Explicit Discontinuance and Switching Signals in Generative AI Applications: A Comparative Analysis across Nine Apps
Yönetim Bilişim Sistemlerinde Güncel Yaklaşımlar , SİNAP VAHİD, Editör, Özgür Yayınları, Gaziantep, ss.133-152, 2026
- Yayın Türü: Kitapta Bölüm / Mesleki Kitap
- Basım Tarihi: 2026
- Doi Numarası: 10.58830/ozgur.pub1539.c6226
- Yayınevi: Özgür Yayınları
- Basıldığı Şehir: Gaziantep
- Sayfa Sayıları: ss.133-152
- Editörler: SİNAP VAHİD, Editör
- Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
- Erzincan Binali Yıldırım Üniversitesi Adresli: Evet
Özet
This study examines explicitly reported discontinuance and platform-switching signals in Turkish- and English-language Google Play reviews of nine generative artificial intelligence applications. A total of 1,145,267 reviews posted between 1 July 2025 and 30 June 2026 were screened using rule-based procedures. Of 11,960 candidate reviews, 821 that met strict contextual rules were evaluated in two blind AI-assisted coding passes. The final high-confidence analysis set comprised 692 records, including 543 signals in which discontinuance or switching was reported as having occurred and 149 signals expressing discontinuance intention. To assess the recall limitation of the strict screen, an additional application-by-language stratified sample of 500 reviews was audited from the broader pool of 11,139 records, revealing 110 missed explicit signals. The stratified weighted estimate of the proportion of missed explicit signals in the broad pool was 21.6% (95% CI: 18.2%-25.2%). Accordingly, rates reported per 10,000 reviews are interpreted as high-confidence detection indicators rather than prevalence estimates. Within the high-confidence set, the most frequent reasons were response quality/task failure, technical problems, privacy/trust, and pricing. Among platform switches reported as having occurred, Gemini and Claude were the most frequent destinations. By separating intention from signals of reported enacted behavior and by explicitly reporting screening sensitivity, this study provides complementary evidence to the post-adoption literature on generative AI.