Grant & Proposal Development

Moderate · Communication & Dissemination track · ~35 min hands-on + readings and quiz

← Back to the Communication & Dissemination track

Brainstorm, search, and critique with AI — but the first draft of your science is yours to write.

What you’ll be able to do

  • Brainstorm framings and draft specific aims, summaries, and budget justifications yourself, with AI as a reviewer-style critic
  • Know what NIH, NSF, and other funders require you to disclose about AI use
  • Verify every claim, citation, and number before it goes in

Overview

Use AI where it genuinely helps a proposal: brainstorming framings and angles, searching and organizing literature, and critiquing a draft you wrote — the way a study-section reviewer would. Do not use it to write the first draft of your science.

This isn’t just good practice; it’s funder policy. NIH will not review applications substantially developed by AI, uses detection technology to identify AI-generated content, and can refer post-award discoveries to the Office of Research Integrity as a misconduct matter. NSF asks proposers to disclose in the project description the extent to which generative AI was used and how. The science, the feasibility, and the citations are yours to get right — and increasingly, funders require that the writing be yours too.

ImportantRead before you draft: what NIH and NSF actually require

NIH (NOT-OD-25-132, “Supporting Fairness and Originality in NIH Research Applications”): Applications “substantially developed by AI,” or containing sections substantially developed by AI, are not considered the applicant’s original ideas and will not be reviewed. NIH uses detection technology to identify AI-generated content, and if it’s discovered after an award, the matter can be referred to the Office of Research Integrity for a misconduct investigation — with possible consequences including disallowed costs, withheld future awards, suspension, and termination. The notice does allow AI “to assist in application preparation for limited aspects or in specific circumstances” (e.g., administrative tasks), but flags plagiarism and fabricated citations as risks. NIH does not require a disclosure form — its mechanism is prohibition plus detection. The policy is effective for applications submitted to the September 25, 2025 receipt date and beyond. (The same notice also caps PIs at 6 new/renewal/resubmission/revision applications per calendar year, excluding T awards and R13s.) Separately, NIH peer reviewers are prohibited from using AI to write their critiques.

NSF (Notice to the Research Community on Use of Generative AI in the Merit Review Process): Proposers are asked to indicate in the project description the extent to which, if any, generative AI was used and how. You remain fully responsible for the accuracy and authenticity of everything submitted, including any AI-generated content — NSF’s existing fabrication, falsification, and plagiarism policies apply. (NSF reviewers, for their part, may not upload proposal content to non-approved AI tools — it becomes public-domain data.)

Foundations and other agencies: policy is inconsistent and evolving. Check the specific funding opportunity’s instructions every time; when policy is silent or unclear, disclose anyway and never let AI write the substantive science.

Activity 1 below is the recommended workflow: brainstorm with AI, draft it yourself, then get a reviewer-style critique.

Practice activities

Activity 1 · Moderate — Brainstorm, draft it yourself, then get a reviewer's critique

Time ~20 min · Tools ChatGPT Edu

Goal. Use AI at the two ends of the process — brainstorming before you write, critiquing after — while the Aims draft itself is entirely your own writing.

Setup. Your project’s goal, gap, and approach (non-sensitive). Example topic: reducing pedestrian injuries near Philadelphia schools through built-environment changes.

Steps.

  1. Brainstorm framings — options, not prose:

    I’m developing Specific Aims for [topic]. Here are my notes: [goal, gap, approach]. Give me 3 different ways to frame the gap and long-term goal as short bullet options — not full paragraphs — so I can pick the strongest angle and write it myself.

  2. Close the AI tool and write your own first draft of the Aims — opening hook, gap, long-term goal, central hypothesis, and 2–3 aims — in your own words, from your notes and the framing you chose. This is your intellectual work, and this is the step that must be yours.

  3. Get a reviewer’s-eye critique of the draft you actually wrote:

    Critique my Aims as a study-section reviewer would: what’s underdeveloped, what’s overreaching, what’s missing?

  4. Revise the draft yourself based on the critique — again in your own words.

Expected result. A self-written, revised one-page Aims draft, informed by AI brainstorming and stress-tested by a reviewer-style critique.

Check your work. Verify every citation and statistic against the actual source, and confirm the hypothesis and aims are your ideas expressed in your words.

Common pitfalls. The tempting shortcut — asking AI to draft the page and lightly rewording it — is exactly what NIH’s NOT-OD-25-132 targets: applications substantially developed by AI will not be reviewed, and NIH uses detection technology to find them. AI also invents plausible citations and overpromises feasibility. Brainstorm and critique with AI; write with your own hands.

Stretch (optional). Ask the AI which of your two or three candidate framings a skeptical reviewer would find most compelling, and why — then decide for yourself.

Activity 2 · Moderate — Fit a section to the funder's rules

Time ~15 min · Tools ChatGPT Edu

Goal. Tighten a section you’ve already written to the funder’s guidelines and page limits without dropping required content.

Setup. A draft section you wrote yourself (e.g., a significance paragraph or budget justification) plus the funder’s instructions.

Steps.

