JRC report reflects on the impact of the use of LLMs in proposal writing and its influence in proposal quality in the context of Horizon Europe.
Since the public launch of ChatGPT in 2022, there has been a sharp increase in the use of generative AI in daily tasks and professional environments, and funding programmes are no exception to this trend. These programmes have witnessed a rise in applications. For example, the European Innovation Council (EIC)’s first call on Advanced Innovation Challenges closed in February 2026 with 709 proposals requesting a budget of €130.7 million – widely exceeding the dedicated budget for the first round. As the EIC President explained, the growing use of AI could be part of the reason. A similar trend is also visible in other EIC calls, such as the EIC Transition calls, which received 611 proposals in 2025, 413 in 2024 and 257 in 2023, demonstrating a clear increase. In this regard, success rates for grants proposals are decreasing due to higher numbers of applications. Therefore, the question remains: does the use of AI impact the success rate of these proposals?
In this context, JRC’s recent report “LLM-assisted proposal writing in competitive R&D fcaunding: Evidence from Horizon Europe” provides insights on the use of large language models (LLMs) by the applicants in grant proposal writing. The report draws on the case of Horizon Europe, the EU’s flagship research and innovation funding programme, and focuses on firm applications between 2021 and 2024. As proposal writing for funding programmes is a significant hurdle and resource-intensive for companies, LLM-assisted proposal writing emerges as a promising solution to reduce costs and lower the participation barriers. Thus, the report aims to examine the following main questions: “i. how widespread is LLM-assisted grant proposal writing, ii. which firms use it, and iii. whether its adoption affects the evaluation outcome”.
Firstly, the results demonstrate that LLM-assisted writing has significantly increased in the period after the public availability of ChatGPT in 2022. By 2024, approximately 40% of Horizon Europe proposal abstracts contained some sort of AI-generated content. Secondly, specific characteristics are observable among the users of LLMs in proposal writing. It is reported that the main users are small and less productive firms. Although some disparities in adoption patterns are noticeable, the usage is more prominent in countries with lower levels of GDP per capita, R&D intensity, English proficiency and digital skills. This contradicts with findings stating that AI is mostly used in larger firms with higher levels of productivity and innovativeness. Lastly, while one may assume that the use of AI would lead to lower proposal quality, the results reflect that the quality of the proposals is not significantly diminished by using LLMs. However, the quality remains strongly dependent on the intensity and level of AI reliance. Ultimately, it appears that proposals written with the help of generative AI do not shift how the quality of the submitted projects is assessed. Consequently, the sole usage of AI does not necessarily give firms an advantage in funding outcomes.
Although limitations on the study remain, this report stands as a starting point for future research on the topic of LLM-assistance in research funding. Moreover, it sheds light on potential positive outcomes of these tools. In fact, AI reduces application costs for firms and opens the doors for companies with limited resources and capacities to apply for funding programmes. Nevertheless, the overburdening of workflow on the evaluators’ side due to the increasing number of applications should not be ignored and remains an open challenge for funding programmes and agencies.
In the future, it remains relevant to explore new ways of funding both for the European Commission (EC) and research funding agencies, especially as human capacity is limited and the current evaluation process might lose its effectiveness in the age of AI. In regards, some reflections have already been raised on the reviewing process. For instance, the EC’s publication from 2023 ponders on alternative ways of decision-making in research funding such as the portfolio approach. On the Swiss side, the Swiss National Science Foundation (SNSF) has experimented with the lottery method, which consists of randomly assigning project funding to research proposals after an initial screening. Considering alternative methods, combining AI with clear guidelines for responsible use and safeguards for research quality could emerge as a new approach to research funding.