Soarin Group · AI at work

Practical ways to put AI to work thoughtfully.

A useful place to start with everyday work: small ideas, ready-to-adapt prompts, and clear guardrails that keep people and their judgment at the center.

Try one low-risk example with fictional or sanitized information. Review the result before you use it.

Latest issue · practical plays

Nine small AI plays for a steadier workday

Nine starting points for common work tasks. Open a play to see the situation, a prompt you can adapt, what to check, and a safe way to try it.

Published November 6, 2026Reviewed November 6, 2026

01Turn a meeting into clear next steps
Long notes can make it hard to spot what was decided, who owns a task, and what still needs an answer.
Prompt to adapt
```
You are helping me organize a meeting summary. Use only the notes below. Do not add facts, decisions, owners, dates, or commitments that are not stated. If something is unclear, label it “Needs confirmation.”

Return:
1. A five-bullet summary
2. Decisions explicitly made
3. Action items in a table with task, named owner (or “Unassigned”), due date (or “Not stated”), and source wording
4. Open questions

Keep the tone neutral and concise. Do not infer agreement from discussion.

Sanitized notes:
[Paste only notes approved for this AI service. Replace names and sensitive details with role labels or placeholders.]
```
What to check
A short summary that separates decisions from discussion, ties action items to the notes, and flags missing owners or dates instead of guessing.
A low-risk first try
Test it on a low-risk internal meeting. The organizer compares the summary with their notes and approves it before sharing. Check consent, retention, and tool rules before using any recording or transcript.
Use it safely
Do not upload recordings or transcripts with employee, client, financial, health, or confidential business details unless both the service and this use are explicitly approved.
02Draft a helpful customer-service reply
A first draft can be calm and clear while staying within approved policy and avoiding promises the team cannot confirm.
Prompt to adapt
```
Draft a warm, plain-language reply to the customer scenario below. Use only the approved facts and policy excerpt I provide. Do not promise a result, make an exception, guess a timeline, or imply that an action has already happened. If the facts do not answer the question, say what needs confirmation and suggest handing it to the right person.

Return a draft under 150 words, then list any facts that the sender must verify before sending. Avoid jargon and keep the customer’s concern in view.

Approved facts/policy:
[Paste approved, non-confidential wording]

Sanitized scenario:
[Describe the issue without names, account numbers, contact details, or unique identifiers]
```
What to check
A respectful, direct reply based only on the supplied policy, with no invented resolution or timeline and a short verification list.
A low-risk first try
Have a staff member check the draft against a maintained source of approved response language. A person handles exceptions and sends the final reply.
Use it safely
Do not paste identifiable customer messages, contact details, account identifiers, private correspondence, sensitive case details, passwords, or payment data into an unapproved tool.
03Make an HR document easier to understand
A dense policy or process draft can hide practical steps or questions that need a person to review.
Prompt to adapt
```
Rewrite the approved, non-confidential text below in plain language for employees. Preserve its meaning. Do not add a rule, entitlement, deadline, legal interpretation, or promise. Keep any defined terms unchanged. If a sentence is ambiguous or appears to conflict with another sentence, flag it rather than resolving it.

Return:
1. A clearer draft
2. A short list of meaning changes or ambiguities to review
3. Any terms that should stay as written

Text:
[Paste only text approved for this AI service]
```
What to check
Shorter sentences and a clearer order, with the original meaning intact and questionable points surfaced for the document owner.
A low-risk first try
Compare the rewrite line by line with its approved source. The HR owner reviews it and seeks qualified advice when an interpretation needs expertise.
Use it safely
Use generic policy wording only. Keep names, performance or disciplinary details, accommodation or health information, applicant records, compensation, benefits, and payroll out of unapproved tools.

04Sort a help-desk request into a useful handoff

A clearer summary of impact, missing information, and safe questions can help prepare an IT handoff.

Prompt to adapt

Help prepare a help-desk triage note from the sanitized request below. Do not diagnose beyond the evidence, request secrets, or suggest bypassing security controls. Do not recommend a configuration or production change. If information is missing, list safe clarification questions.

Return:
- Issue summary in one sentence
- Reported impact and scope (mark unknown items as unknown)
- Safe clarification questions
- Low-risk checks the user can do without admin access
- Suggested routing category and why, marked as a draft
- Any sign that a human should review promptly

Sanitized request:
[Remove names, email addresses, device identifiers, public IPs, internal hostnames, ticket links, log contents and any unique business details]

What to check

A concise handoff that separates reported facts from unknowns, asks for safe details, and leaves diagnosis to a person.

A low-risk first try

Use fictional tickets first. Keep routing suggestions advisory and follow the service desk’s usual priority, identity-check, and escalation rules.

Use it safely

Never paste passwords, codes, tokens, private keys, unredacted logs, vulnerability details, network diagrams, or live customer or employee identifiers. Do not disable protections or grant access based on a generated suggestion.

