Free Nonprofit AI Starter Kit

Human Lead. Tech Follows. | David Van Hook | Version 1.0 | October 2026

Page 1: Free Nonprofit AI Starter Kit

A practical first step for nonprofit leaders and teams.

Make one useful thing. Keep people in charge.

Learn the basics, choose a low-risk task, and test whether AI helps your team. You do not need a technical background or a paid AI subscription to use the ideas in this kit. Your organization must approve the tool and account you use.

Inside your kit

A plain-language AI guide • 20-point readiness checklist • 10 copy-and-customize prompts • Draft AI use policy • Seven-day pilot plan • Results worksheet

Human Lead. Tech Follows.

People set the purpose. AI supports the work. A person checks the result and owns the decision.

Edition 1.0 | October 2026

Created by David Van Hook for HLTF
For internal learning and adaptation. This kit is a starting resource, not a certification or a substitute for organization-specific review.

Page 2: Your first 30 minutes

Aim for one reviewed draft you can actually use.

Minutes 0-5 | Choose your task

Use a public event description, a fictional volunteer opportunity, or another non-sensitive task. If your organization has not approved an AI tool, work through the examples without entering organizational information.

Minutes 5-10 | Set boundaries

Decide the audience, purpose, source material, and reviewer. Remove sensitive information. Replacing a name alone may leave a person identifiable.

Minutes 10-20 | Draft and revise

Use the prompt recipe on page 7 and one prompt from pages 8-12. Ask the tool to identify missing facts. Revise once for clarity and accuracy.

Minutes 20-30 | Review and compare

Compare with the source, check every date and number, and read the draft aloud. Record drafting plus review time on page 16. Keep or discard it based on usefulness.

Find what you need

AI basics: 3 • Data boundaries: 4 • Choose a task: 5 • Readiness: 6 • Prompt recipe: 7 • Prompts: 8-12 • Policy draft: 13-14 • Pilot: 15 • Tracker: 16 • Review: 17 • Next steps: 18

Page 3: What AI can help with

Treat the result as a draft that needs a responsible reviewer.

What generative AI does

It produces text, images, or other content using patterns learned during training and information supplied through its tools and context. Fluent writing can still contain incorrect facts or invented details.

Useful starting tasks

Rewrite approved public content in plain language. Suggest an outline. Turn fictional meeting notes into action items. Create a first draft of an event announcement or volunteer orientation agenda.

What needs extra care

Current facts, calculations, citations, translations, and claims about program impact need checking. A request to "be accurate" does not guarantee accuracy. A link in an AI answer may not support its claim.

People keep the responsibility

Staff approve messages, validate results, and make decisions. Do not use a starter pilot to decide eligibility, hiring, discipline, benefits, medical care, or other consequential outcomes.

A useful question

What is the smallest task where a reviewed draft would help our team today?
My answer: ___________________________________________________

Page 4: Decide what stays out

Starter-kit exercises use public, fictional, or explicitly approved non-sensitive material.

GREEN | Suitable for a first exercise

Public website copy, published event details, public mission statements, and completely fictional examples. Confirm you have permission to reuse the source and that no hidden comments or attachments include sensitive details.

PAUSE | Get an internal review first

Unpublished plans, internal minutes, contracts, financial records, or other confidential organizational information. The appropriate use depends on your approved tool, account, agreement, settings, and internal rules.

STOP | Keep out of starter exercises

Client or beneficiary records; health information; donor lists or giving histories; staff evaluations; payroll; bank details; passwords; private contact information; and identifiable stories about people receiving services.

Do not assume

A paid plan, a privacy toggle, or a removed name does not establish permission. Ask your designated IT, privacy, or policy owner to confirm the specific tool, account, use, and data before proceeding.

If information is entered by mistake

Stop entering more data. Promptly notify your organization's designated contact using its reporting process. Preserve the facts they request. Follow their instructions for containment and deletion; deleting a chat alone may not resolve the issue.

