Are ChatGPT, Gemini and DeepSeek adequate for intralingual translation of clinical reports in Spanish? A readability quantitative study

Abstract Accessibility refers to the conditions that processes, products and services must meet to be understandable and usable for everyone. This includes specialized information like medical reports. The current legislation in Spain and other European countries grants patients access to their medical history. However, their clinical reports often include numerous technicalities that hinder patients’ comprehension and access to potentially important health information. Large Language Models (LLMs) can be used for intralingual translation of specialized texts with the aim of adapting them for lay users. However, their suitability for this task has not been evaluated with clinical reports in Spanish. Here, ChatGPT, Gemini and DeepSeek were used with two different prompts to adapt 250 anonymized clinical reports in Spanish and the difficulty of the translated texts was evaluated with the Spanish-validated Inflesz readability scale. Results showed that the three LLMs significantly improved readability, although the translated texts still fell into the “difficult” or “standard” categories. Despite positive results with the three LLMs, DeepSeek consistently outperformed ChatGPT, while the performance of Gemini depended markedly on the prompt and showed the largest variability. Further research is needed to assess the adequacy of these intralingual translations from the medical and the usability points of view.