Scholarship Program AI
Task Reference Guide

A practical reference mapping common scholarship administration tasks to single-prompt AI solutions and Retrieval-Augmented Generation (RAG) setups with custom instructions. Designed for NSPA member programs: platform-agnostic, privacy-first, human-oversight built in.

Key:
Single Prompt One-shot prompt, no external data needed
RAG Requires retrieval from program documents or knowledge base
Both Can work either way; RAG improves accuracy and consistency

Applicant Communications

Task Approach AI Automation Method Suggested Tools / Notes
Drafting award and congratulations letters Both Single prompt for one-off letters. RAG to enforce approved voice, program-specific language, and award conditions. "Draft a congratulations letter for a scholarship recipient. Tone: warm and encouraging. Mention the award amount [X], next steps for acceptance, and program contact. 3 short paragraphs. Do not include the applicant's name or ID." → RAG: Applicant Communications (templates + voice) Staff must personalize and verify award amount before sending. AI draft is a starting point only. Claude Project, Custom GPT
Mail merge integration (Word, Google Docs)
BoodleBox
Drafting declination / non-selection letters Both Single prompt from a compassionate template. RAG to apply program-specific language and ensure consistency across all declinations. "Draft a respectful, compassionate non-selection letter for a scholarship applicant. Do not cite specific reasons for the decision. Acknowledge effort, encourage reapplication if eligible. Under 150 words. Use second-person ('you') but do not include the applicant's name." → RAG: Applicant Communications (templates + voice) Never paste applicant names or ID numbers into a prompt. Staff approve every letter before sending. Claude Project, BoodleBox
Scholarship management platform (Submittable, Salesforce Scholarships)
Responding to common applicant questions RAG Ground responses in your program's eligibility rules, deadline FAQs, and requirements documentation so answers are accurate and consistent. → RAG: Program Eligibility & FAQ Knowledge Base Website chatbot (Tidio, Intercom)
Claude Project loaded with program docs
BoodleBox with program knowledge
Drafting waitlist notification letters Single Prompt Prompt for a clear, honest letter that explains waitlist status without implying guaranteed selection. "Draft a waitlist notification letter. Acknowledge the applicant's strong application. Explain they are on the waitlist and will be contacted if a position opens. Do not promise a specific timeline. Compassionate, direct tone. Under 120 words. No applicant identifiers." Staff review each draft before sending. Do not input applicant names or demographic details. Any LLM (Claude, ChatGPT, Gemini, Copilot)
Proofreading outgoing applicant-facing emails Single Prompt Paste draft (with any names or IDs removed); prompt the model to check tone, clarity, and equity of language. "Review this applicant communication for clarity, compassionate tone, and any language that could feel dismissive or inequitable. Flag passive-aggressive phrases, jargon, or assumptions about applicant background. Do not alter meaning." Claude, Grammarly Business, Copilot

Application Review & Scoring

Task Approach AI Automation Method Suggested Tools / Notes
Summarizing applicant essays for triage Single Prompt Paste a single, anonymized essay (remove name, school, location); prompt for a neutral factual summary only. Summary supports triage, not final decisions. "Summarize this scholarship essay in 3-4 sentences. Report only what the applicant wrote. Note the central theme, any specific examples given, and the stated goal. Do not evaluate quality, make inferences about the applicant, or add interpretation." Staff must read the original before scoring. AI summary is for orientation only, not a substitute for full review. Claude, ChatGPT, BoodleBox
Remove all PII before pasting
Checking essays against rubric criteria Both Single prompt when rubric is pasted inline. RAG when rubric lives in a knowledge base and should be retrieved automatically. "Using the rubric criteria below, identify which criteria this anonymized essay directly addresses and which it does not. Quote only brief phrases as evidence. Do not assign a score. Flag criteria where evidence is weak or absent. [Paste rubric. Paste anonymized essay.]" → RAG: Reviewer Knowledge Base (rubrics + scoring guides) AI analysis is a first-pass aid only. Reviewers make all scoring decisions independently. Claude Project, BoodleBox
Scholarship management platforms with AI integrations
Flagging potential completeness issues Single Prompt Prompt the model to compare a submitted application checklist against required components. No essay content or PII needed. "Compare this list of submitted components to the required checklist below. List any required items that appear missing or incomplete. Do not evaluate quality. Output as a simple checklist. [Paste required list. Paste submitted components list.]" Any LLM, Excel Copilot for batch review
Calibrating reviewer scoring consistency Single Prompt Paste anonymized score distributions (no applicant IDs); prompt the model to surface patterns, outliers, or potential calibration issues for staff review. "Review this anonymized reviewer score distribution table. Identify reviewers with scores more than 1.5 standard deviations from the group mean, and note any rubric categories with high inter-rater variance. Do not suggest any individual applicant outcome. Output a brief summary for program staff." Use aggregate data only. No individual applicant identifiers in this prompt. Claude, ChatGPT, Excel Copilot
Generating reviewer debrief summaries Single Prompt Paste anonymized reviewer panel notes after deliberation; prompt for a structured debrief summary for program records. "Summarize these anonymized reviewer panel notes into a structured debrief. Include: key themes noted across applications, rubric categories where the pool was strongest/weakest, and any process concerns raised. Do not reference individual applicants. Under 300 words." Claude, Otter.ai (for live meeting notes) + Claude for summary

