AI for grant writing, with the funder's rules on the table

I am preparing an application to [FUNDER] under [PROGRAM]. Below are (1) the full solicitation text, (2) [number] public award abstracts, and (3) our program summary.

Using only that text:
1. List the funder's stated priorities, quoting the phrases they use.
2. For each priority, say which part of our program speaks to it, or write "no match".
3. Note where our summary uses different words from the solicitation.
4. List anything in the solicitation that says anything about artificial intelligence, with the quoted sentence.

Do not add facts about the funder that are not in the text I pasted.

Help me draft a needs statement for [FUNDER] / [PROGRAM] using only the data below. Each figure includes its source.

Population and geography: [..]
Data: [figure | source]
Our theory of change (my words): [..]
Funder priorities to reflect: [paste from stage 1]

Write 400 to 600 words. Do not invent statistics, quotes or anecdotes. Where a claim needs evidence I did not supply, write [NEEDS SOURCE] instead. Afterwards, list the two weakest claims and what evidence would strengthen them.

Here are the activities, outputs and outcomes I decided on for [PROGRAM]:

[paste your list]

Arrange them as a table: Inputs | Activities | Outputs | Short-term outcomes | Long-term outcomes. Do not add any new activity or outcome. Then list gaps: outcomes with no activity behind them, activities with no output, and outputs that no outcome depends on.

Attached is an export of our budget (CSV) and the funder's budget instructions.

For each line item, draft one or two sentences of justification in plain language. Then list every place where the narrative and the budget figures disagree, or where a funder instruction is not addressed. Do not do arithmetic for me: quote the figures as given and I will recompute the totals myself.

Attached are (1) the complete solicitation and (2) our draft application.

Build a table with the columns: Requirement | Where the solicitation states it | Where the draft addresses it | Status (Met, Partial, Missing). Include page limits, formatting rules, required sections, attachments, certifications, and anything about AI use. Quote the solicitation's wording in the second column. List Partial and Missing items at the end.

NIH notice NOT-OD-25-132, released July 17, 2025 and effective for applications submitted to the September 25, 2025 receipt date and beyond, says NIH will not consider applications that are substantially developed by AI, or that contain sections substantially developed by AI, to be the original ideas of applicants. It adds that AI may be appropriate for limited aspects of preparation and warns of plagiarism and fabricated citations. The notice does not define substantial, so ask your program officer or research office if you are unsure where the line sits.

In its December 14, 2023 notice, NSF says proposers are encouraged to indicate in the project description the extent to which generative AI was used and how. That is encouragement, not a blanket requirement in that notice. NSF also says it may publish further guidelines, so check the current PAPPG and your solicitation before you submit.

We found no general AI rule on the DOL ETA grants overview page, and foundation rules differ. Where a rule exists it will be in the funding announcement or the foundation's guidelines. Search those documents for artificial intelligence and generative, and ask the program officer if they are silent.

That depends on the funder's rules and on honest authorship. The NIH notice itself says AI use may result in plagiarism, fabricated citations or other research misconduct. Our rule of thumb: be able to explain and defend every section, open every source you cite, and follow whatever disclosure the funder asks for.

NIH's notice says it will continue to employ the latest technology in detection of AI-generated content. We cannot tell you how reliable any detector is, and this page does not recommend writing to avoid one. The safer frame is authorship: the ideas are yours, and the draft reflects them.

We have not measured it and do not publish a figure. Instrumentl's roundup, citing Charity Science, gives 15 to 20 hours for a foundation grant and up to or over 100 hours for a federal grant, which is only a rough baseline. Time one stage yourself, with and without AI, including the time you spend checking.

We have not published a ranked comparison, so we will not name a winner. Run the same inputs through two or three models in one ZeroTwo chat and score them against your own checklist: stays within your facts, marks missing sources, needs little rewriting.

The NIH notice applies to all activity codes except T codes and R13 conference grants, so NIH SBIR and STTR applications are covered by it. Other agencies set their own rules in their solicitations. Read the specific solicitation for AI language, page limits and required sections.

Anything you do not have permission to share. ZeroTwo's Privacy Notice says that using AI features may send your content to the model provider you select, that some providers may retain or use submitted data under their own terms, and that you are responsible for having permission to submit confidential or regulated information. NSF also prohibits reviewers from uploading proposal content they are reviewing to non-approved generative AI tools. Test with non-confidential text first.

ZeroTwo is a general chat workspace with many models, not a grant database. Grant platforms such as Instrumentl describe funder discovery and pipeline tracking as their focus; check their sites for current features. A reasonable split is a funder database for discovery and ZeroTwo for drafting support, checking and matrices.

ZeroTwo has a Free plan with 100 free credits to start and +20 bonus credits every day you log in; no card is required. Plus is $14.99/month and Pro is $29.99/month ($26.99/month billed annually).