Grant Administration with AI

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

← Back to the Communication & Dissemination track

Use AI for the administrative side of grants — deadlines, formatting, compliance — without touching sensitive budget or personnel details.

What you’ll be able to do

  • Use AI to track deadlines, summarize requirements, and format compliance documents
  • Recognize which grant-administration tasks are safe for AI and which aren’t
  • Build reusable prompts for recurring admin tasks (progress reports, no-cost extensions, subaward tracking)

Overview

Grant administration involves a lot of recurring, structured work: tracking deadlines, formatting budget justifications, drafting progress report narratives, summarizing sponsor requirements, and compiling compliance documentation. AI is good at all of these — and the data is often non-sensitive (public funding announcements, published guidelines, de-identified summaries).

A note on how this page differs from Grant & Proposal Development: that page covers drafting the scientific content of grant applications, where funder policies like NIH’s NOT-OD-25-132 apply and the stakes of AI use are much higher. The administrative and progress-report work covered here is a lower-risk category — but progress report narratives are still official submissions to a federal sponsor, so verify every fact before you submit, and disclose your AI use if your sponsor or institution asks. See the proposal-development page for the fuller discussion of funder AI policies.

The boundary is clear: AI can help with formatting and summarizing public or non-sensitive grant information. It should not be used with identifiable personnel data (salaries, effort allocations with names), real budget numbers that are confidential, or anything involving sponsored program compliance decisions (those require human judgment and institutional sign-off).

Practice activities

Activity 1 · Novice — Summarize sponsor requirements

Time ~15 min · Tools ChatGPT Edu

Goal. Extract and organize key requirements from a funding announcement.

Setup. Use a real, publicly posted funding opportunity announcement (FOA). Here’s one to use:

Go to NIH Guide and find any current R21 or R01 FOA in injury prevention, environmental health, or a topic in your area. Copy the “Application and Submission Information” section.

Alternatively, use this summary of typical NIH requirements:

Funding Opportunity: PA-24-XXX, NIH R21 Exploratory/Developmental Research Grant. Page limits: Specific Aims (1 page), Research Strategy (6 pages for R21), Budget (modular up to $275,000 direct costs over 2 years). Required sections: Specific Aims, Significance, Innovation, Approach, Protection of Human Subjects (if applicable), Data Management and Sharing Plan (required per 2023 policy), Authentication of Key Biological/Chemical Resources (if applicable). Letters of support required from subaward sites. Biosketch format: NIH format (5 pages). Submission via Grants.gov with eRA Commons validation. Due dates: standard NIH cycles (February 5, June 5, October 5 for new R21s).

Steps.

  1. Extract the key requirements:

    From this FOA summary, create a pre-submission checklist organized by: (a) documents to prepare, (b) page/format limits for each, (c) institutional approvals needed, and (d) key dates. Flag anything that’s easy to miss.

  2. Verify each item against the actual FOA — AI may miss recent policy changes (e.g., the 2023 Data Management and Sharing Plan requirement).

  3. Save the checklist as a reusable template for your next submission.

Expected result. A clean pre-submission checklist you can reuse. The AI should catch most items but may miss institution-specific requirements (e.g., Drexel’s internal routing deadline, which is typically 5 business days before the sponsor deadline).

Check your work. Compare to your grants office’s checklist if they have one. The AI version is a starting point, not a replacement for institutional guidance.

Common pitfalls. AI doesn’t know your institution’s internal deadlines, routing requirements, or F&A rate. Layer those in yourself.

Stretch (optional). Create checklists for two different mechanisms (e.g., R01 vs. foundation LOI) and note what differs.

Activity 2 · Novice — Draft a progress report narrative

Time ~18 min · Tools ChatGPT Edu

Goal. Use AI to draft a progress report narrative from bullet points, then verify it against what actually happened.

Setup. Use these fabricated project milestones (or substitute your own non-sensitive project details):

Project: Pedestrian Safety Infrastructure and Injury Outcomes (NIH R01, Year 2 of 5)

Year 2 milestones completed: - Geocoded 14,200 crash records from PennDOT 2018-2023 files - Linked crash data to road-segment characteristics (lanes, speed limit, crosswalk presence) from PennDOT and OpenStreetMap - Completed missing speed-limit imputation (nearest-neighbor method, validated against 500 manually checked segments) - Submitted IRB continuing review (approved March 2026) - Published 1 paper (Injury Prevention, accepted January 2026) - Presented at APHA annual meeting (November 2025)

In progress: - Multilevel models for KSI outcome — preliminary results show 2.3x odds on 4+ lane roads - Hiring replacement research coordinator (previous coordinator left February 2026)

Challenges: - 2023 PennDOT file format change broke ingestion script — 3 weeks to fix - Coordinator turnover delayed data cleaning by approximately 6 weeks

Steps.

