AI Grant Writing Assistant for Nonprofits

AI Prompts to Build This

Copy these build prompts into Claude, Cursor, or your AI coding tool. Create a free account to unlock the full research behind them.

1. 1. Project Setup

   Create a Next.js 15 App Router + TypeScript + Tailwind project called "GrantDraft" for nonprofit grant writing. Provision Supabase with these tables:
   
   - organizations(id, name, mission_statement, ein, budget_range, primary_contact_email, plan)
   - programs(id, org_id, name, description, target_population, annual_budget)
   - impact_metrics(id, org_id, program_id, metric_name, value, year, source_note)
   - funders(id, org_id, name, funder_type ENUM[foundation, government, corporate], priorities JSONB, format_spec JSONB, next_deadline)
   - proposals(id, org_id, funder_id, type ENUM[loi, full_proposal], status ENUM[drafting, submitted, funded, declined, no_response], draft_content TEXT, submitted_at)
   - outcomes(id, proposal_id, result, amount_requested, amount_awarded, decided_at, notes)
   
   Enable row-level security so every row is scoped to organizations the authenticated user belongs to. Wire Stripe with four products: Free, Starter ($39/mo), Growth ($89/mo), Agency ($199/mo). Install pgvector as a Supabase extension and add an embeddings column to programs and impact_metrics for retrieval.

2. 2. Funder-Format Parser + Draft Generator

   Build the core drafting feature: parse a funder's RFP text into a structured format spec, then generate a matched draft.
   
   Step 1 — Parse the RFP:
   Accept pasted RFP/LOI guideline text. Send to Claude with a strict JSON schema output:
   { sections: [{ name: string, max_words: number|null, required: boolean, guidance: string }], overall_page_limit: number|null, required_attachments: string[], stated_priorities: string[], tone_notes: string }
   Store this as funders.format_spec.
   
   Step 2 — Retrieve org context:
   Given the funder's stated_priorities, run a pgvector similarity search over this organization's programs and impact_metrics embeddings to pull the 5-10 most relevant facts (program descriptions, outcome numbers, budget figures).
   
   Step 3 — Generate the draft:
   Prompt Claude: "You are drafting a grant [proposal/LOI] for [org_name] to [funder_name]. Follow this exact section structure and word limits: [format_spec.sections]. Use ONLY the following verified organization facts — never invent statistics or outcomes: [retrieved facts]. Match the funder's stated priorities: [stated_priorities]. Write in a confident, specific, non-generic nonprofit voice." Return one draft per required section, respecting each section's max_words. Flag any section where the org's available data is too thin to answer specifically, rather than filling with generic filler language.
   
   Persist the draft to proposals.draft_content and show word counts against each section's limit in the editor UI.

3. 3. Deadline Tracker + Outcomes Loop

   Build the deadline tracking and outcomes-feedback system.
   
   Deadline tracker:
   - Calendar view of all funders.next_deadline across the org, color-coded by days remaining (red under 7 days, yellow under 21, green otherwise).
   - Vercel Cron job runs daily, queries proposals with status='drafting' or funders with upcoming next_deadline within 14/7/3/1 days, and sends a Resend digest email to the org's contacts.
   
   Outcomes loop:
   - When a proposal's status changes to 'funded' or 'declined', prompt the user for amount and a short free-text reason (e.g. "budget didn't match our program size," "board wanted more community partnerships").
   - Store in outcomes. Build a simple dashboard: win rate by funder_type, win rate by proposal length, and a ranked list of which stated_priorities correlate with funded outcomes for this org.
   - Surface this on the funder-matching screen: when a new funder is added, show "similar funders you've won with" and "similar funders that declined you, and why" pulled from the outcomes table, so staff can decide whether a funder is worth the time before drafting.