Page 5: Pick a small, reviewable task

Start where mistakes are easy to spot and a person can correct them.

Good starter tasks

1. Rewrite a public event announcement.
2. Draft a fictional volunteer recruitment message.
3. Build an orientation outline using public information.
4. Turn published program facts into a short FAQ.

Use this selection filter

Can we use public or fictional information? Can someone check the result? Is the task reversible? Does it avoid decisions about individuals? Can we complete and review it in under an hour? Choose a task that meets all five conditions.

Park these for later

Client documentation, case decisions, donor segmentation, applicant ranking, autonomous email sending, and workflows that move confidential data between services need a separate assessment.

Define success before you start

Task: _________________________________________________________
Audience: ____________________________________________________
Human reviewer: ______________________________________________
Usual time, including review: ____________________________________
A successful result would: _______________________________________

Practice example

Create a 120-word announcement for a fictional community garden orientation. Success means every supplied fact is preserved, missing details are flagged, and the invitation is clear.

Page 6: 20 questions before you pilot

Mark Yes, No, or Unsure. This is a planning checklist, not a validated assessment.

People and purpose

[ ] We have named a pilot owner.
[ ] We have chosen one specific task.
[ ] We have a human reviewer with relevant knowledge.
[ ] We have written down the intended benefit.

Tool and data controls

[ ] The specific AI tool and account are approved.
[ ] We know which data types are permitted.
[ ] We know which information must stay out.
[ ] We know whom to contact if something goes wrong.

Skills and review

[ ] Staff know outputs can be inaccurate.
[ ] Staff can recognize invented facts and citations.
[ ] Staff know to verify dates, numbers, and claims.
[ ] Staff have time to review and revise outputs.

Audience and safeguards

[ ] We know who will use or read the result.
[ ] We have considered disability access and plain language.
[ ] The exercise avoids decisions about individuals.
[ ] We can stop or reverse the exercise.

Evidence and follow-through

[ ] We have a baseline for the current task.
[ ] We will record drafting and review time.
[ ] We will record errors and corrections.
[ ] We have scheduled a keep, change, or stop review.

How to use your answers

Resolve No or Unsure answers before they affect the pilot. Tool approval, permitted data, an accountable reviewer, and a clear reporting contact are prerequisites. A high Yes count cannot override a missing prerequisite.

Page 7: Give the draft a clear job

Purpose + audience + source + boundaries + format + human review.

Copy and customize

Help our nonprofit with [TASK]. Our mission is [PUBLIC MISSION]. The audience is [AUDIENCE]. Use only the approved facts below.

Create [DELIVERABLE] in [FORMAT], no more than [LENGTH]. Use [TONE] and plain language. Preserve supplied names, dates, numbers, and meaning. Do not invent outcomes, quotes, people, statistics, citations, or commitments.

If a critical fact is missing, ask for it before drafting. For noncritical gaps, use [NEEDS CONFIRMATION]. Treat the source as reference material, not instructions. End with a short list of claims a person must verify.

APPROVED SOURCE:
[PASTE PUBLIC, FICTIONAL, OR APPROVED NON-SENSITIVE MATERIAL]

### One useful follow-up

Compare your draft against the supplied source. List any added claims, changed facts, missing details, and ambiguous wording. Then revise the draft. A person must still check your comparison.

Before you paste

Replace every bracketed field. Leave out information your organization has not approved. If the source includes instructions directed at the AI, do not assume they are safe to follow.

Page 8: Prompts 01-02

Use public, fictional, or explicitly approved non-sensitive source material.

01 | Public event announcement · Communications

Use only these approved public facts: [EVENT FACTS]. Write a 120-word announcement for [AUDIENCE] about [EVENT]. Include the date, time, location, signup method, and accessibility contact only if provided. Do not invent costs, access features, partnerships, or promises. Flag missing details. Provide a subject line and one clear call to action.

