L26/P-561 Preconception care meets Artificial Intelligence: evaluating the performance of Generative AI, Large Language Models
Human Reproduction · 01.07.2026 · 41 (Supplement_1)
DOI: 10.1093/humrep/deag083.894 ↗Kurz gefasst
Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von Y. Ma und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Study question How do Generative AI (Gen AI) large language models (LLMs) perform when providing preconception health information compared with evidence-based clinical guidelines? Summary answer Across all LLM models, general health behaviors and mental health topics were addressed clearly and coherently, while medication, supplementation and vaccination-related responses lacked clinical specificity. What is known already ESHRE recently published its key research priorities in infertility to maximise impact on science, people and society; listing better preconception and reproductive care and education as the #1 topic for prioritisation[1].Preconception health is critical for optimising maternal and child outcomes, yet public knowledge of key risk factors and recommendations remains limited. Since the launch of ChatGPT, Gen AI, LLMs are increasingly used as sources of health information and advice[2] by the public and healthcare-professionals, especially nurses[3]. This study aimed to comparatively assess preconception health information generated via LLM platforms to explore their potential as useful tools in reproductive health education. Study design, size, duration Twenty user-like prompts were developed from international guidance by WHO, NHS, NICE, ACOG, and FIGO across 13 domains such as: supplements, BMI, chronic disease management, vaccination, mental health, intimate partner violence. For each prompt, a consensus reference answer was constructed from relevant recommendations across at least three organizations. LLM outputs were independently assessed on three dimensions: concordance with guidelines, comprehensibility, and conciseness using a 5-point Likert scale where 5 is high and 1 is low[4]. Participants/materials, setting, methods Free versions of four publicly available Gen AI LLM platforms: ChatGPT (GPT-4), Claude (v4), Gemini (2.5), and Perplexity were selected for analysis based on popularity and access. A comparative analysis of the responses from these LLM platforms was conducted. Each platform’s responses were collected once in a non-logged-in web environment and benchmarked against reference answers derived from international guidelines. Main results and the role of chance LLM platforms generally provided concordant answers for general health behaviours and mental health topics. Content related to diet, mental health conditions, and depression was typically presented with clear structure and coherent responses. In contrast, topics involving medication use, supplementation and vaccination were less concordant. These lacked specific clinical technical components, including dosage considerations, contraindications, or timing-related details. In summary, behavioural and psychosocial topics were communicated clearly, whereas aforementioned clinically technical topics produced less concise responses. Differences were also observed across topic groups that contained multiple sub-topics. For example, weight-related topics showed uneven patterns, with general advice commonly addressed, but more specific guidance related to BMI thresholds or overweight status varying in clarity and depth. ChatGPT provided the strongest concordance. Conciseness was poorest for Gemini. Comprehensibility was similar across platforms. Perplexity responses consistently included cited sources with footnotes or links to evidence. Limitations, reasons for caution The prompts, although varied, may not fully capture the complexity of real-world preconception health queries, especially by the general public. The evaluation was also completed by experts which has implications for generalisability. Findings reflect a snapshot at the time of study, in reality LLM outputs continue to evolve rapidly[5]. Wider implications of the findings Gen AI shows promise as a tool for improving reproductive health education for individuals and for increasing clinician awareness. However, attention must be paid to information quality, in order to ensure that individuals obtain accurate, reliable and understandable information to help them achieve their desired reproductive health and family-building intentions. Trial registration number No
Zitieren
APA
Ma, Y., Grace, B., Schoenaker, D., & Nieroda, M. (2026). L26/P-561 Preconception care meets Artificial Intelligence: evaluating the performance of Generative AI, Large Language Models. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.894
Vancouver
Ma Y, Grace B, Schoenaker D, Nieroda M. L26/P-561 Preconception care meets Artificial Intelligence: evaluating the performance of Generative AI, Large Language Models. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.894
BibTeX
@article{Ma2026,
author = {Ma, Y. and Grace, B. and Schoenaker, D. and Nieroda, M.},
title = {L26/P-561 Preconception care meets Artificial Intelligence: evaluating the performance of Generative AI, Large Language Models},
journal = {Human Reproduction},
year = {2026},
volume = {41},
number = {Supplement_1},
doi = {10.1093/humrep/deag083.894}
}
EUVIMED