Reviewer Onboarding & Support

Task Approach AI Automation Method Suggested Tools / Notes
Preparing reviewer orientation packets RAG AI assembles role-specific onboarding packets from the rubric, program policies, conflict-of-interest guidelines, and scoring instructions. → RAG: Reviewer Knowledge Base (rubrics + policies) Claude Project, BoodleBox
Loaded with current-cycle reviewer docs
Drafting reviewer welcome and instruction emails Single Prompt Provide reviewer role, cycle dates, platform name, and key instructions; model drafts a clear, welcoming onboarding email. "Draft a welcome email to a volunteer scholarship reviewer. Include: their role, the review window dates [X to Y], how to access the scoring platform [Platform Name], the conflict-of-interest policy reminder, and who to contact with questions. Friendly, professional tone. Under 200 words." Any LLM
Answering reviewer questions about rubric or process RAG Chatbot or Claude Project grounded in the current cycle's rubric, scoring guide, training materials, and FAQs. → RAG: Reviewer Knowledge Base Claude Project, BoodleBox
Creating reviewer training scenarios and calibration examples Single Prompt Prompt the model to generate synthetic (fictional) applicant essay excerpts that illustrate different rubric performance levels for training purposes. "Create three short synthetic scholarship essay excerpts (100 words each) illustrating three different performance levels on this rubric criterion: [paste criterion]. Label them Level 1, Level 3, and Level 5. Make them realistic but entirely fictional. Include a brief annotation explaining the score for each." Use only synthetic examples for training. Never use real applicant essays as calibration samples without explicit policy approval. Claude, ChatGPT, Gemini

Donor & Board Reporting

Task Approach AI Automation Method Suggested Tools / Notes
Drafting donor impact reports Both Single prompt when staff paste anonymized aggregate stats and key themes. RAG when donor history, fund-specific data, and prior report language live in a knowledge base. "Draft a one-page donor impact report section for a named scholarship fund. Use these anonymized aggregate stats: [awards made, total dollars, fields of study, geographic reach]. Write in a warm, narrative tone. Do not name individual recipients. Highlight impact themes only. ~250 words." → RAG: Donor & Board Reporting Staff verify all statistics against source data before sharing with donors. AI draft is a first pass. Claude Project, BoodleBox
Loaded with prior donor reports and fund summaries
Summarizing program outcomes for board packets Single Prompt Paste anonymized program-level data (totals, averages, demographics in aggregate); prompt for an executive narrative for governance review. "Summarize this scholarship cycle's aggregate outcome data for a board packet. Structure: Program Snapshot (3 sentences), Key Highlights (3 bullets), and Equity Indicators (2 sentences noting any demographic patterns visible in aggregate data). Audience: board members, not program staff. Under 200 words. Data: [paste aggregate stats only]." Claude, ChatGPT, Copilot in Word
Drafting thank-you letters from award recipients to donors Single Prompt Prompt for a template recipients can personalize, or draft a lightly personalized version from recipient-provided talking points (no PII in prompt). "Draft a scholarship thank-you letter template that recipients can personalize. Tone: genuine and specific, not generic. Structure: opening (gratitude), middle (what the award makes possible), closing (forward-looking). Provide [brackets] for recipient to insert personal details. Under 200 words." Any LLM
Writing narrative summaries of review cycle for stewardship Both Single prompt from a structured debrief notes input. RAG to align with prior cycle language and donor-specific fund guidelines. "Using these anonymized review debrief notes, write a 200-word narrative summary of this cycle's applicant pool and selection process for a donor stewardship report. Highlight pool strengths, committee themes, and selection criteria emphasis. Do not reference individual applicants." → RAG: Donor & Board Reporting Claude Project, BoodleBox

