byKing's College London

Schematic representation of the 2-stage transcription pipeline. Credit:Journal of Dental Research(2025). DOI: 10.1177/00220345251382452

A new study from King's College London has revealed that artificial intelligence (AI) automatic speech recognition (ASR) tools could dramatically improve how dental professionals record patient information—saving time and reducing administrative burden.

However, while transcriptional accuracy of these tools is high, they can struggle with more technical language, and their reliability is not currently sufficient to support unreviewed use.

The findings were recentlypublishedin theJournal of Dental Research. Researchers tested 10 different ASR systems to see how well they could transcribe spoken orthodontic clinical records into written text.

The best-performing system was an experimental pipeline combining OpenAI's GPT-4o transcription with alarge language modelforerror correction, closely followed by the Heidi Health digital scribe and GPT4oTranscribe speech-to-text application programming interface.

Dentists spend significant time typing upclinical notesoften during the consultation, which can reduce face-to-face time with patients.

ASR tools can allow clinicians to dictate their clinical notes naturally, freeing them up to focus more on direct interaction with the patient. The most advanced systems were faster and more accurate than manual typing, with up to 60% in time savings.

The AI-enhanced experimental pipeline (GPT4oTranscribeCorrected) was found to have the lowest error rate, especially with technical dental terms. Commercial systems like Heidi Health also performed well, but others—such as Dragon Anywhere—had high error rates and even introduced clinically significant mistakes. Importantly,background noiseand accent had minimal impact on the best systems, making them suitable for real-world clinical settings.

Caution remains, however. While the technology is promising, researchers warn that clinically significant errors—such as misidentifying teeth or treatment plans—can still occur. They recommend a "human-in-the-loop" approach, where clinicians review and edit transcripts rather than relying on them blindly.

Lead author Ruairi O'Kane said, "AI speech tools can streamline documentation and improve efficiency, but we must remain vigilant. Even subtle transcription errors can potentially impact patient care."

The team suggests future systems should include confidence indicators to flag uncertain terms and be trained on larger, more diverse dental datasets. Ultimately, the goal is to help clinicians become editors of their notes—not just authors—while maintaining safety and accuracy.

More information: R. O'Kane et al, Transcription Accuracy of Automatic Speech Recognition for Orthodontic Clinical Records, Journal of Dental Research (2025). DOI: 10.1177/00220345251382452 Journal information: Journal of Dental Research

Provided by King's College London