  1. Draft the narrative:

    From these project milestones, draft a 300-word progress report narrative for an NIH annual progress report. Tone: factual, no adjectives like “exciting” or “significant.” Structure: Accomplishments, Challenges, and Plans for Next Year. Include specific numbers where available.

  2. Verify every claim — did the draft invent any accomplishments not in the milestones? Did it minimize the challenges?

  3. Add the one thing AI can’t know: your assessment of whether the project is on track, and any changes to the scientific direction.

Expected result. A clean, factual narrative that saves 30–60 minutes of writing time. You should need to add: (1) your PI assessment of progress, (2) any changes to specific aims or approach, and (3) institutional details the AI doesn’t know.

Check your work. Every number in the narrative must match the milestones. AI tends to inflate (“over 14,000 records” when the number is 14,200 — use the exact number).

Common pitfalls. AI-drafted progress reports often minimize challenges (“a minor delay”) when the actual impact was significant. Be honest — NIH reviewers appreciate candor, and understating delays can backfire in a site visit. And remember that a progress report is an official record submitted to the sponsor — it’s not governed by NIH’s application-focused NOT-OD-25-132 policy, but every fact in it still needs to be accurate and verified by you.

Stretch (optional). Draft the progress report for a foundation grant using the same milestones but a different format and audience (shorter, less technical).

Activity 3 · Novice — Build a reusable deadline tracker with Claude Code

Time ~25 min · Tools Claude Code (personal/PI-purchased — not Drexel-supported; Low Risk Data only)

Goal. Instead of re-running the Activity 1 prompt every time you add a funding opportunity, have Claude Code build a small script you can rerun forever: it reads a list of FOA deadlines from a file and prints a sorted deadline calendar.

Setup. Claude Code runs in your computer’s terminal — the text-based window where you type commands (Terminal on Mac, PowerShell or Windows Terminal on Windows). Unlike a chat window, Claude Code can read and write real files on your computer and run code, which is what makes a reusable tool possible. Install it from Anthropic’s Claude Code page and open a terminal in an empty folder (e.g., foa-tracker).

Data reminder: Claude Code is approved for Low Risk data only — use public FOA information and the fabricated data below, never real budget figures or personnel names/salaries.

Create a file called foas.csv in that folder (any text editor works — or ask Claude Code to create it for you) with this fabricated sample:

title,sponsor,page_limit,due_date
Community Air Quality Sensor Networks (R21),NIH/NIEHS,6,2026-10-16
Pedestrian Injury Prevention in Urban Corridors (R01),NIH/NICHD,12,2026-11-05
Regional Wastewater Surveillance Capacity,CDC,10,2026-09-30
Heat Resilience Planning for Older Adults,RWJF,8,2027-01-15

Steps.

  1. In the folder containing foas.csv, type claude to start Claude Code, then describe the tool you want:

    I’m a grants administrator tracking multiple funding opportunities. Read foas.csv in this folder — it has columns title, sponsor, page_limit, due_date. Write a simple Python script called deadline_tracker.py that reads this file and prints a table of all opportunities sorted by due date (soonest first), showing title, sponsor, page limit, due date, and days remaining from today. Flag anything due within 60 days. Keep the script simple enough for a non-programmer to rerun, and tell me the exact command to run it.

  2. Claude Code will show you what it plans to create and ask permission before writing files or running anything — read each request before approving. When it finishes, run the command it gives you (typically python deadline_tracker.py).

  3. Test that it’s genuinely reusable: add a fifth made-up FOA row to foas.csv yourself, rerun the script, and confirm the new opportunity appears in the right sorted position.

  4. Ask for one refinement in plain English:

    Update the script so it also writes the sorted table to a file called deadline_report.txt that I can paste into an email to my study teams.

Expected result. A working deadline_tracker.py you keep and rerun each grant cycle — add a row to the CSV, run one command, get an updated sorted deadline calendar. No re-prompting required.

Check your work. Count days-remaining by hand for at least two rows against a real calendar, and confirm the sort order matches the due dates. Just like Activity 2: verify every number the tool produces before anyone relies on it.

Common pitfalls. The script only knows what’s in your CSV — it won’t know about internal routing deadlines, so consider adding a column for your grants office’s internal due date (usually days before the sponsor deadline). A typo in a date (e.g., 2026-13-01) can break the script or silently mis-sort; if that happens, just paste the error message into Claude Code and ask it to fix it. And remember the classification rule: this file lives on your computer and passes through a non-Drexel tool, so keep it to public, fabricated, or otherwise Low Risk information only.

Stretch (optional). Ask Claude Code to add a status column (e.g., “drafting,” “in routing,” “submitted”) and print submitted opportunities in a separate section — a first step toward the pre-submission checklist from Activity 1 living in a file instead of a chat window.

Check your readiness

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

Useful resources