Program Operations & Content

Task Approach AI Automation Method Suggested Tools / Notes
Building and updating program FAQs Both Single prompt to draft FAQ from eligibility rules and policy docs pasted inline. RAG for ongoing updates that draw from a living program knowledge base. "Using the eligibility rules and requirements below, draft a 10-question FAQ for prospective scholarship applicants. Write questions as applicants would ask them. Keep answers under 60 words each. Plain language, no jargon. Flag any question where the answer is ambiguous in the source docs. [Paste program rules]." → RAG: Program Eligibility & FAQ Knowledge Base Claude Project, BoodleBox
Website CMS integration
Drafting program web page copy Single Prompt Prompt for GEO-optimized (answer-first) web copy from program details. Front-load eligibility, deadlines, and key facts so AI tools and search can surface accurate info. "Write a scholarship program web page. Lead with a one-sentence description of who is eligible. Follow with: award amount, application deadline, eligibility requirements (bulleted), how to apply (numbered steps), and a contact line. Plain language, scannable format. No fluff. Details: [paste program details]." Claude, ChatGPT, Copilot
Review against source documents before publishing
Summarizing reviewer comment themes for program improvement Single Prompt Paste anonymized open-text reviewer comments (remove any applicant identifiers); prompt for thematic analysis to inform next cycle. "Analyze these anonymized reviewer comments for recurring themes. Identify: (1) what reviewers found most compelling in strong applications, (2) common weaknesses noted, (3) any rubric criteria that generated confusion. Output 3-5 themes per category. Under 300 words total. Source: reviewer comment export with all applicant names and IDs removed." Claude, ChatGPT, NotebookLM
Drafting program policy documents and SOPs Both Single prompt for first drafts from staff notes. RAG to check alignment with existing program policies and governance requirements. "Using these staff notes, draft a standard operating procedure for [program task, e.g., conflict-of-interest recusal]. Format: Purpose, Scope, Step-by-step process, Escalation path. Plain language. Include an 'Effective Date' and 'Owner' field. ~300 words." → RAG: Reviewer Knowledge Base (policy alignment) Claude Project, Notion AI, Google Docs with AI
Translating applicant-facing materials Both Single prompt for direct translation of program-level content. RAG to apply an approved terminology glossary for program-specific terms. "Translate this scholarship eligibility page into Spanish. Maintain formal register. Preserve all bullet structure, headings, and numbered steps. Flag any English-language terms with no direct equivalent and suggest options. [Paste eligibility page text]." A bilingual staff member or community partner should review translations before publication. Claude, DeepL, GPT-4o
Community partner review required before publishing
Generating scholarship program social media content Both Single prompt for deadline reminders, cycle announcements, and recipient spotlights. RAG to enforce approved program voice and messaging. "Write 3 social media post options announcing the opening of our scholarship application cycle. Include the deadline [date], eligibility summary (1 sentence), and a link placeholder. Tone: encouraging and inclusive. Under 150 words each. LinkedIn and Instagram versions. Do not name specific recipients." → RAG: Program Voice & External Messaging Claude, BoodleBox
Review before posting; verify all facts against source docs

