Grant Writing Assistance: Specific Aims, Specifically

Grant proposals are among the most structured documents in research. NIH and NSF both publish explicit scaffolds: specific aims, significance, innovation, approach, broader impacts. Structure is what LLMs do best. The danger is that structure also lets bad content hide — you can produce a perfectly-formatted proposal that says nothing new.

The specific aims workflow

1. 1Write a one-paragraph description of the scientific problem, in plain English
2. 2Ask the LLM to propose 3 specific aims that would address the problem
3. 3Refine the aims with your domain expertise — the LLM doesn't know your field well enough
4. 4Ask the LLM to pressure-test each aim: 'what would a skeptical reviewer say?'
5. 5Draft the specific aims page with the refined aims

The broader impacts section (NSF)

Broader impacts is where proposals often get weakest — researchers tack on outreach bullet points at the end. Use the LLM to pressure-test your broader impacts: 'Would a program officer find these specific, funded-to-execute, and measurable? If not, rewrite.'

• Save every prompt and response — funders may require AI-use disclosure
• Your institution likely has an AI-use policy for grants; check it
• For NIH, the 2024 guidance discourages AI use for peer review but allows AI-assisted writing with disclosure

The big idea: AI drafts grant scaffolds beautifully. The science has to come from you — and the institutional disclosure rules matter.

Budget justifications are sponsor-format-specific drafting; AI handles the format so PIs spend their time on the substantive science.

What AI does well here

• Draft personnel justifications with appropriate effort percentages and salary basis explanations
• Generate equipment justifications tying each item to specific aim activities
• Produce travel justifications connecting conference attendance to project deliverables
• Draft indirect cost rationale paragraphs matching the institution's negotiated rate

What AI cannot do

• Substitute for the sponsored programs office's compliance review
• Make budget-strategy decisions (where to invest, what to cut)
• Generate justifications for costs the PI hasn't justified to themselves

Funders read hundreds of proposals; tightness signals seriousness. AI is a ruthless line editor that won't fall in love with your sentences.

What AI does well here

• Cut filler phrases and academic hedging
• Surface buried claims and promote them
• Flag passive voice in critical paragraphs
• Suggest stronger opening hooks

What AI cannot do

• Judge what your program officer actually values
• Add domain credibility you haven't earned