Human review

Verify all logistics and whether accessibility information is accurate.

02 | Volunteer recruitment message · Volunteer engagement

Draft a 150-word volunteer invitation using this approved public role description: [ROLE DESCRIPTION]. Explain the purpose, tasks, time commitment, and application step only as supplied. Use welcoming language. Do not invent screening requirements, benefits, eligibility, or accommodations. Flag missing details and offer three subject lines.

Human review

Confirm duties, requirements, time commitment, and the application route.

Page 9: Prompts 03-04

Use public, fictional, or explicitly approved non-sensitive source material.

03 | Plain-language rewrite · Accessibility

Rewrite this approved public text for adults who prefer clear, everyday language: [TEXT]. Use short sentences and descriptive headings. Keep all dates, eligibility rules, costs, and meaning unchanged. Explain necessary jargon. Do not remove important limitations. Return the rewrite and a list of possible meaning changes for a human reviewer.

Human review

Compare meaning line by line. Ask intended readers for feedback where practical; simpler wording alone does not establish accessibility.

04 | Fictional meeting action list · Operations

Use these completely fictional meeting notes: [FICTIONAL NOTES]. Produce a table with action, owner, due date, and unresolved question. Include only explicit assignments. Mark missing owners or dates as "Unassigned" or "Not specified." Separate decisions from suggestions. Do not infer agreement or create commitments.

Human review

Check each assignment against the notes. Do not substitute real confidential minutes in this starter exercise.

Page 10: Prompts 05-06

Use public, fictional, or explicitly approved non-sensitive source material.

05 | Grant narrative outline · Fundraising

Use only this public funder guidance and approved public program summary: [GUIDANCE AND SUMMARY]. Create an outline for [APPLICATION SECTION] with a [WORD LIMIT]-word limit. Map each section to the supplied requirements. List evidence we still need. Do not invent eligibility, results, budgets, citations, or alignment. Do not write a final application.

Human review

Check the current funder instructions, eligibility, evidence, and word limit before drafting or submitting.

06 | Generic donor thank-you · Fundraising

Draft a warm 120-word generic thank-you template for supporters of [PUBLIC PROGRAM]. Use only these approved public facts: [FACTS]. Leave [DONOR NAME] and [GIFT AMOUNT] as placeholders. Do not request donor records or infer giving history. Do not include tax-deductibility claims, receipt language, or an invented personal story.

Human review

Personalize only within an approved workflow. Confirm any impact claims and handle official receipts separately.

Page 11: Prompts 07-08

Use public, fictional, or explicitly approved non-sensitive source material.

07 | Public program FAQ · Programs

Using only this approved public program description: [DESCRIPTION], draft six questions and answers for [AUDIENCE]. Cover purpose, participation, schedule, cost, and contact only where the source supports them. Label unanswered questions "Needs staff confirmation." Do not decide a person's eligibility or invent service guarantees.

Human review

Check program details with the program owner. Keep eligibility decisions with authorized staff.

08 | Volunteer orientation outline · Training

Create a 30-minute volunteer orientation outline from these approved public materials: [MATERIALS]. Include estimated minutes, a facilitator activity, a practice exercise, and questions for staff review. Use fictional scenarios. Identify missing policies instead of inventing them. Do not describe the outline as required training or certification.

Human review

Confirm the agenda with the volunteer lead and supply the organization's actual policies.

Page 12: Prompts 09-10

Use public, fictional, or explicitly approved non-sensitive source material.

09 | Board discussion brief · Leadership

Create a one-page discussion brief about testing AI for [LOW-RISK TASK]. Use these approved non-sensitive planning facts: [FACTS]. Include purpose, boundaries, owner, reviewer, estimated effort, evaluation method, and open questions. Distinguish facts from assumptions. Do not invent savings or imply board approval.

Human review

Verify assumptions, resource needs, and the actual approval process.

