Table 2. Challenges encountered with LLM use in research among adopters (n=496).

Challenge	N endorsed (%)
Inaccurate or misleading responses	390 (78.6%)
Lack of transparency about sources and citations	298 (6.1%)
Ethical concerns (eg, data privacy/confidentiality, authorship, plagiarism)	227 (45.8%)
Bias in responses	134 (27.0%)
Over-reliance on AI-generated content	96 (19.4%)
Institutional or journal policies restricting LLM use	80 (16.1%)
Technical limitations (eg, token/context length, inability to handle large datasets)	131 (26.4%)
No challenges encountered	18 (3.6%)
Other (unique challenges not covered above referenced two times or more)	26 (5.2%)
Environmental impact	3 (<1%)
Financial limitations	2 (<1%)
Responses not useful	10 (2.0%)

AI, artificial intelligence; LLM, large language model.