English speaking is considered the most difficult and anxiety-provoking language skill for EFL learners due to lack of access to authentic language use, fear of making mistakes, and peers' negative comments. With automatic speech recognition and natural language processing, intelligent personal assistants (IPAs) have potential in foreign language interaction applications. Further empirical research is required to validate the usefulness of IPA interactions in improving EFL speaking skills. Therefore, this study investigated the impact of IPAs, especially IPA feedback presentation mode, on adolescent EFL learners' speaking proficiency. A total of 88 ninth-grade EFL learners were recruited from three intact classes and randomly divided into two experimental groups (i.e., one using Google Home Hub with multimodal feedback and one using Google Home Mini with audio feedback only) and one control group (that interacted with a teacher and peers in a conventional classroom). Both quantitative (English speaking test results) and qualitative data (semi-structured interview responses) were collected and analyzed in the 10-week study. The results revealed that interacting with Google Assistant through Google Home Hub improved the adolescent EFL learners' speaking proficiency most markedly. Detailed analyses of the participants' interviews demonstrated that IPA-mediated interaction (1) added variety and enjoyment to EFL speaking, (2) expanded exposure to English and learner-centered speaking practice with immediate feedback, (3) offered greater authenticity and flexibility in interactions, and (4) encouraged peer collaboration. Furthermore, multimodal presentation of IPA feedback was superior to single-mode presentation in enhancing EFL speaking proficiency. The narration, on-screen text, and graphics were crucial to the EFL learners' successful interaction with the IPA.