ChatGPT prompts for grant applications that get read correctly by a panel

The short version: ChatGPT can draft 70 to 80 percent of a grant application in a fraction of the time it takes by hand, but only if you feed it your own numbers, your own beneficiaries, and your own budget lines first. Used blind, it writes fluent nonsense that a panel rejects in under two minutes. Used with the prompts below, it becomes a fast first draft you edit down, not a source of facts.

Why ChatGPT is useful here, and where it fails

Grant panels read hundreds of applications a year and most of them sound the same: vague need, vague outcomes, a budget that does not add up to the ask. ChatGPT is good at structure. It can turn a messy pile of notes into a needs statement, an outcomes framework, or a budget narrative in the right order and the right tone. It is bad at facts. It will invent statistics, invent named studies, and invent numbers if you let it. Every figure it produces has to come from you, or from a source you paste in, never from the model's own memory.

The prompts below are grouped by the sections a typical grant application asks for: the project idea, the needs statement, goals and objectives, the budget narrative, and the executive summary. Paste your own facts into every bracket before you send a prompt.

Prompts for shaping the project idea

• Prompt: "I run a [type of organisation] serving [population] in [location]. We want funding for [rough idea]. Ask me ten questions that a grant panel would want answered before I write a word, one at a time, waiting for my answer each time."
  What to change: the organisation type, population, and rough idea. This prompt stops the model guessing and forces it to interview you, which surfaces gaps before you waste time on a full draft.
• Prompt: "Here are my answers to your ten questions: [paste answers]. Turn this into a one paragraph project summary of no more than 120 words, written for a funder who reads 200 applications a month and skims the first line."
  What to change: paste your own answers, not a summary of them, so nothing gets lost.
• Prompt: "List five ways this project could fail in year one, based on this summary: [paste summary]. For each, give one sentence on how we would mitigate it."
  What to change: nothing but the summary. Funders increasingly want to see risk awareness, not just optimism.
• Prompt: "Rewrite this project summary so a funder with no background in [sector] understands it on the first read: [paste summary]."
  What to change: the sector name. Use this when your first draft is full of jargon only insiders understand.

Prompts for the needs statement

• Prompt: "Here is our raw data on the problem we are solving: [paste stats, waiting list numbers, survey quotes]. Write a needs statement of 250 to 300 words that opens with the sharpest number, not a general claim."
  What to change: the raw data. Never let the model supply the number itself.
• Prompt: "Take this needs statement and add one direct quote from a beneficiary, marked clearly as a placeholder for me to replace with a real quote: [paste needs statement]."
  What to change: nothing, but replace the placeholder with a real quote before you submit. A fabricated quote in a grant application is the fastest route to a rejection and a reputation problem.
• Prompt: "Compare this needs statement against this funder's stated priorities: [paste funder priorities]. Tell me which of my points map directly to their language, and which do not map at all."
  What to change: the funder's own wording, copied from their guidelines page, not paraphrased.
• Prompt: "This needs statement is 400 words. Cut it to under 250 without losing the strongest number or the strongest quote: [paste statement]."
  What to change: nothing but the pasted text. Most panels reward tight writing over comprehensive writing.

Prompts for goals and objectives

• Prompt: "Turn this goal into three measurable objectives using specific numbers and a timeframe: goal is [paste goal]. We currently serve [current number] people and want to reach [target number] by [timeframe]."
  What to change: the goal, current number, target number, and timeframe. Funders fund objectives, not aspirations.
• Prompt: "For each of these three objectives, suggest one way we could measure it that does not need new software or a big budget: [paste objectives]."
  What to change: nothing but the objectives list. Cheap, believable measurement beats an ambitious plan you cannot run in practice.
• Prompt: "Check these objectives against the SMART framework and tell me exactly which letter each one fails on, if any: [paste objectives]."
  What to change: nothing. Use this as a final check before submission.

Prompts for the budget narrative

• Prompt: "Here is our line item budget: [paste budget lines and amounts]. Write a budget narrative that explains each line in plain language, in the order the lines appear."
  What to change: the actual budget. This is the section panels scrutinise hardest, so accuracy matters more than style here.
• Prompt: "This grant caps administrative costs at [percentage] of the total ask. Here is our budget: [paste budget]. Tell me which lines count as administrative under a typical funder definition, and whether we are over the cap."
  What to change: the percentage and the budget. Then verify the funder's own definition in their guidelines, because "administrative" is defined differently by different funders.
• Prompt: "Write one sentence justifying each of these staff time allocations, based on the tasks they do day to day: [paste staff roles and hours]."
  What to change: the roles and hours. Vague justification for staff costs is one of the most common reasons a budget line gets queried or cut.

