AI · 17 November 2026
AI-Assisted Grant Drafting Linked to Less Distinctive Research Ideas
A study of U.S. federal grant applications, reported by PsyPost, found scientists who use generative AI tend to propose research closer to already-funded topics. At the NIH, AI-heavy proposals were funded more often.
What happened
A new study of U.S. federal grant applications, reported by PsyPost, suggests that scientists who use generative AI tend to propose research closer to topics that have already been funded. In short, heavier AI involvement in drafting was associated with less distinctive research ideas.
The findings are not one-directional. At the National Institutes of Health, proposals classed as AI-heavy were funded more often than those that were not, and they were followed by modestly more routine publications, according to the same reporting.
Why it matters
This is a signal about how generative tools shape what gets proposed, not only how quickly proposals are written. When a drafting aid pulls applicants towards established patterns, the pool of ideas reaching reviewers can narrow before any judgement is made. The NIH result adds a complication: the pattern that reduces distinctiveness also coincided with success in funding. That suggests current review outcomes may be rewarding convergence, which could make the problem self-reinforcing.
For leaders responsible for AI adoption in government and public services, the lesson is that these tools change the character of inputs, not just throughput. Organisations that automate drafting, triage or applications should measure what diversity they may be losing alongside the time they save.
The Renascence take
Most coverage will frame this as a question of AI quality or applicant integrity. The more useful reading is behavioural. Generative tools offer a strong default, and defaults shape choices. Applicants who begin from a model's suggestion tend to anchor on what already exists, and reviewers who reward familiar framing reinforce that loop.
Treat AI drafting as a source of options, not a starting point. Build a deliberate step where applicants or teams must generate at least one idea the tool did not suggest, and score originality explicitly during review. If your assessment process rewards polish over novelty, you are effectively paying for convergence, and this study suggests you will get it.
Sources
This briefing was written by our Newsdesk, synthesising reporting from the outlets below. Follow the links for the original coverage.
FAQ
Questions we get on this topic
More in AI
Stay ahead of CX
Get the signal, not the noise.
The stories shaping customer experience — plus the Journal and Experience Loom — in your inbox.


