12 August 2026 — United States: Artificial intelligence may be giving university researchers a new advantage in the fierce competition for funding but that advantage could come at the cost of originality.

A newly released study from Northwestern University’s Kellogg School of Management suggests that grant proposals displaying stronger signs of large language model assistance were approximately four percentage points more likely to receive funding from the National Institutes of Health.

However, researchers uncovered a troubling pattern. Proposals with greater indications of AI involvement were also more likely to resemble projects that funding agencies had recently supported.

The findings suggest that AI may be helping researchers package their ideas more effectively while quietly pulling scientific proposals towards familiar territory.

Thousands of Proposals Examined

Researchers analysed about 5,700 confidential grant applications and approximately 131,000 publicly available funding awards associated with the National Institutes of Health and the National Science Foundation.

The proposals covered the period between 2021 and 2025, allowing the researchers to compare grant-writing patterns before and after generative-AI tools became widely available.

Signs of AI-assisted writing increased sharply after the public release of ChatGPT in late 2022. Proposals eventually appeared to divide into two groups: those with little indication of AI involvement and those showing much heavier use.

Across both confidential applications and publicly announced awards, proposals with stronger AI signals were consistently less distinctive when compared with projects previously funded by the same agency.

The Northwestern University announcement was published on 12 August 2026.

Better Presentation or Less Innovation?

AI can help researchers organise complicated ideas, improve grammar and communicate technical information more clearly. These capabilities may be particularly valuable to scientists working in a second language or institutions without dedicated grant-writing departments.

But the same technology learns from existing documents and established patterns. When asked to improve a proposal, it may naturally produce language and structures resembling applications that were successful in the past.

That can make a proposal appear polished, practical and familiar to reviewers. It could also discourage unconventional thinking.

At the National Institutes of Health, applications showing stronger AI involvement were associated with a higher probability of receiving funding. The resulting projects also produced more early-stage publications—but not more papers belonging to the most highly cited category.

At the National Science Foundation, researchers found no significant relationship between AI signals and either funding success or subsequent publication output.

This difference indicates that AI’s influence may depend on an agency’s evaluation culture, research priorities and expectations.

An Important Limitation

The findings demonstrate an association, not definite proof that AI caused proposals to become less original or more successful.

Researchers estimated AI involvement by examining patterns in the language used in proposal abstracts. They could not independently confirm exactly how every applicant used an AI tool.

It is therefore possible that researchers were already developing relatively conventional projects and simply used AI to present them more effectively.

The original study published in PNAS warns that the results should be interpreted as correlations rather than direct evidence of cause and effect.

Why This Matters for Universities

Research funding determines which medical treatments, technologies and scientific ideas receive an opportunity to develop. If AI-assisted proposals repeatedly favour familiar approaches, universities could produce more studies while exploring fewer genuinely new directions.

The challenge is not necessarily to prohibit artificial intelligence. Instead, universities and funding organisations may need clearer disclosure rules, reviewer training and assessment methods capable of recognising originality beneath polished language.

AI may make a proposal sound more convincing. The greater responsibility for educators and research institutions is ensuring that it does not make the future of science sound exactly like its past.