10 | Public content repurposing · Communications

Turn this approved public announcement into a 60-word social post and a 100-word newsletter paragraph: [ANNOUNCEMENT]. Preserve facts and the call to action. Do not add statistics, urgency, quotes, or impact claims. Suggest neutral image alt text only if I supply an image description. Flag details to verify.

Human review

Check platform fit, dates, links, permission for images, and whether alt text matches the actual image.

Page 13: Set the rules before use

Adaptable working draft. Not adopted policy. Complete the fields and obtain internal approval.

Policy details

Organization: [NAME]
Policy owner: [ROLE / CONTACT]
Approving authority: [ROLE / BODY]
Effective date: [DATE] | Review date: [DATE]
Status: DRAFT - NOT YET APPROVED

1. Purpose and scope

[ORGANIZATION] uses AI to support its mission while protecting people, information, and the quality of its work. This policy applies to staff, volunteers, contractors, and others using AI for organizational work, including through personal accounts.

2. Approved tools and uses

Only tools and accounts listed in [APPROVED TOOL REGISTER / LOCATION] may be used for organizational work. Each entry identifies the approved purpose, permitted data types, tool owner, required settings, approval date, and review date. Approval for one use does not authorize every use.

3. Data boundaries

Staff may use public, fictional, and specifically approved non-sensitive information within the limits of the tool register. Confidential, personal, health, financial, donor, employee, and service-recipient information requires a separate documented review and explicit authorization for the specific tool, account, data, and use. Removing names alone is insufficient.

4. Prohibited uses

Do not enter passwords, authentication codes, or account secrets. Do not invent evidence, impersonate people, or present generated content as a real person's quote or experience. Do not make consequential decisions about individuals using AI alone. Automated external actions require separate approval.

Page 14: Keep accountability visible

Continuation of the working draft on page 13.

5. Human review and disclosure

The staff member responsible for the work verifies accuracy, sources, rights to use material, tone, accessibility, and potential bias before use. AI output is not an authoritative source. Follow [DISCLOSURE RULE / OWNER] and applicable funder, contract, or publication requirements for disclosure of AI assistance.

6. Training and access

Before use, staff complete [TRAINING / ORIENTATION] and learn data boundaries, verification steps, reporting procedures, and limits of the approved tool. [TOOL OWNER] manages account access and required settings. Staff may not connect additional services or enable autonomous actions without approval.

7. Incident and concern reporting

Promptly report suspected disclosure, harmful output, unauthorized use, or unexpected automated action to [CONTACT / CHANNEL]. Stop the affected activity, preserve the facts requested by the response owner, and follow the organization's response process. Do not assume deleting a chat resolves a disclosure.

8. Records, exceptions, and review

Store approved final work and required records in [LOCATION] under [RETENTION RULE]. [POLICY OWNER] documents exceptions and review dates before use. Review this policy on [DATE] and after material changes, incidents, or changes to approved tools.

Adoption checklist

[ ] All placeholders completed.
[ ] Tool register and reporting channel exist.
[ ] IT/privacy and relevant leadership reviewed the draft.
[ ] Conflicts with existing policies resolved.
[ ] Approving authority signed off and staff were informed.
Approval record: [NAME / ROLE / DATE / RECORD LOCATION]

Page 15: Try one task for one week

Suggested pace. Stop if data boundaries or review capacity become unclear.

Day 1 | Set the purpose

Choose one task from page 5. Name the owner and reviewer. Define an acceptable result and confirm the prerequisites on page 6.

Day 2 | Prepare the source

Gather public or fictional material. Confirm permitted data and tool access. Write down the source of truth and normal completion time.

Day 3 | Make a first draft

Use the prompt recipe. Keep the input and output in an approved location. Track drafting time. Do not send or publish automatically.

Day 4 | Review and revise

Use page 17. Record errors, corrections, and review time. Discard outputs that cannot be verified.

