Straight answer: ChatGPT, Claude and Perplexity are the workhorses for drafting, structuring and researching grant applications, while tools like Instrumentl and Grantable exist specifically to match you with funders and manage the paperwork. None of them will win you a grant on their own because reviewers can spot generic AI language a mile off, and I’ve watched that exact problem sink an application.
Why I got pulled into this in the first place
I’m not a grant writer. I’m an AI and marketing consultant. But a friend who runs a small arts education charity in Bristol asked me to look over an application she’d drafted with ChatGPT before she submitted it to a local authority fund worth £12,000. She’d spent three evenings on it and was proud of how fast it came together.
It read like every other charity application in the pile. Phrases like “we are committed to fostering meaningful engagement within the community” and “our innovative approach seeks to empower diverse stakeholders” appeared four times in two pages. It was smooth, grammatically perfect, and completely forgettable. We rewrote it together, pulled in her actual numbers (47 children on the waiting list, 3 volunteer tutors, one specific story about a boy called Reece who’d gone from refusing to read aloud to volunteering for the school assembly), and cut the AI-flavoured filler by about 60%. She got the funding. The tools didn’t fail her. The way she used them did.
The tools that earn their place
Here’s what I’d point people toward, broken down by what each one is good at.
ChatGPT (GPT-5, paid tier)
Best for turning a messy brain-dump into a structured first draft. If you talk through your project for ten minutes into a voice memo or paste rough notes, ChatGPT can organise that into the standard sections a funder expects: need, approach, outcomes, evaluation, budget justification. I use the same technique for client proposals, and the same trick I mentioned in my roundup of AI news for small business applies here too: the newer models are noticeably better at following a specific word count than they were a year ago, which matters when a funder caps your answer at 300 words and means it.
Claude (Anthropic)
Better than ChatGPT at holding a long, detailed brief in its head without losing track of your organisation’s specific facts. I’ve found it handles multi-page grant guidelines better, meaning fewer moments where it invents a criterion the funder never mentioned. If you’re applying to something like the National Lottery Community Fund or an Arts Council England grant, both of which run to dozens of pages of guidance, paste the whole guidance document in and ask Claude to extract exactly what’s being scored against, before you write a single word.
Perplexity
This is your research tool, not your drafting tool. Use it to find comparable funded projects, sector statistics you can cite, or the actual named priorities of a funder this year rather than three years ago. Funder priorities shift constantly and outdated stats in your application are an instant red flag to an experienced reviewer.
Instrumentl
A paid platform (plans start around $179 a month) built specifically for US nonprofits to find matching grants and track deadlines. It’s not an AI writing tool in the ChatGPT sense, but it does use matching algorithms to surface funders you’d never have found by googling, and that alone can be worth the subscription if you’re applying for multiple grants a year.
Grantable
A newer tool built around the idea of a “response library”: you feed it your past successful answers and it suggests reused, adapted content for new applications. useful if you apply to a lot of similar funders (say, five different local council pots a year with overlapping questions), less useful for a one-off big application where every answer needs to be bespoke.
Otter.ai or similar transcription tools
Undervalued in this whole conversation. The best grant content usually comes from a spoken conversation, not a blank page. Get your programme lead or a beneficiary on a call, record it, transcribe it, and feed the transcript to ChatGPT or Claude as raw material. It captures specific detail (names, numbers, moments) that nobody produces when staring at a form.
A step-by-step process that works
This is roughly the process I now recommend to small charities and small businesses applying for grants:
- Step 1: Paste the full funder guidance into Claude or ChatGPT and ask it to list every scoring criterion and word limit, in order.
- Step 2: Record a 10 to 15 minute voice conversation about the project, including at least one specific example, number, or named person. Transcribe it.
- Step 3: Feed the transcript and the criteria list into the AI tool and ask for a rough first draft matched to each section and word limit.
- Step 4: Read the draft out loud. Cross out every sentence that could apply to any organisation applying for any grant. If a sentence would still make sense with your charity’s name swapped for a random competitor’s, it’s not specific enough and it goes.
- Step 5: Add back in the real detail: exact figures, a named beneficiary story, your actual evaluation method, not “we will measure impact through evaluation” but “we will track reading age using the York Assessment before and after the six-week programme.”
- Step 6: Run the final draft through a second AI tool as a fresh reviewer, asking it specifically “does this sound generic or specific, and where?” It’s a decent proxy for a tired reviewer reading application number 40 of the day.
That whole process took my friend about four hours the second time round, compared to the three evenings she’d spent before. The time saving is real. It’s the quality of what goes in at Step 2 and Step 5 that determines whether it works.
