Ask around your department and you will hear it. A PI used ChatGPT to tighten her aims page. A postdoc drafted a biosketch narrative in Claude. A new investigator asked Gemini to summarize a NOFO before deciding whether to apply.
None of this went through your office. Most of it never will.
Faculty are not waiting for a policy. They are under pressure to submit more, faster, with less support, and AI is open in the next browser tab. The real question for research offices is no longer whether AI will touch your institution's proposals. It already does. The question is whether it happens in a tool you can see and stand behind, or in a dozen tools you cannot.
The Two Default Responses, and Why Neither Works
Response 1: Ban It
Some offices have tried to discourage AI use in proposals entirely. The instinct is understandable. NIH has drawn a clear line on originality, and no one wants their institution's name on a misconduct referral.
But bans have three problems.
- They are not enforceable. You cannot see what a PI does on a personal laptop at 11 p.m. before a deadline.
- They push use underground. Faculty who would have asked for guidance now just do not mention it.
- They leave the time problem unsolved. The workload that drove faculty to AI in the first place is still there.
A ban does not remove AI from your proposals. It removes your visibility into it.
Response 2: Let Everyone Pick Their Own Chatbot
The opposite approach is to allow AI use with a general caution. This keeps faculty productive, but it hands every decision to individuals using general-purpose tools that were never designed for grants.
- No consistency. Ten PIs using ten tools produce ten different approaches, and ten different levels of quality.
- No structure. A blank chat window will happily "write the aims" if asked, which is exactly the use NIH warns against.
- No connection to your process. Chatbots do not know your routing, your internal deadlines, the NOFO requirements, or the SF424. Your staff still catch everything at the end.
- No shared view. Your office cannot see where a proposal stands until it lands on your desk.
Scattered chatbots do not make your office's job easier. They move the risk and the cleanup downstream, to you.
What NIH NOT-OD-25-132 Actually Requires
It helps to be precise about the rules, because they are narrower than many faculty assume.
Under **NOT-OD-25-132**, "Supporting Fairness and Originality in NIH Research Applications," NIH will not consider applications, or sections of applications, substantially developed by AI to be the original ideas of the applicants. The same notice limits each PI to six new, renewal, resubmission, or revision applications per calendar year.
Two things follow for research offices.
- First, the standard is originality, not abstinence. NIH has not banned AI in application preparation. It has said the ideas must belong to the investigators. The governing question is how AI is used, not whether.
- Second, every application now counts more. With a six-per-year cap, a weak or incomplete submission costs a PI one of a small number of chances. Quality per proposal matters more than volume.
A sanctioned tool should be built around both points.
The Third Option: A Governed Platform Like ResearchGrants AI
Between a ban and a free-for-all is a better answer: give faculty a sanctioned tool that is more complete than any chatbot and designed to keep the science in their hands.
That is what ResearchGrants AI is built to be. Here is what a governed option looks like in practice.
1. Human First, AI Augmented
The platform starts from the researcher's own thinking: background, rationale, preliminary data, and early ideas for the aims. It provides structure, drafting support, and review on top of those inputs. Specific Aims scaffolding organizes the argument; it does not invent it.
That design keeps faculty on the right side of the originality line by default, rather than relying on each PI to police their own prompts.
2. The Whole Proposal, Not Just the Paragraphs
A chatbot helps with sentences. An R01 needs far more. ResearchGrants AI covers the full proposal in one place:
- Opportunity discovery with GrantMatch and NOFO summaries
- Parsing of preliminary data for Research Strategy figure use
- NOFO-optimized Specific Aims draft, including iteration tool
- SciENcv-ready biosketches and supporting attachment draft generation (if not uploaded)
- A Research Strategy draft that builds on the aims, preliminary data and figures, and uploads
- Easy to use budget tool and budget justification drafting
- Completeness, consistency, and strength checker
- SF424 build and review
- Submission as a downloaded package or routed for approval and S2S submission
When faculty have a tool that does the whole job, they have less reason to go looking for others.
3. Formatting and Compliance Built In
Page limits, required components, and NOFO requirements are part of the workflow, not a checklist your staff run at the deadline. Fewer proposals arrive needing last-minute rescue.
4. Visibility Your Office Does Not Have Today
Each proposal has a dashboard that shows completeness at a glance. Real-time collaboration means your staff and the researcher can work in the same proposal, instead of trading attachments by email. Problems surface weeks earlier, not days.
5. It Fits Your Process, Wherever a Proposal Starts
Faculty can start a new proposal, upload one already in progress, bring in a prior submission, or upload reviewer comments for a resubmission. The platform fits into your existing pre-award process rather than replacing it.
Ban vs. Chatbots vs. Governed Platform
| Feature | Ban | Scattered chatbots | ResearchGrants AI |
|---|---|---|---|
| Faculty actually comply | Rarely | Yes, but unevenly | Yes |
| Office visibility | None | None | Per-proposal dashboard |
| Built around NIH originality | Relies on avoidance | Relies on each PI | Human-first by design |
| Covers the full R01 | No | Writing only | End to end, through SF424 and submission |
| Formatting and compliance | Manual | Manual | Built in |
| Consistent quality | No change | Varies by tool and user | Same structure for every PI |
| Saves office time | No | Often adds cleanup | Yes |
Questions to Ask Any AI Tool Before You Sanction It
Whether you choose ResearchGrants AI or another option, a sanctioned tool should hold up to these questions:
1. Does it start from the researcher's own ideas and data, or generate content from a copied application (whether the researcher's own or not)?
2. Does it cover the full application, or only the writing?
3. Does it build NOFO requirements, page limits, and required components into the workflow?
4. Can your office see proposal status and collaborate in real time?
5. How does it protect institutional data, unpublished results, and confidentiality?
6. Does it fit your existing routing and submission process?
If a tool cannot answer these well, it is just another chatbot with a login.
The Bottom Line
Your faculty are going to use AI. The choice in front of your office is where.
A ban leaves you blind. A patchwork of chatbots leaves you cleaning up. A governed platform gives faculty something better than what they are using now, keeps the science in their hands, and gives your office the visibility and consistency it has never had.
The grant process is automated. The writing is not. Until now.
About ResearchGrants AI
ResearchGrants AI is the most complete, end-to-end grant development platform for NIH and other government applications. It takes a proposal from opportunity search through Specific Aims, Research Strategy, biosketches, budget, SF424, and submission in one connected workflow, with support for resubmissions and every mechanism from R21s and K awards to R01s, SBIR/STTR, and P01s. Built on a human first, AI augmented approach, it drafts from each researcher's own inputs so PIs can claim originality with confidence, while research offices collaborate in real time and track every proposal's completeness.
See how ResearchGrants AI fits your office's pre-award process. Contact us at researchgrants.ai
Build aims that hold up under review.
ResearchGrants AI guides you through structuring scientifically-sound Specific Aims — sharpening your hypothesis, rationale, and approach before a single word of the proposal is written.