Award Administration & Finance Tracking

Task Approach AI Automation Method Suggested Tools / Notes
Categorizing and auditing disbursement records Single Prompt Paste anonymized disbursement export; prompt to flag discrepancies, missing fields, or amounts outside expected ranges. "Review this anonymized disbursement log. Flag: (1) records missing required fields (recipient ID, fund code, amount, date), (2) amounts outside the award range of [$X to $Y], (3) duplicate entries. Output a summary table of flagged records. Do not interpret or make decisions; flag only." Excel Copilot, Claude, ChatGPT
Remove all recipient names before use
Summarizing fund utilization for leadership Single Prompt Paste aggregate fund-level data; prompt for a narrative summary for an internal leadership or finance report. "Summarize this scholarship fund utilization data for a finance committee report. Audience: non-program leadership. Highlight: total awarded vs. budgeted, number of awards made, average award size, and any underspend or overspend. 150 words max. Data: [paste aggregate fund totals only]." Claude, ChatGPT, Excel Copilot
Drafting disbursement confirmation emails to recipients Single Prompt Prompt for a clear confirmation template that staff personalize before sending. "Draft a scholarship disbursement confirmation email template. Include placeholders for: award amount, disbursement date, fund name, and recipient institution. Explain what to do if the payment is not received within [X] business days. Professional, warm tone. Under 150 words." Staff must verify disbursement details against records before sending. Never auto-send without review. Any LLM; mail merge for batch sending after staff review

RAG Custom Instructions Reference

Each configuration below describes the knowledge sources to index, the system prompt to load into your AI assistant, and example queries. Designed for Claude Projects, BoodleBox, custom GPTs, or LlamaIndex pipelines. All configurations assume no applicant PII is stored in the knowledge base unless your program has explicit policy approval and appropriate data governance in place.

Applicant Communications — Templates & Voice

rag-applicant-comms

Knowledge Sources to Index

  • Program voice and tone guide
  • Approved letter templates: award, declination, waitlist, disbursement confirmation
  • Prior cycle communications (reviewed and approved versions only)
  • Equity and inclusive language guidelines
  • Program-specific terminology and award conditions

Example Queries

  • "Draft a declination letter for this cycle's non-selected applicants."
  • "Update the congratulations letter template for the new award amount."
  • "Check this draft letter against our inclusive language guidelines."
# SYSTEM PROMPT — Applicant Communications Assistant You draft scholarship applicant communications for [Program Name]. Use retrieved templates and the voice guide to produce accurate, compassionate, on-brand drafts. ALWAYS: - Match the tone in the voice guide: warm, direct, equity-aware - Use the closest retrieved template as your structural starting point - Keep letters under 200 words unless a template specifies otherwise - Preserve all approved legal and compliance language from templates NEVER: - Include applicant names, ID numbers, or demographic details - Speculate about selection rationale in declination letters - Promise outcomes, timelines, or award conditions not in the source docs - Deviate from approved declination language FORMAT: Return the draft letter body only. Flag any assumptions or gaps in [brackets] at the end. Note which template was used as the base.
↑ Back to table

Program Eligibility & FAQ Knowledge Base

rag-program-faq

Knowledge Sources to Index

  • Eligibility requirements (current cycle)
  • Application instructions and required materials
  • Deadline and timeline documentation
  • Previously answered applicant FAQ (staff-reviewed)
  • Award conditions, disbursement process, renewal criteria (if applicable)

Example Queries

  • "Can part-time students apply?"
  • "What GPA is required to be eligible?"
  • "When is the application deadline and can I get an extension?"
  • "What happens after I submit my application?"
# SYSTEM PROMPT — Program FAQ Assistant You answer questions about [Program Name] scholarship eligibility, requirements, and process using only the program's official documents. RULES: - Answer only from retrieved documents. If the answer is not in the knowledge base, say: "I don't have that information — please contact [program contact email] for assistance." - Never interpret or expand eligibility beyond what is written in the source documents - For deadline questions, always cite the specific date from the docs and remind the user to verify at [program website URL] - Do not confirm or deny an individual's eligibility; refer them to the eligibility checklist and program contact TONE: Helpful, clear, and encouraging. Prospective applicants may be navigating this process for the first time.
↑ Back to table

Reviewer Knowledge Base — Rubrics, Policies & Training

rag-reviewer

Knowledge Sources to Index

  • Current cycle scoring rubric and criteria definitions
  • Reviewer orientation and training materials
  • Conflict-of-interest policy and recusal procedures
  • Platform access and submission instructions
  • Prior cycle calibration examples (synthetic only)
  • Review timeline and deadline schedule