  1. Ask it to check compliance and tighten:

    Here is my draft and the funder’s formatting and length rules. Flag where I don’t meet the guidance and tighten the text to fit — without dropping anything required. [draft] [rules]

  2. Confirm nothing required was cut.

  3. Surface gaps:

    List anything the guidelines require that my draft is still missing.

Expected result. A tightened, compliant section plus a gap checklist.

Check your work. Re-read against the funder’s actual instructions yourself — AI can miss a requirement.

Common pitfalls. Trimming can quietly remove mandatory elements; verify. And disclose AI assistance per the funder’s actual instructions — NSF, for example, asks proposers to state in the project description whether and how generative AI was used.

Stretch (optional). Draft a cover-letter paragraph on the proposal’s fit to the funding opportunity.

Activity 3 · Moderate — Cross-check a multi-document application package with Claude Cowork

Time ~20 min · Tools Claude Cowork (consumer/PI-purchased — see below)

Data classification — read this first. Claude Cowork is not a Drexel-supported tool and is approved for Low Risk Data only unless separately approved through Drexel’s Third-Party Risk Assessment process. Real grant application content is normally unpublished and competitively sensitive — it is not low-risk. Do this activity only with the fabricated files below (or your own clearly fictional materials). Real proposal documents belong in an approved Drexel tool.

Goal. Use an agentic, folder-based tool to catch cross-document inconsistencies in an application package you wrote yourself — the kind a study section or grants office would flag, and a single-document chat review would miss. No drafting: AI reads, compares, and critiques; you make every fix.

Setup. Create a folder on your computer called mock-grant containing three plain-text files with the fabricated excerpts below (all fictional — invented investigator, institution, and data), then connect Claude Cowork to that folder. At least one inconsistency is deliberately planted.

  • specific-aims.txt:

    Pedestrian injuries near the fictional city of Harborview’s school zones have risen 40% since 2020 (invented figure). Building on our preliminary dataset of 18 months of intersection-camera observations at 12 school crossings, we will test whether low-cost signal-timing changes reduce risky mid-block crossings (Aim 1) and whether effects differ by time of day (Aim 2). Co-Investigator Dr. Elena Vasquez, a biostatistician, will lead the Aim 2 modeling. PI Dr. Sam Okafor will oversee all data collection and analysis.

  • budget-justification.txt:

    Dr. Sam Okafor (PI, 0.6 calendar months) will provide overall scientific direction. Dr. Elena Vasquez (Co-I, 1.2 calendar months) will conduct statistical modeling for Aim 2. A Community Engagement Coordinator (3.0 calendar months) will recruit and convene the parent advisory board described in the outreach plan.

  • biosketch-okafor.txt:

    Dr. Sam Okafor is an Associate Professor of urban health at the fictional Marchmont University with 10 years of experience in transportation-injury epidemiology. Dr. Okafor recently completed a municipal collaboration mapping sidewalk conditions across Harborview. Dr. Okafor’s current projects focus on qualitative interviews with crossing guards.

Steps.

  1. Ask Cowork to review the whole package as a reviewer would — comparing across files, not editing them:

    This folder contains three files from a draft grant application I wrote myself: a Specific Aims page, a budget justification, and the PI’s biosketch personal statement. Act as a study section reviewer and a grants-office pre-award checker. Read all three files and flag cross-document inconsistencies only: claims in one document that another document doesn’t support, effort or personnel that don’t match, named collaborators or resources with no corresponding documentation, and anything a reviewer would question. For each flag, cite which files conflict and quote the conflicting passages. Do not rewrite or edit anything.

  2. Compare its findings to the planted problems. It should catch at least: the Aims page claims a preliminary intersection-camera dataset that the biosketch never mentions or supports; the budget funds a Community Engagement Coordinator and parent advisory board that appear nowhere in the Aims; and the Aims says the PI will “oversee all data collection and analysis” while the budget gives the PI only 0.6 months.

  3. Decide which flags are real problems and revise the files yourself, in your own words. Then ask for a re-check:

    I’ve revised the files. Re-read all three and tell me whether the inconsistencies you flagged are resolved, and whether my revisions introduced any new mismatches.

Expected result. A cross-document inconsistency report citing specific files and passages — then, after your own revisions, a clean (or cleaner) re-check.

Check your work. Verify each flag against the actual files yourself; agentic tools can misquote or over-flag. And confirm it caught all three planted problems — if it missed one, that’s a useful lesson in why AI review supplements, never replaces, your grants office.

Common pitfalls. Letting the tool “fix” the files for you — that crosses from critique into drafting, which NIH and NSF policy (and this course’s rule) put squarely on you. Also, connecting Cowork to a folder that contains more than you intended — it can read everything in the folder, so keep practice materials isolated. And never repeat this workflow with real, unpublished proposal documents in Cowork; that requires an approved Drexel tool.

Stretch (optional). Add a fourth fabricated file — a one-paragraph letter of support from a fictional partner — and ask Cowork whether every collaborator and resource named in the Aims now has corresponding documentation somewhere in the package.

Check your readiness

Answer these, then check — your score suggests whether to dive in or skim the readings first.

Useful resources