Study of 17,000 AI App Reviews Shows What Really Makes Users Quit

A new study took one of the largest looks yet at what ordinary people actually say about AI chatbots. Researchers collected 17,012 English-language reviews from Google Play and Apple's App Store for six popular apps: ChatGPT, Gemini, Microsoft Copilot, Claude, DeepSeek, and Perplexity. They then used AI tools to sort the reviews into themes and score whether each one was positive or negative, and double-checked the results against 300 reviews that people read by hand.

The biggest finding is that people aren't mainly complaining about the intelligence of these tools. They're complaining about the packaging around them. Among reviews mentioning ads, 91% were negative. Logging in and account problems: 89% negative. Servers crashing or slow responses: 83% negative. Subscription pricing: 73% negative. In other words, the frustration is rarely "the AI gave a bad answer" — it's "the app made me watch an ad," "it logged me out again," or "it went down when I needed it."

The apps also had noticeably different reputations. Claude stood out with the highest share of negative reviews (47.7%) — but also a strongly enthusiastic fan base, meaning its users are unusually split between loving it and hating it. Reviews of DeepSeek, the Chinese-made chatbot, brought up data privacy and geopolitical concerns tied to where the company is based, something the other apps largely avoided.

The study also proposes a "Trust Friction Score" — a simple way to sum up how much each app's quirks get in users' way. The honest catch: these are people who bothered to leave a review, a grumpier and more technical crowd than the average user, and the paper is still under peer review. Still, the practical lesson is clear — the companies that fix their plumbing, not just their AI, are the ones people stick with.