Intel / article

What Users Think of Generative AI: A Cross-Platform NLP Analysis of Trust and Friction in App Store Reviews

Source: arxiv.org
Author: Md Jafrin Hossain, Umme Nusrat Jahan, Shouvaggo Sharif Shammo
Date: Published Sep 18, 2026

Context

An analysis of 17,012 English-language reviews of six generative AI apps found substantial dissatisfaction with the experience surrounding the model. Negative sentiment was especially prevalent in reviews about advertising, signing in, service reliability and subscription pricing. The researchers used automated topic and sentiment analysis, checking both against human assessments of 300 reviews. Study abstract.

The findings identify concrete obstacles between an AI app's capabilities and a satisfactory user experience. A capable assistant still has to be accessible, dependable and acceptably priced. The reported percentages describe sentiment within review topics; they do not measure the proportion of all customers who are dissatisfied. Methods and findings.

Key takeaways · AI-distilled

Analyzing 17,012 app-store reviews across six GenAI apps, negative sentiment concentrated overwhelmingly in specific complaint categories: advertising (91% negative), authentication (89%), server reliability (83%), and subscription pricing (73%).
Claude showed the highest negative-sentiment share among the six apps (47.7%) alongside a strongly enthusiastic user base -- a statistically significant polarization pattern, not uniformly negative reception.
A subset of DeepSeek reviews raised geopolitical and data-privacy concerns tied specifically to its Chinese origin, an exploratory finding the authors flag as distinct from the main sentiment analysis.
The sentiment and topic models were validated against 300 human-coded reviews before being applied at scale, and findings held up despite unequal review counts across apps.

Read the source arxiv.org