Improving pronunciation has long been one of the toughest challenges for English learners — especially outside of formal classroom settings.
Now, artificial intelligence is starting to change that. And a new academic study shows just how much of a difference the right AI tool can make.

In early 2025, researchers at Iowa State University published a study in the Journal of Second Language Pronunciation exploring whether generative AI-powered chatbots could improve second-language (L2) learners’ comprehensibility — how clearly they are understood.

From Traditional CAPT to AI-Powered Pronunciation Practice

Computer-Assisted Pronunciation Training (CAPT) tools have been around for decades. But historically, they relied on repetitive drills and lacked the ability to offer natural conversation or personalized feedback.

The rise of generative AI changes that. Today’s AI chatbots can simulate free-flowing dialogue, offering learners more authentic speaking opportunities. As the researchers note, “Generative AI-based chatbots enable more natural, free-form dialogues, offering immediate and personalized feedback that aligns with contemporary communicative language teaching (CLT) principles.”

However, the study highlights an important distinction: not all AI chatbots are equally effective for pronunciation practice.

Inside the Experiment: Pronounce vs. Gemini

To explore the impact of AI chatbots on pronunciation, researchers compared two tools:

• Pronounce AI: Built specifically for language learning, Pronounce offers explicit feedback on pronunciation, grammar, vocabulary, and fluency.
• Google Gemini: A general-purpose chatbot that offers only implicit feedback through real-time transcription of user speech.

Over three weeks, English learners participated in nine 30-minute sessions with their assigned chatbot. Practice tasks included free conversations, read-aloud exercises targeting difficult sounds, and structured role-plays.

As the researchers explained, learners used Pronounce “to receive feedback on their pronunciation,” while Gemini users relied on “the transcript of their spoken words as implicit feedback.”

What the Study Found

Although the study's small sample size (n=15) limited group-wide statistical significance, individual patterns were clear — especially among Pronounce users. One participant described Pronounce AI as “like a tutor who points out my mistakes,” while others appreciated how it raised their consciousness about previously unnoticed pronunciation errors.

1. Motivated Learners Improved Their Comprehensibility

Learners who practiced consistently and seriously — completing at least seven practice sessions in the presence of a researcher — showed notable improvement in how understandable their English became.

“We found improvement in some (motivated) individual learners’ comprehensibility ratings after the training with chatbots,” the authors noted.

One participant described their experience:

“Before using the chatbot, I always neglected my pronunciation. After using it, it’s just like a tutor who points out my mistakes. Now I know my mistakes and will pay more attention to them.”

2. Explicit Feedback Made a Difference

Learners using Pronounce became more aware of specific pronunciation issues, such as vowel length, word stress, and linking sounds — compared to Gemini users, who tended to notice pronunciation challenges only in general terms.

“Conversing with Pronounce helped learners recognize issues like vowel differentiation, word stress, and linking words.”

In contrast, Gemini’s transcript-based feedback often required learners to guess what they had done wrong if the bot misunderstood them.

“You have to give it instructions and remind it of those instructions... otherwise it won’t be a conversation; it will be an answer to a question.” — Gemini user reflection

3. Structured Practice Multiplied the Benefits

Students who practiced in a guided setting — often with a mentor prompting them to reflect or ask questions — improved much more than those practicing independently.

The researchers emphasized:

“Learners who completed their speaking practice in the researcher’s office showed higher comprehensibility gains compared to those who practiced independently.”

This points to a critical lesson: AI works best when integrated into structured learning environments, with guidance and feedback loops.

Why This Matters for Language Learners and Educators

The study offers several takeaways for anyone teaching or learning English pronunciation:

• Purpose-built AI tools (like Pronounce AI) outperform general-purpose chatbots when it comes to improving speaking skills.
• Explicit, actionable feedback accelerates learner awareness and improvement.
• Consistent, mentored practice leads to much stronger results than unguided solo practice.
• AI should complement, not replace, the human elements of teaching — such as motivation, reflection, and tailored advice.

As the researchers conclude:

“These chatbots could be very helpful for learners’ pronunciation improvement, but their effectiveness is maximized when learners are motivated and have the guidance of a teacher.”

The Future: AI as a Learning Partner — Not a Replacement for Teachers

The researchers emphasize that while AI chatbots like Pronounce offer powerful, on-demand speaking practice, they are most effective when used alongside human guidance, not as a substitute for it.

Learners who practiced with mentor support — asking questions, reflecting on feedback, and receiving encouragement — made the greatest improvements. In contrast, those practicing entirely on their own showed smaller or inconsistent gains.

AI tools can amplify practice opportunities, but they cannot fully replace the personalized support, motivation, and instructional guidance that teachers and mentors provide. For the best results, Pronounce AI should be part of a structured, guided learning routine.