Comparative Assessment of Generative Artificial Intelligence Platforms in Providing Multilingual Pediatric Dental Postoperative Instructions: A Mixed-Method Study.

Purpose: To comparatively evaluate the accuracy, readability, and concision (word count) of 7 generative artificial intelligence platforms (GenAIs) that developed multilingual postoperative instructions (POIs) for pediatric dental procedures.

Methods

Standard prompts were developed and uploaded to 7 trained or untrained GenAIs to create POIs for 5 pediatric dental procedures in 3 languages: English, Spanish, and Polish. Two masked, calibrated, bilingual, simulated parents assessed the accuracy, clarity, cultural neutrality, and language appropriateness of 70 GenAI-developed POIs using an ordinal scale (AI-instructify) for each language pair (English-Spanish, English-Polish). The readability, understandability, actionability, and concision of GenAIs were calculated using evidence-based indices, and the data were statistically evaluated (a =0.05).

Results

Trained GenAIs improved the accuracy and readability of multilingual POIs (P<0.05). POIs in English consistently received higher AI-instructify scores than POIs in Spanish and Polish. However, POIs from untrained GenAIs or in the Polish language were concise (lower word count) as compared to those in English and Spanish languages (P<0.001). Most POIs received a score indicating they were "easily" or "fairly easily" readable, understandable, and actionable at a sixth-grade level (the standard literacy level for the US population).

Conclusions

Trained generative artificial intelligence platforms produced accurate, readable postoperative instructions. Despite generic content, GenAIs showed promise in generating multi-lingual instructions suitable for diverse populations with lower health literacy.