The uncomfortable bit nobody selling these tools wants to say
Grant reviewers, especially the ones who read hundreds of applications a year for the same fund, have started noticing AI writing patterns. Certain phrases (“holistic approach”, “meaningful impact”, “diverse stakeholders”, “innovative solution”) now function almost like a tell, the same way a poker player has a tell. Several funders I’ve spoken to informally, including one programme officer at a UK regional arts fund, have told me they now specifically discuss AI-flavoured applications in scoring meetings, not to disqualify them automatically but to flag that the application probably needed more human editing before submission. That’s not a rumour, that’s a shift in how these panels talk about what lands in front of them.
The tools are not the problem. Submitting the first draft as the final draft is the problem. AI is very good at structure, grammar, and speed. It is not good at conviction, and grant panels are, in the end, funding conviction and evidence, not prose.
Where the budget and evaluation sections need different tools
Most people forget this part until the last minute: the numbers section of a grant application often carries as much weight as the narrative, and it’s where AI is weakest without your input. ChatGPT can help you structure a budget table and check your maths, but it cannot invent realistic costs for your region. If you’re budgeting for freelance staff time, get real quotes rather than letting the AI suggest a plausible-sounding figure, because funders who work in a sector do know roughly what things cost, and a wildly wrong number damages trust in the whole application.
For the evaluation section, ask yourself what you’ll measure before you ask any AI tool to write about “measuring impact.” If your real plan is a feedback form and a headcount, say that plainly rather than dressing it up as “a comprehensive mixed-methods evaluation framework.” Reviewers who’ve read that phrase five hundred times will trust the plain version more.
When it’s worth getting outside help instead
If you’re applying for grants regularly, several times a year, at a value where a professional grant writer’s fee (typically 5 to 15% of the award, or a flat fee of £500 to £2,000 for a single application in the UK) makes financial sense, it’s worth weighing that against doing it yourself with AI tools. There’s also a middle path a lot of small organisations don’t think about: bringing in someone to set up a proper AI-assisted workflow and template system once, rather than paying per application forever. That’s the kind of one-off setup work I cover in more detail on my page about what it costs to work with an AI consultant, and it’s usually far cheaper than a single grant writer’s fee if you’re going to be applying repeatedly.
If you don’t have the time to run the process yourself, a good virtual assistant, briefed, can manage the deadlines, the funder research and the first-pass formatting for you, freeing you up to focus on the actual conviction and voice, which is the part no AI tool can fake. I go into what a VA can realistically take off your plate, including for admin-heavy work like this, in my complete guide to what a virtual assistant does.
Small habits that make the AI output better
A few things I’ve noticed make a real difference to output quality, none of which are complicated:
- Give the AI your actual past successful application (if you have one) as a style reference rather than a generic prompt about “professional tone.”
- Ask for three different versions of the opening paragraph and pick the least AI-sounding one, not the most polished one.
- Never let the tool write your closing “why fund us” paragraph from scratch. Write that one yourself, badly if needed, then ask AI to tidy the grammar only.
- Keep a running document of specific facts, quotes and numbers about your organisation so you’re never starting from a blank brain when you sit down to draft.
If you want to build broader habits around using AI well day to day, not just for grant writing, I’ve written before about which tools are worth your attention generally in this list of apps for sharpening how you think and work, and several of the underlying habits (structured prompting, treating AI as a first draft not a final one) apply directly to grant writing too.
What I’d tell you if you only remember one thing
Use AI for speed and structure. Use yourself, or someone who knows the project intimately, for the parts that make a reviewer believe you. The charities and small businesses I’ve seen get funded aren’t the ones with the smoothest prose. They’re the ones whose applications sound unmistakably like a real project run by real people who know exactly what £12,000, or £50,000, or £200,000 would let them do.
Frequently asked questions
Can AI write a whole grant application for me?
It can produce a full draft quickly, but submitting that draft unedited is a mistake because experienced reviewers now recognise generic AI phrasing, and it reads as less credible than a rougher, more specific human draft.
Is ChatGPT or Claude better for grant writing?
ChatGPT is generally faster for turning notes into a structured first draft, while Claude tends to handle long, detailed funder guidance documents more accurately, so many people use both at different stages of the same application.
Are paid grant-specific tools like Instrumentl worth it?
They’re worth it mainly for funder discovery and deadline tracking if you’re applying for several grants a year, typically running $179 a month and up, but they don’t replace a general AI tool for the actual writing.
What’s the biggest mistake people make using AI for grants?
Treating the first AI draft as the final one, which leaves in generic phrases like “innovative” and “holistic approach” that experienced reviewers have read hundreds of times and now read as a warning sign rather than a strength.
Primary sources
Related reading: AI Consultant for Charities and Nonprofits: Grant Writing and Admin That.