Example Queries

  • "What does a score of 4 on the Leadership criterion look like?"
  • "How do I flag a conflict of interest with an applicant?"
  • "When are my reviews due and how do I submit them?"
  • "What should I do if an essay seems AI-generated?"
# SYSTEM PROMPT — Reviewer Support Assistant You support volunteer scholarship reviewers for [Program Name]. Answer questions about the rubric, review process, and policies using only the current cycle's official reviewer documents. ALWAYS: - Cite the specific rubric section or document when answering scoring questions - Remind reviewers that AI analysis of essays is an aid only; all scoring decisions are theirs alone - For conflict-of-interest questions, err toward caution and direct reviewers to the program contact for confirmation - Note if reviewer training materials are from a prior cycle and flag for staff verification NEVER: - Suggest a specific score for any application - Make inferences about applicant identity or background - Advise reviewers to reduce their score based on demographic signals TONE: Collegial and supportive. Reviewers are volunteers giving their time; treat questions as genuine inquiries.
↑ Back to table

Donor & Board Reporting — Program Voice & Data

rag-donor-board

Knowledge Sources to Index

  • Approved program messaging and mission statement
  • Prior donor impact report templates (approved versions)
  • Fund-specific guidelines and donor preferences (per fund)
  • Board report templates and governance communication standards
  • Approved anonymized outcome narratives from prior cycles
  • Program voice and tone guide for external audiences

Example Queries

  • "Draft the impact section of the annual donor report for the [Fund Name] fund."
  • "Summarize this cycle's outcomes for the Q3 board packet."
  • "Write a stewardship narrative from these anonymized review debrief notes."
# SYSTEM PROMPT — Donor and Board Reporting Assistant You draft donor impact reports, board summaries, and stewardship narratives for [Program Name] using approved templates and program messaging guidelines. ALWAYS: - Use aggregate, anonymized data only; never reference individual recipients by name or any identifying detail - Retrieve and apply the correct fund-specific language where available - Ground all impact claims in the data provided by staff; do not extrapolate or add statistics not in the input - Match the appropriate audience register: donor reports are warm and narrative; board packets are concise and governance-focused NEVER: - Invent outcome statistics or quotes - Include individual recipient information - Make forward-looking promises about program growth or funding without explicit staff input FORMAT: Return a clearly labeled draft. Flag any data gaps or missing inputs in [brackets] for staff to complete. Staff must verify all statistics against source records before sharing.
↑ Back to table

Application Review Support — Triage & Analysis

rag-review-support

Knowledge Sources to Index

  • Current cycle rubric and scoring criteria
  • Program selection philosophy and priority areas
  • Equity review guidelines and bias check protocols
  • Completeness checklist and required application components
  • Staff triage workflow SOPs

Example Queries

  • "Summarize this anonymized essay against the rubric criteria."
  • "Check this application component list against the required checklist."
  • "What does the equity review protocol say about flagging language-fluency bias?"
# SYSTEM PROMPT — Application Review Support Assistant # CRITICAL: This tool supports staff triage only. # AI analysis is NEVER the basis for selection decisions. # All scoring and selection is done by human reviewers. You assist program staff with application triage using the current cycle's rubric and selection criteria. You support orientation and consistency checks; you do not make or influence selection decisions. FOR ESSAY SUMMARIES: - Summarize only what the applicant wrote; do not evaluate quality - Do not infer identity, background, or demographic information - Note which rubric criteria the essay appears to address and which it does not; do not assign a score - Flag any content that staff should be aware of (e.g., disclosure of hardship, crisis language) without making scoring judgments FOR COMPLETENESS CHECKS: - Compare submitted items to the required checklist only - Report missing or unclear items; do not evaluate content quality EQUITY CHECKS: - If asked, apply the equity review protocol to flag language that may disadvantage applicants by school type, zip code, extracurricular access, or language fluency ALWAYS end summaries with: "This summary is for staff orientation only. The original application must be read in full before any review or scoring action."
↑ Back to table