As artificial intelligence tools rapidly become part of the academic research process, understanding how to use them responsibly is more important than ever.

This new guide, based on the PLOS Computational Biology article, 'Ten Simple Rules to Leverage Large Language Models for Getting Grants', authored by Elizabeth Seckel, MA, Brandi Stephens, PhD, and Fatima Rodriguez, MD, translates the original publication into actionable tips for grant writers.

This resource arrives at a particularly timely moment: just last week, the NIH released updated guidance on the use of AI in grant applications (NOT-OD-25-132). If you're preparing a proposal - or simply trying to stay ahead of evolving best practices - this guide is designed to help you integrate AI tools thoughtfully and in line with federal expectations.

Rule 1: Always check the funder’s rules

Funding agencies have different policies on AI use. Some allow it if you disclose it, others restrict it. For example, the NIH issued a policy July 17th, 2025 (NOT-OD-25-132) that prohibits applications that are substantially developed with AI. Before using AI tools like ChatGPT, first check the funder’s website for the latest guidance.

Rule 2: Protect your ideas

Due to the nature of their training, AI tools may save your input to update future models, which risks exposing your original research ideas to other users. As a tip, don't paste any unpublished data, budget details, or your Specific Aims into public chatbots. Instead, use secure AI tools that may be available for use at your institution.

• Example of a secure AI tool: A locally hosted LLM like PrivateGPT or LM Studio can be installed and run entirely offline. These are better for privacy but may be less powerful than tools like ChatGPT. Stanford also has its own secure GPT tool, called the AI Playground.

Rule 3: Don’t let AI write the first draft

Start with your own words and ideas. AI can help polish later, but don’t use it to write content. Grant reviewers can often tell when language feels generic or disconnected from your own voice. Your grant must reflect you as a scientist — your scientific ideas, your preliminary data, and your novel approach, described in your own words.

Rule 4: Be specific in your prompts

The more detailed prompt that you provide AI, the better the feedback. Don’t forget to include what grant you’re applying for and what section you want help with. Ask it to check for tone, clarity, or grammar—not to rewrite. The researchers also found that it's better at providing feedback on shorter sections.

• Example prompt: “I’m a postdoctoral scholar at Stanford writing a Career Development Award for the American Heart Association. Here’s a draft of my four training goals for the Career Development Plan section. Can you help me improve the clarity and make sure it aligns with the American Heart Association’s Career Development Plan  review criteria? Please focus on structure and tone, not just rephrasing everything.”

Rule 5: Double-check everything

AI can make up fabricated sources, citations, and even data. Always verify every fact, reference, and claim that it generates. If you're using AI to brainstorm ideas, remember to fact-check before you write anything into your final draft.

Rule 6: Use AI as a helper, not a copy-paste machine

AI-generated content may contain plagiarism or bias. Use it to brainstorm edits, not as a replacement for your writing. Keep your voice and expertise front and center.

Rule 7: Learn from the process

AI isn’t just a shortcut — it can help you become a stronger grant writer. Each time you receive a suggestion, ask yourself why it works. Does it improve clarity? Strengthen your argument? Make your language more concise? Use this back-and-forth as a way to sharpen your own writing instincts.

Rule 8: Try AI to brainstorm visual ideas

You can use image-generating AI tools to sketch out figure ideas or get feedback on your visuals. (Just make sure the images are accurate, and that using AI-generated figures is allowed by the funder.)

Example tools:

1. DALL·E 3 (via ChatGPT Pro)

2. Midjourney (requires Discord)

3. Adobe Firefly

• Example prompt: "Create a simple conceptual figure showing a timeline of clinical trial phases from early discovery to implementation, including key milestones like IRB approval and patient recruitment.”

Rule 9: Human feedback still matters

Despite the power of AI, there are many limitations when it comes to grant writing. Human colleagues and mentors can catch things that AI is not yet capable of doing, such as understanding scientific logic, cultural tone, or strategic fit. Combine AI support with human insight for best results.

Rule 10: Play around and experiment

Test different tools and prompts to see which one works best for you. Practice asking the same question in different ways. Use prompts on content you’re already familiar with to understand the specific AI model’s strengths and weaknesses.