AI for Everyday Work and Prompting
Module 1 Choose a useful and safe task

Aabha AI Academy · Codeless AI

An AI assistant can help turn rough notes into readable text. You still decide what to share, what the task means, and whether the result is safe to use. In this module, you will make those decisions before preparing a small, checked workplace update.

What you will learn

By the end, you will be able to:

- Choose a small, low-risk task with a defined input, output and human reviewer.
- Explain why a convincing AI response can still need correction.
- Decide whether to proceed, change the input, or stop and ask before sharing information.
- Compare a three-bullet AI output with its source and identify unsupported details.

What you need

You need basic browser skills and a way to copy text and save your work. No AI experience or programming is required. Use an approved general-purpose text-chat assistant and a text editor or notebook. The exercises require only pasted text; no paid subscription, connected app, upload or image-generation feature is needed.

Use only the fictional material supplied here. Do not substitute real employee, customer or business records. If you cannot use a chat tool, complete the checks with the supplied sample responses and return to the live practice when access is available.

Plan for about 60 minutes: five short lessons and demonstrations, 25 minutes; guided and independent practice, 25 minutes; self-check and review, 10 minutes. This is a planning estimate; take longer where needed.

Your workplace scenario

You support the fictional Willow Team, which is preparing a community workshop. Your coordinator wants a brief internal progress update. You are helping prepare a draft, with no responsibility to book anything, contact anyone or make promises.

Everyone, every organisation and every operational detail in this module is invented for learning. The source notes are the evidence for this exercise; they are not claims about a real event.

Lesson 1 Understand the job an AI assistant is doing

Suggested time: 5 minutes.

Artificial intelligence, or AI, is a broad term for computer systems that perform tasks such as recognising patterns or generating text. Generative AI produces new content in response to an input. In this course, the content is ordinary written text.

A prompt is the instruction or question you give the assistant. It can include material you want it to work on. A response is what the assistant produces. A conversation may contain several prompts and responses.

For example, you could provide three rough notes and ask the assistant to rewrite them as three clear bullets. The notes are the source material. “Rewrite these as three clear bullets” is the instruction. The resulting bullets are the response.

AI can be useful for drafting, shortening, rewriting, grouping ideas and extracting information from text. These tasks are especially manageable when you already have the facts and can compare the response with them.

Producing a readable answer does not establish that its contents are true. A language model can produce plausible wording that includes a wrong date, an invented explanation or a reference that does not exist. This is often called a hallucination. A confident tone is not evidence. OpenAI describes these limitations in its guidance on whether ChatGPT tells the truth (https://help.openai.com/en/articles/8313428-does-chatgpt-tell-the-truth).

Worked example

Fictional source:

“Room requested. Venue has not confirmed.”

Weak response:

“The venue is booked and preparations are on track.”

The sentence is fluent, but both claims go beyond the source. A request is not a booking. “On track” is a judgement that the note does not establish.

Better response:

“The room has been requested; venue confirmation is still pending.”

This keeps the useful information and preserves the uncertainty. It does not make the update less professional. It makes it more accurate.

Pause and notice: The assistant can change the wording. It cannot turn an unconfirmed fact into a confirmed one by writing confidently.

Lesson 2 Choose a task you can actually check

Suggested time: 5 minutes.

A bounded task has a clear starting point and a clear finish. You can say what information will go in, what should come out, and who will review it. The task is small enough that the reviewer can inspect the result.

Compare these requests:

- Too broad: “Manage the workshop and make it a success.”
- Bounded: “Turn these three fictional notes into a three-bullet internal progress draft. I will check each bullet against the notes.”

The broad request hides many decisions: budget, bookings, responsibilities and what success means. The bounded request asks for one transformation of supplied material.

Before choosing a task, complete this sentence:

“I will use [permitted input] to produce [specific output] for [reader or purpose], and [person] will check it against [evidence].”

For our scenario:

“I will use three fictional workshop notes to produce a three-bullet internal update for the coordinator, and I will check every bullet against the notes.”

What makes a good first task

Choose work where a mistake is easy to notice and correct before it affects anyone. Rewriting an invented announcement is a better first exercise than deciding who should receive a real service or whether a contract should be signed.

Also ask whether AI is useful here. If you need the exact total of 18 plus 7, an ordinary calculator is simpler to check. If the job requires a legal, medical or other high-stakes professional judgement, use an appropriate qualified person and approved process. AI-generated wording does not supply that authority.

The workshop update is a useful first task because the input is short, the output has a fixed size, and the learner can verify every statement. The human reviewer remains responsible for accepting, correcting or rejecting the result.

Try it mentally: “Write something about the workshop” is missing a specific output and evidence boundary. “Summarise these notes in three bullets for my own review” supplies both.

Lesson 3 Decide what information may go in

Suggested time: 5 minutes.

Before entering anything, separate two questions:

1. Is this information permitted in this tool for this purpose?
2. Is all of this information necessary for the task?

Access is not the same as permission to share. Being able to open a workplace document does not automatically mean you may paste it into an external assistant. If permission is unclear, stop and use your organisation’s approved guidance or ask the responsible person.

Use the minimum information needed. A formatting exercise does not need real names, contact details or customer histories. For this course, the safest practice input is synthetic data: material invented for the exercise, rather than copied from real records.

Removing names alone is not a reliable privacy solution. A job title, unusual incident, date or location may still identify someone. When you do not need a real record, replace the whole scenario with an invented example. Do not paste the real record into AI and ask the assistant to anonymise it; you would already have shared it.

Worked input decision

Unsuitable learning input:

“Here is our real attendee spreadsheet, including contact details and access needs. Use it to practise writing a progress update.”

The task does not require those details, and some may reveal sensitive information. Do not enter the spreadsheet. Use the supplied fictional notes instead.

