Study Finds AI-Drafted Grant Proposals More Likely to Win NIH Funding

Grant proposals written or edited with AI chatbots are more likely to win funding from U.S. agencies — but at a cost that should concern every researcher chasing originality, according to research reported by Nature in February 2026.

The finding comes from a study by Qian et al., published as a preprint on arXiv, which analyzed patterns in funded versus unfunded proposals submitted to U.S. agencies including NIH. The core discovery: AI-assisted proposals tend to be more similar to previously funded projects, and that conformity appears to correlate with higher success rates.

The Conformity Trap

The mechanism is straightforward. Large language models trained on vast corpora of scientific text naturally gravitate toward established framings, common methodologies, and safe research directions. When researchers use these tools to draft or polish proposals, the output subtly shifts toward what has already worked — the kind of incremental, low-risk science that review panels have historically favored.

This creates a paradox. Individual researchers gain a competitive edge by using AI. But if everyone does it, the entire applicant pool converges toward sameness, potentially crowding out the high-risk, high-reward proposals that drive scientific breakthroughs.

NIH's Response

NIH has already moved to address AI-driven volume. In 2025, the agency limited scientists to six applications per year, partly in response to fears that AI tools would flood review panels with machine-polished submissions. The new research suggests the quality concern may be more nuanced than the quantity concern.

What Researchers Should Consider

The study doesn't argue against using AI in grant writing — that ship has sailed. The strategic takeaway is to use AI for mechanics (clarity, structure, compliance) while preserving what makes a proposal distinctive: the original scientific question, the unexpected methodology, the contrarian hypothesis. Reviewers can spot boilerplate. What they fund is conviction.

Tools like Granted help researchers identify the right funding opportunities, so the human effort can focus on crafting proposals that stand apart rather than blend in.