The apps people are leaving: what 34,000 one-star reviews say

Since August we have collected 34,469 one-, two- and three-star reviews from the App Store and Google Play across 682 apps, plus a smaller set of posts from Hacker News, Product Hunt, GitHub and product forums. A language model reads each one and decides whether it describes an unmet need. 16,079 do. We then group the ones that say the same thing. This is what came out.

Where the complaints pile up

Grouping by category, the most complaint-dense corners of the app stores in our sample are not the ones you would guess from a startup pitch deck. They are utilities people are forced to use.

A “repeated complaint” is a cluster of at least two independent reviews that, once summarised, say the same thing. The EV charging category is remarkable: 48 distinct repeated complaints, and one of them, about apps that force a prepaid wallet instead of letting drivers pay per session, has 547 people behind it. That single cluster is larger than most whole categories. We wrote it up separately in EV charging apps are the most complained-about software we track.

The apps with the most low-star reviews in our sample

These are counts of one-to-three-star reviews we collected, not totals on the stores. Our collectors search by niche keyword and read the most recent low reviews of the apps they find, so a popular app in a category we track is over-represented relative to a popular app in one we do not. Read the list as “where dissatisfaction is loud”, not as a ranking of quality.

1. MyFitnessPal, 485 reviews
2. Opal: Screen Time Control, 375
3. Apple Fitness, 331
4. TalkingParents, 326
5. A Better Routeplanner, 306
6. Shopify, 300
7. Fing network scanner, 298
8. Workday, 297
9. OurFamilyWizard, 282
10. SmartLife, 281

Three of the ten are co-parenting apps or close to it, which is the most concentrated market on the list: 289 reviews across three products, most of them about price. A court can order two parents to use one of these apps, and reviewers describe paying a subscription each to comply. The strongest single complaint in the category, from 224 people, is exactly that.

What people actually complain about

Every review gets a short list of complaint tags. Across the 16,079 that describe a real need, the most common tags are dull and consistent, which is the point: a market where the incumbent loses data or does not answer support is a market you can enter with reliability alone.

• Data loss, 43 clusters mention it as the top complaint
• Slow performance, 36
• Bait-and-switch pricing, 34, plus another 19 under a near-identical tag
• Crashes, 33 and 22 under two phrasings
• No phone or human support, 30, 23 and 21 under three phrasings
• Confusing or unintuitive interface, 21 and 21
• No refund policy, 19
• Sync failures, 19
• Feature bloat, 19

Two things stand out. Pricing complaints are not about price. They are about a change: a free tier that shrank, a one-time purchase that became a subscription, a feature that moved behind a paywall after people had built a habit on it. And support complaints are about absence rather than rudeness. The phrase that recurs is “no human”.

Willingness to pay is high

The model also tags whether a review signals willingness to pay, either explicitly (“I would pay for”) or implicitly (the reviewer is already a paying customer complaining about what the money buys). 75.7 percent of pain-describing reviews carry one of those signals. Average emotional intensity, on a one-to-five scale, is 3.56. These are not idle complaints. They are customers.

How to use this

The ranked list scores each repeated complaint on demand, pain, supply gap and feasibility, and scales the score down when a cluster rests on few reviews. The top three are free in full, including the reviews behind them and the apps they name. Each category page shows which apps draw the complaints in that market and the themes that repeat.

Method: reviews collected 19 August to 1 September 2026 from the US and UK storefronts of the App Store and Google Play, plus Hacker News, Product Hunt, GitHub issues and Stack Exchange. Only reviews rated one to three stars are read. Counts are of our sample. See how it works.