Finance & Admin
Using AI to Prepare Grant and Funding Applications
US$49 a month. That is the published starting price, in a tool roundup checked on 2026-10-25, for one dedicated grant-writing tool — and it is close to the only hard number in this whole topic. Everything else about preparing a funding application with AI is a workflow question, and the workflow has a strict order: verified facts first, the funder’s rules second, drafting third. Reverse that order and the AI will happily draft an application around figures nobody has checked.
Disclaimer: This article is general information only. It is not funding, financial, or professional advice, and it does not assess your eligibility for any programme. Grant and funding programmes set their own criteria, deadlines, and rules on AI use — check the funder’s current guidance for any programme you apply to, and certify only information you have verified yourself.
The numbers first: three routes, costed as far as sources allow
| Route | Sourced cost | Arithmetic, labelled as a scenario | Where it wins |
|---|---|---|---|
| Dedicated grant tool (Grantable class) | Plans start around US$49 per month for small teams (roundup checked 2026-10-25) | One application prepared in a single month: about US$49. Left running for twelve months: 49 × 12 = US$588 — the scenario that punishes forgetting to cancel | Repeated applications, where a stored fact base is reused |
| Free document tool (NotebookLM class) | Free, per the same roundup | US$0 in subscription; the cost is entirely your review time | A single application, when you upload both the request for proposals and your previous grant materials — the combination the roundup says it works best with |
| Custom-priced tool (TurboInnovate class) | Pricing is custom, per the same roundup | No honest per-application figure can be stated without a quote | Only after a quote makes the comparison possible |
| Human grant writer | No sourced rate was available for this comparison | Not costed here — a price invented to complete a table would be worse than a gap | High-value or complex applications where certification risk is highest |
The kill criterion belongs here, before any drafting: if a subscription route will be used for one application only, set the cancellation date the day you subscribe. US$49 for a month of preparation is a tool cost; US$588 for a forgotten subscription is a leak. And if the funder’s own rules, checked in step two below, do not permit the way you planned to use the tool, stop — no saving justifies an application a funder would reject on process grounds.
Step one: build the verified fact base
- Assemble the facts only you can certify. Eligibility facts, financial figures, and evidence of need go into one document, each checked against its source record. This division of labour is the core finding of the guidance on AI grant writing: AI is best at organising factual material, drafting sections, checking consistency, and identifying questions that need a human answer.
- Mark what the AI must never supply. The same guidance is blunt about the other side: AI cannot reliably invent eligibility, financial figures, programme impact, or evidence of need. Treat any such content appearing in a draft as an error to remove, not a gap it helpfully filled.
- Remember who certifies. Many funders require applicants to certify that submitted information is accurate. That signature is the owner’s, and it covers every sentence — including the ones a tool drafted.
Step two: check the funder’s AI rules before drafting
- Read the funder’s own guidance on AI use, first. Funders publish expectations. One funder’s guide to applicants, for example, instructs applicants to write the initial draft themselves and then use AI to edit and polish — the reverse of the workflow most tool marketing assumes.
- Plan the disclosure. That same guidance asks applicants to be transparent about which tools they used and how. Note the tools and their role as you work, so disclosure is a record, not a reconstruction.
- Adjust the workflow to the rules, not the other way round. If the funder expects a human-written first draft, the AI’s job shrinks to editing, consistency checks, and rubric scoring. The runbook below still works — the drafting step simply changes hands.
Step three: draft in sections, against the rubric
- Load the context. Give the tool the call document and your previous application materials together — the combination identified above as the one the free document tool works best with — plus the verified fact base from step one.
- Draft one section at a time. Each section is drafted only from the fact base. Where the fact base has no answer, the correct output is a flagged question for a human, not a sentence.
- Check consistency across sections. Figures, dates, and claims must agree everywhere they appear. Consistency checking is one of the tasks the guidance assigns to AI — use it there, where an error is mechanical and catchable.
What success looks like: a complete draft in which every figure traces back to the fact base, every section answers the criterion it was written against, and the list of flagged human questions is empty because each one was answered from a record — not because the tool stopped asking.
Step four: self-score, then the human pass
- Score the draft against the funder’s criteria. Work criterion by criterion: what evidence does this section offer, and where does it sit in the text? Weak or missing evidence goes back to the fact base, not to more fluent drafting.
- Do the final human read as the certifier. Read the submission as the person who will certify its accuracy — because that is who the owner is. You are accountable for the submission; that accountability is stated in the funder guidance and cannot be delegated to a tool.
- Apply the decision framework for the next application. One application, tight deadline: the free route plus owner time. A pipeline of applications across the year: the dedicated tool’s stored fact base starts to justify its monthly cost. A funder whose rules restrict AI drafting: human draft, AI polish, whichever route supplies the polish.
What never goes into the prompt
Keep sensitive or confidential data out of open AI platforms — the funder guidance states this directly, and it applies to financial records, personal data about staff or beneficiaries, and anything a funder or a client shared in confidence. Draft from the verified fact base, which contains the facts the application needs, rather than uploading raw records and hoping the tool ignores the rest. The fact base is not only an accuracy control. It is also the boundary that decides what the tool ever sees.
Reminder: this article is general information only, not funding or financial advice. Programme criteria, deadlines, and AI-use rules are set by each funder and change — verify them with the funder before you apply.
— Sterling Cole, Money Writer — Earning Desk