05Explain a small reporting snapshot
A readable summary can help a manager see what aggregate figures show and what questions remain.
Prompt to adapt
```
Summarize the aggregate table below for a business manager. Use only the figures shown. Do not calculate a percentage unless you show the formula, do not claim a cause, and do not fill in missing values. Separate observed changes from possible explanations.

Return:
1. Three factual observations with the relevant figures and period
2. Missing, inconsistent, or unusually small samples to check
3. Up to three follow-up questions
4. A short draft summary with no causal claim

Table and metric definitions:
[Paste aggregate figures, units, time period and definitions approved for this AI service]
```
What to check
Observations connected to the supplied figures and period, transparent calculations, visible data gaps, and questions instead of causal stories.
A low-risk first try
Check totals and formulas against the source spreadsheet. Include definitions and comparison periods; a human owner interprets the results and makes any decision.
Use it safely
Use aggregate or synthetic data where possible. Do not paste row-level employee, customer, payroll, account, or transaction exports into an unapproved service; small groups can also identify people.
06Prepare a sales follow-up draft
A timely follow-up should reflect what was actually discussed without overstating what the business can deliver.
Prompt to adapt
```
Draft a concise follow-up email using only the approved notes and service description below. Do not invent a customer need, quote, capability, result, price, deadline, or commitment. If a detail is missing, use a neutral placeholder in brackets. Include one clear next step and make the tone helpful, not pushy.

After the draft, list every factual claim the sender should verify.

Approved service description:
[Paste current, approved wording]

Sanitized call notes:
[Use role labels and remove names, contact details, confidential plans and identifying account details]
```
What to check
A short message grounded in the notes and current approved service wording, with uncertain details left for the sender to confirm.
A low-risk first try
The account owner checks the message against their notes and approved service language, then handles personalization and sending.
Use it safely
Do not include identifiable prospect or client correspondence, private business plans, contact lists, negotiated pricing, or account credentials in an unapproved AI service.
07Make an internal process easier to follow
A recurring task can be difficult to hand off when its steps live mostly in people’s heads.
Prompt to adapt
```
Turn the approved process notes below into a draft checklist for a trained teammate. Preserve the sequence and control points. Do not invent approvals, system access, deadlines, policy, or exceptions. Mark gaps as “Owner input needed.”

Return:
- A short purpose statement
- Numbered steps, each starting with a verb
- Decision points and handoffs
- A list of unanswered questions

Process notes:
[Use a fictional or sanitized example, with no credentials, client details, employee records or internal secrets]
```
What to check
A reviewable checklist with decisions and handoffs visible and missing steps clearly marked.
A low-risk first try
Ask the person who does the task to validate the sequence. Test the draft with a sample case before it becomes an approved procedure.
Use it safely
Leave out login instructions, secrets, security-control bypasses, and sensitive client or employee examples. Keep internal process documents in approved storage with appropriate access controls.
08Compare leadership options without handing over the decision
A balanced view of tradeoffs and unknowns can help a leadership team prepare for its own discussion.
Prompt to adapt
```
Act as a neutral discussion facilitator. Compare the options below using only the supplied goals and facts. Do not select an option, invent costs or benefits, claim certainty, or treat an assumption as a fact.

Return a table with: option, stated benefits, stated tradeoffs, assumptions, information still needed, and a low-risk next question. Then list perspectives the leadership team may be missing.

Goals and constraints:
[Add only approved, non-confidential information]

Options and known facts:
[Describe options without client identifiers, employee records, credentials or confidential financial details]
```
What to check
A balanced comparison that separates facts, assumptions, and missing information and raises questions for people to discuss.
A low-risk first try
Use the result to prepare a meeting, not replace one. Leaders verify facts, hear affected perspectives, and decide through the organization’s normal process.
Use it safely
Keep board materials, acquisition plans, confidential financials, client contracts, identifiable personnel cases, credentials, and other restricted information out of unapproved tools.
09Check a draft for accidental oversharing
A draft email, procedure, or public response may contain details its intended audience does not need.
Prompt to adapt
```
Review the sanitized draft below for information that may be private, confidential, security-sensitive, or more specific than the intended audience needs. Do not reproduce or expand sensitive details. Do not decide whether disclosure is legally allowed.

Return:
- The type of detail to review (describe the category, not the value)
- A safer placeholder or generalization
- Any audience or approval question for a human reviewer
- “No issues spotted in the text provided” only if none are apparent

Draft:
[Use a test or sanitized draft. Never paste credentials, employee or client records, private keys, account numbers or unredacted incident details]
```
What to check
A short list of detail categories to review without echoing their values, plus an audience or approval question for a person.
A low-risk first try
Use this as an extra editing pass after your usual review. Approved audience rules and incident-response steps remain the source of truth.
Use it safely
This is not a security scanner and does not make it safe to paste a secret into a tool. Start with a sanitized copy; follow your organization’s incident process if a real secret or record was exposed.

People stay responsible

A few safety habits before you try a prompt.

Use an approved tool for the information involved. The prompts here are workflow starters, not instructions to change a live system.

Use only an AI service and account your organization has approved for the information involved. Ask the responsible person to check its current privacy, retention, training, access, and administrative settings.
Never enter passwords, access tokens, authentication codes, recovery details, private keys, or other credentials.
Keep identifiable employee records, health or accommodation details, payroll or benefits data, applicant materials, customer records, confidential contracts, financial account details, and security logs out of public or unapproved tools. Use fictional or properly sanitized examples.
Removing a name may not make information anonymous. Small details can identify a person, customer, device, or business. When unsure, leave the detail out and ask your IT or privacy contact.
Treat generated summaries, classifications, and recommendations as drafts. Check names, dates, figures, sources, tone, and completeness. A person with the right responsibility reviews, decides, and approves the final result.
Do not let generated text make hiring, discipline, compensation, access, security, customer-commitment, or other consequential decisions on its own. Follow your organization’s normal approval and change process.

About this guide

Useful starters, with human judgment in the loop.

This is educational information for small-business readers. AI-generated text is a draft: a person must review it and approve any final use. This guide does not provide individualized legal, tax, HR, or security advice.

Dated editions

Issue history

Each issue shows when it was published and last reviewed, so you can find the current starting point and revisit earlier guidance.

Nine small AI plays for a steadier workdayNovember 6, 2026 Current issue