep you honest — you're reading users, not a model's summary of users.
3. Watch rivals' releases without using their app. The versions endpoint gives you per-version review stats, so a competitor's bad update shows up as an average-rating cliff with timestamps — the Spotify numbers above, for example. Add a webhook and the new 1★ wave arrives in your Slack instead of you remembering to check.
4. Let your AI agent do all of it. App Pain API is a native MCP server: paste https://hawkmill.com/mcp into Claude, ChatGPT or Cursor — keyless, 3 fresh analyses a day, cached apps unlimited — and ask "what do users hate about app X, and what's the gap vs app Y?" The tools are read-only, and review text is treated as data, never instructions. Every tool result is structured JSON your agent can cite.
5. Mine your own app's reviews. Sort by severity, look at what users beg for in the quotes, and you have next sprint's backlog argued for you — with receipts.
How to start
| Path | Setup | What you get |
|---|---|---|
| Browse hawkmill.com/pains | none | real output in 10 seconds, 100+ apps |
Paste https://hawkmill.com/mcp into your AI client | none | agent-driven teardowns |
curl the REST API | none | 3 fresh analyses a day, cached apps unlimited |
| Free key (email verification, 2 min) | 2 min | 10 fresh analyses a month (+10 signup bonus; referrals add more) + 2 months of pain-point history |
| Starter, $9/mo | checkout | 1,500 calls, 50 fresh analyses, 2-app competitor watch with weekly digests |
| Pro, $29/mo | checkout | 10,000 calls, unlimited fresh analyses, 15-app watch on a daily cadence, full history archive |
Cached apps answer in ~5ms; a fresh analysis of an uncached app takes 30–120 seconds (we fetch, filter and cluster on the fly), then it's cached for 7 days. Metadata caches 24h, reviews 6h. Every response carries an X-Cache header so you know which one you got.
Honest limitations
- iOS depth is shallow, on purpose. Apple aggressively rate-limits datacenter IPs: ~50 iOS reviews per analysis window vs 300–500 on Android, with occasional full-outage days. We honor the limits instead of rotating around them, so treat iOS results as directional and read the quotes. Android depth is solid.
- LLM cluster titles can be imperfect. That's why percentages are deterministic, and why every item carries raw quotes. And when an analysis quota runs out, the API doesn't just refuse: agent mode kicks in and hands your own model the raw filtered reviews with clustering instructions, so the pipeline never dead-ends.
- The painpoints history endpoint is fresh — the archive only started collecting in late October, so trends need another month or two to become interesting.
What's next
The next post in this series is the engineering deep-dive: what it takes to scrape app stores from the edge in 2026 — Apple's rate-limiting behavior, the relay-and-circuit-breaker design, why 70% of the codebase is defense. If that sounds like your kind of problem, the API itself is the demo: hawkmill.com/docs has the OpenAPI spec and copy-paste examples for all five stores.
Mine any app yourself
Everything in this post is a public API: pain points, raw reviews, version stats — for any app on iOS, Android, macOS, WordPress or Chrome. Keyless in Claude, ChatGPT or Cursor.
Read the API docs Browse the pain indexBuilt from public app-store reviews · not affiliated with Apple or Google · clusters are LLM-generated from real user quotes.