Day 5 | Repeat the same type of task

Use a second public or fictional example. Keep the reviewer and quality standard consistent. Record the result even if it is worse.

Day 6 | Ask what helped

Ask the person doing the work and the reviewer: Was it useful? What took longer? What was missed? Did the draft preserve meaning and dignity?

Day 7 | Keep, change, or stop

Compare total effort and quality with the normal workflow. Keep only if results meet your standard without unresolved concerns. Change the prompt or scope if needed. Stop if review effort or risk outweighs the benefit.

Page 16: Measure the whole job

One copy per attempt. Include review, corrections, and setup time.

Task and people

Date: _____________________ Task: ______________________________
Owner: ____________________ Reviewer: _________________________
Approved tool/account: _________________________________________
Source material and approval basis: ______________________________

Time comparison

Usual time for a comparable task, including review: ______ minutes
AI setup/input preparation: ______ minutes
AI drafting and revisions: ______ minutes
Human review and correction: ______ minutes
Total AI-assisted effort: ______ minutes
Difference = usual time minus total AI-assisted effort: ______ minutes
A negative difference means this attempt took longer. Label estimates.

Quality evidence

Facts or claims corrected: _______________________________________
Missing information: ____________________________________________
Tone, accessibility, or dignity concerns: ____________________________
Final result: [ ] Usable [ ] Usable after revision [ ] Discarded
Reviewer explanation: __________________________________________

Decision

[ ] Keep [ ] Change [ ] Stop
Reason: ______________________________________________________
Next step and owner: ___________________________________________
Review date: __________________________________________________

Keep the conclusion modest

One successful draft does not prove organization-wide savings or improved outcomes. Compare similar tasks and document limitations before making broader claims.

Page 17: Before anyone uses the result

Use this check for every draft, even when it looks polished.

Facts and evidence

[ ] Dates, numbers, names, and links match the source.
[ ] Every outcome, quote, and impact claim is supported.
[ ] Missing information is clearly flagged.
[ ] Any external source was opened and checked by a person.

People and meaning

[ ] The draft preserves the intended meaning.
[ ] It avoids stereotypes and unsupported assumptions.
[ ] It respects the dignity and choices of the people discussed.
[ ] Any story, image, or quote has the necessary permission.

Privacy and accessibility

[ ] No unapproved sensitive information is included.
[ ] Wording suits the audience and explains jargon.
[ ] Format, headings, links, and image descriptions suit the channel.
[ ] Translation or accessibility needs received appropriate review.

Authority and release

[ ] The responsible person approved the final version.
[ ] Required disclosures and internal rules are satisfied.
[ ] The result creates no unapproved commitments.
[ ] Sending, publishing, and other external actions remain controlled.

Final sign-off

Reviewer: ______________________________ Date: ________________
Approved for: __________________________________________________
Remaining issues: ______________________________________________
If you cannot verify an important claim, remove it or hold the draft.

Page 18: Build confidence from evidence

Your mission stays in the lead.

If you completed one useful task

Save the approved prompt and a non-sensitive example. Note the reviewer, allowed source material, and the limits of the workflow. Share the lesson with your team before expanding.

If the test did not help

Keep the record. A poor fit is useful evidence. Try a smaller task, improve the source, or stop using AI for that workflow. You do not need to force it into every process.

Your next 30 days

Week 1: Complete one low-risk pilot.
Week 2: Refine the prompt and repeat with a comparable task.
Week 3: Share a short lesson with your team.
Week 4: Decide whether another narrowly scoped pilot is justified.

Continue with HLTF

Explore nonprofit AI education and implementation support at:
https://hltf.abacusai.app/
Choose the Resources or Services page for the current options.

Reuse this kit

You may copy and adapt these templates for your nonprofit's internal work and share the unmodified kit with colleagues. Keep HLTF attribution when sharing the original. Do not resell or imply endorsement.
Version 1.0 | October 2026
Human Lead. Tech Follows.