Assessing chatbots ability to produce leaflets on cataract surgery: Bing AI, chatGPT 3.5, chatGPT 4o, ChatSonic, Google Bard, Perplexity, and Pi.

PURPOSE: To evaluate leaflets on cataract surgery produced by 7 common free chatbots.

SETTING

UK-based ophthalmologists carrying out online research.

DESIGN

Data were collected from the responses of 7 freely available online chatbots.

METHODS

Analysis of answers given by 7 chatbots (Bing AI, chatGPT 3.5, chatGPT 4o, ChatSonic, Google Bard, Perplexity, and Pi) was prompted to "make a patient information leaflet on cataract surgery." Answers were evaluated using the DISCERN instrument, Patient Education Materials Assessment Tool (PEMAT), presence of misinformation, the Flesch-Kincaid Grade level readability score, and material reliability.

RESULTS

The highest overall scored response was from ChatSonic, followed by Bing AI and then Perplexity. The lowest scoring was ChatGPT 3.5. ChatSonic achieved the highest DISCERN and PEMAT scores, and had the highest Flesch-Kincaid Grade level. The lowest DISCERN and PEMAT scores were for Pi. Only ChatGPT 3.5 included some misinformation in its response. Bing AI, ChatSonic, and Perplexity included reliable references; the other chatbots provided no references.

CONCLUSIONS

This study demonstrates a range of answers given by chatbots creating a cataract surgery leaflet, suggesting variation in their development and reliability. ChatGPT 3.5 scored the most poorly. However, ChatSonic indicated promise in how technology may be used to assist information giving in ophthalmology.