Prompts for the executive summary and cover letter

• Prompt: "Using only the sections I have already written, draft a 150 word executive summary. Do not add any claim, number, or outcome that is not already in the text I am about to paste: [paste full draft]."
  What to change: nothing but the full draft. This constraint stops the model inventing a stronger claim than the body supports.
• Prompt: "Write a two paragraph cover letter to [funder name], referencing our past relationship with them if any: [describe relationship or write 'none']."
  What to change: the funder name and the relationship detail. A cover letter that names a real prior grant or conversation reads far better than a generic one.

A weak prompt versus a strong prompt

Weak: "Write me a grant application for a youth mentoring charity."

This gives the model nothing to work with. It will invent a charity name, invent statistics about youth mentoring outcomes, invent a budget, and hand you 800 words of confident fiction that sounds like every other application a panel has read that year.

Strong: "I run Bright Futures, a youth mentoring charity in Leeds. We currently have 40 young people on a waiting list and mentor 60 at a time, with a 6 month average wait. Our funding ask is 18,000 pounds to hire one part time coordinator at 15,000 pounds and cover 3,000 pounds of training costs. Write a 250 word needs statement using only these numbers, and flag with [NEED SOURCE] anywhere you think a claim needs a citation I have not given you."

The strong version gives real numbers, a real ask, and a real instruction to flag anything unsupported. That single instruction, telling the model to mark gaps instead of filling them, is the difference between a draft you can trust and one you have to fact check line by line.

Mistakes that get applications rejected

• Letting the model invent statistics. If you did not give it a number, it made one up. Read every figure back against your own source before it goes in the application.
• Submitting the same generic paragraph to five funders. Panels notice when a summary does not mention their name, their priorities, or their region. Rerun the funder priority comparison prompt for every single application.
• Ignoring the word count and page limit. ChatGPT will happily write 500 words when the form allows 250. Always paste the funder's exact limit into the prompt and ask for a hard cut afterwards.
• Copying budget narrative language that does not match the budget spreadsheet. If the narrative says "one full time coordinator" and the spreadsheet shows a part time salary, that mismatch is the first thing a reviewer flags.
• Skipping the human read through. Read the whole thing aloud once before you submit. AI drafted text often has a rhythm that repeats itself, and your own ear will catch it faster than another prompt will.

I have written more around this on the site: AI Consultant for Charities and Nonprofits: Grant Writing and Admin That Doesn’t Eat Your Week, How Much AI Writing Is Acceptable in a Research Paper?, What Are the Best AI Assisted Writing Tools for Small Business Owners.

If you want the full set, start at the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.

I built a free prompt builder for exactly this.

Related: writing for us on nonprofit.

Frequently asked questions

Can a funder tell an application was written with ChatGPT?

Some can, especially when the phrasing is generic and repeats the same sentence patterns across paragraphs. The fix is not to avoid the tool, it is to feed it your own specific numbers, quotes, and context so the output reads like your organisation rather than a template. A heavily edited AI draft with real detail is indistinguishable from one written from scratch.

Is it against the rules to use ChatGPT for a grant application?

Almost no funders ban the use of AI tools outright, but a growing number ask applicants to declare if AI was used to draft significant sections. Check the specific funder's guidelines each time, because policies differ and change.

How much editing does an AI grant draft need?

Plan for at least one full editing pass per section, checking every number against your own records and every claim against something you can prove. Treat the first output as a structural draft, not a finished one.

Should I paste the funder's guidelines directly into ChatGPT?

Yes, this is one of the most useful things you can do. Pasting the actual priorities and eligibility language lets the model check your draft against the funder's own words rather than a generic idea of what grant writing should sound like.

What is the biggest time saving compared to writing without AI?

Most of the saving comes from structure, not final wording. A first draft that would take three or four hours by hand can be scaffolded in twenty minutes, leaving your remaining time for fact checking, budget accuracy, and the human touches a panel responds to.