Give examples
Paste a past email or document you liked (with private details removed) and ask for "in this style."
Resources
Tips & tricks for using AI well — and safely.
Free, practical guidance from our engineers. Share it with your team, print it for your break room, or use it to start your own AI policy.
Better prompts
Talk to AI like a new colleague. Context in, quality out.
AI doesn't know your organization, your audience, or what "good" looks like unless you tell it. A simple structure makes a big difference.
- Role — who should the AI act as?
- Context — who's it for, and what background matters?
- Task — exactly what you want produced
- Format — length, tone, structure
- Check — ask it to flag assumptions or questions
prompt-template.txt
ROLE: You are a warm, clear communications
assistant for a mid-sized church.
CONTEXT: Our fall food drive runs Nov 1–15.
Readers are members of all ages; many are
new to volunteering.
TASK: Draft a newsletter blurb inviting
volunteers and listing 3 ways to help.
FORMAT: 120 words max, friendly tone,
one clear call to action.
CHECK: List anything you assumed so
I can verify it before sending.
# No names, phone numbers, or member details
# are needed — so we don't include them.Iterate, don't restart
Treat the first answer as a draft. "Shorter," "warmer," or "add a deadline" usually beats a brand-new prompt.
Ask for options
Request three versions with different tones, then pick and combine. It's faster than perfecting one.
Break big tasks down
Outline first, then draft each section. Long, complex requests in one go produce weaker results.
Ask it to critique
"What's unclear or missing in this draft?" turns AI into a helpful second reader.
Save what works
Keep a shared library of prompts that work well for your organization's recurring tasks.
What never goes into an AI tool. Unless your policy explicitly approves it.
Never enter
- Passwords, API keys, or security codes
- Social Security, driver's license, or passport numbers
- Credit card and bank account numbers
- Health, medical, or counseling information
- Prayer requests and pastoral care notes
- Children's personal information
- Confidential client, legal, or HR matters
Do this instead
- Replace names and details with placeholders like [MEMBER] or [CLIENT]. Our free redaction tool does it for you, privately
- Describe the situation in general terms
- Use only organization-approved AI tools and accounts
- Ask your AI lead when you're unsure — that's what they're for
- Report accidental exposure right away; quick reporting limits harm
- Approved tool? Am I using an organization-approved AI tool and account?
- Safe data? Have I removed personal and confidential information?
- Verified? Have I checked facts, numbers, names, quotes, and links?
- Fair? Could this output treat any group of people unfairly?
- Human decision? If this affects someone's job, money, or care, did a person decide?
- Disclosed? Should I tell the reader AI helped create this?
- Mine to share? Am I respecting copyright and others' work?
Safe-use checklist
Seven questions before you hit send. Print it. Pin it.
Responsible AI use comes down to habits. This checklist covers the most common mistakes.
Ideas to try this week. Low-risk, high-value starting points.
Turn a messy process into a checklist
Describe how a task gets done; ask for a step-by-step SOP.
Explain a spreadsheet
Ask for formulas, pivot ideas, or a plain-English summary of trends (non-sensitive data only).
Draft polite, firm emails
Overdue invoices, vendor follow-ups, schedule changes.
Repurpose one piece of content
Turn an article into a newsletter blurb, three social posts, and an email subject line.
Write for different audiences
Rewrite the same update for staff, customers, and the board.
Check readability & accessibility
Ask for a simpler version, alt text for images, or a plain-language summary.
Pressure-test a decision
Describe a plan and ask for risks, objections, and questions you haven't considered.
Prepare for a meeting
Summarize long public reports and draft an agenda with time boxes.
Draft your AI policy
Start with our free AI policy generator, then review it with counsel.
Plan a volunteer schedule
Generate a rotation template and reminder messages (use roles, not names).
Translate announcements
Draft translations for multilingual members — and have a fluent speaker review them.
Build small-group guides
Turn sermon notes into discussion questions for small groups.
Common pitfalls. And how to avoid them.
Trusting confident answers
AI can "hallucinate" — state false facts, quotes, or citations confidently. Verify anything that matters.
Personal accounts for work
Free personal accounts often lack admin controls and may use your inputs for training. Use organization accounts.
Over-connected AI
Giving an AI tool access to "everything" means it can surface anything. Connect only what's needed.
Hidden instructions
Documents, emails, and web pages can contain text that tries to redirect an AI ("prompt injection"). Be careful letting AI act on untrusted content.
Automating judgment
Screening applicants or deciding eligibility with AI alone invites bias and legal risk. Keep a human decision-maker.
Skipping the people part
Rolling out tools without explaining why — or addressing job concerns — breeds distrust. Communicate early.
Glossary
AI terms, in plain English
Large language model (LLM)
The technology behind tools like ChatGPT, Claude, Gemini, and Copilot. It's trained on huge amounts of text to predict and generate language.
Prompt
The instructions or question you give an AI tool. Better context in your prompt generally means better results.
Hallucination
When AI produces information that sounds right but is false or made up. Always verify important facts.
Generative AI
AI that creates new content — text, images, audio, or code — rather than only analyzing existing data.
AI agent
An AI system that can take actions (send emails, update records, run steps) toward a goal, not just answer questions. Agents need careful permissions and approval steps.
Retrieval-augmented generation (RAG)
A technique that lets AI answer using your own documents by retrieving relevant passages first. It improves accuracy and allows citations.
Human-in-the-loop
A design where a person reviews or approves AI outputs before they take effect — essential for consequential decisions.
Prompt injection
An attack where hidden instructions in content (a web page, email, or document) try to make an AI do something it shouldn't.
Shadow AI
AI tools used by staff without the organization's knowledge or approval — often through personal accounts.
Model training vs. inference
Training is how a model learns; inference is when you use it. Enterprise plans typically promise not to use your inputs for training — confirm it in writing.