Suitable learning input:

“Fictional practice note: The team has drafted a workshop handout. A colleague still needs to review it.”

This lets you practise the same writing skill without exposing a real person’s information.

Privacy settings do not replace permission

Controls for model training, retention and personalisation serve different purposes. A training opt-out does not mean information is never stored, never accessible or approved for your workplace use. Temporary chat is not a guarantee of confidentiality either. OpenAI’s guidance explains that temporary chats may be retained for safety purposes. Check the current data controls guidance (https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt) and temporary chat guidance (https://help.openai.com/en/articles/8914046-temporary-chat-in-chatgpt). Other providers have their own policies.

For this module, changing a privacy setting is unnecessary: use only the invented inputs provided.

Lesson 4 Make a task decision before prompting

Suggested time: 5 minutes.

Use these three decisions:

- Proceed: The tool and input are permitted, the task is low-risk and bounded, and you have evidence and a human reviewer.
- Change the input: You can keep the useful task but replace unnecessary or unsuitable material with a safe input before sharing anything.
- Stop and ask: Permission is unclear, the task needs an authorised or specialist decision, or you cannot adequately check a consequential result.

“Change the input” is not permission to share a shortened confidential document. It means preparing an independently safe alternative, such as the fictional material in this course. If you cannot establish that the replacement is permitted, stop and ask.

A task decision path

Illustration: Five checks before using AI: permission, sensitivity and minimum data, consequence of error, evidence, and human review. Any unresolved check leads to changing the task or stopping before data is entered.
See the accompanying decision-tree diagram.

Text equivalent of the diagram

1. Do I have permission to use this tool and share this input? If no or unsure, stop and ask before entering it.
2. Does the input avoid unnecessary sensitive or confidential information? If no, replace it with a permitted, minimal or wholly fictional input; then start the checks again. If a safe replacement is not possible, stop.
3. Is the consequence of an error low enough for this exercise? If no, reduce the task to a safe draft or seek the appropriate authorised process.
4. Do I have evidence to check the result? If no, obtain the evidence or narrow the task. Do not present an unsupported factual conclusion as checked.
5. Is a human responsible for review before use? If no, arrange review. If all five checks are satisfied, proceed with the bounded draft.

Apply it to the workshop

The supplied notes are invented and intended for this exercise. They contain no real personal details. The task produces a practice draft, not an event booking. Each bullet can be compared with a note. You are the reviewer. Decision: proceed.

Compare a different request: “Use a private employee complaint to decide who should lose their job.” This involves sensitive material and a consequential decision. Do not enter the complaint or delegate the decision to a general-purpose chat. Stop and follow the appropriate authorised process.

Lesson 5 Review the answer before accepting it

Suggested time: 5 minutes.

The first response is a candidate draft. A human review means comparing the output with the source and making an explicit decision about whether it is usable.

For a short output, check each bullet using three questions:

1. Where does this statement come from?
2. Did the response keep important conditions such as “pending,” “proposed” or “not yet reviewed”?
3. Did it add a fact, reason, person, date, promise or judgement that the source does not support?

You may ask AI to help locate a problem, but a second AI response is not independent verification. Asking “Are you sure?” does not replace looking at the original note yourself.

Full worked example

Fictional source notes:

- N1: The room request has been sent. The venue has not confirmed availability.
- N2: The handout draft is complete. Review is still pending.
- N3: A practice session has been suggested. No date has been agreed.

Copyable prompt:

This is a fictional learning exercise. Turn the notes below into exactly three short bullets for an internal progress draft. Use only the notes. Keep pending and unconfirmed details clear. Do not add dates, owners or promises. Put the supporting note label at the end of each bullet.

N1: The room request has been sent. The venue has not confirmed availability.
N2: The handout draft is complete. Review is still pending.
N3: A practice session has been suggested. No date has been agreed.

Illustrative response, not a record of a live tool run:

- The room has been requested; venue availability is still unconfirmed. [N1]
- The handout draft is complete and awaiting review. [N2]
- A practice session has been suggested, with no date agreed. [N3]

The labels make checking easier, but a label is not proof by itself. Read N1 and confirm that the first bullet preserves both the request and the lack of confirmation. Repeat for N2 and N3.

Illustrative weak response:

- The room is booked. [N1]
- The handout is ready to distribute. [N2]
- The practice session will happen next week. [N3]

All three labels point to real notes, but none of these claims is supported as written. Booking is unconfirmed; review is pending; no session date exists. Correcting the wording restores the source meaning. Your review decision would be: “Revise before accepting.”

Guided practice Choose and check

Suggested time: 15 minutes.

Part A Sort six tasks

Spend about 7 minutes on this part. For each task, choose proceed, change the input, or stop and ask. Write one reason. Use the stated circumstances; do not assume extra permissions.

1. Use the supplied fictional workshop notes to create three bullets for your own review in an approved chat tool.
2. Practise formatting an attendee list by pasting real names, phone numbers and access needs into the chat. The exercise does not require real records, and permission to share them has not been established.
3. Paste a confidential venue contract into a personal AI account and ask whether the team should sign. You do not know whether your organisation allows this.
4. Rewrite this supplied fictional sentence in plain language: “The handout draft is complete; review is pending.” Keep the meaning and check the response yourself.
5. Practise summarising a complaint by pasting a real colleague’s identifiable complaint. A wholly invented complaint would teach the same skill. Permission to share the real complaint has not been established.
6. Ask the assistant to confirm that the venue is booked, although the only evidence says that a request was sent. You need to report a confirmed booking to the coordinator.

Use this decision-sheet format:

Task number:
Decision:
Reason:
Safe next step, if needed:

For tasks where you choose “change the input,” describe the replacement before using the assistant. Do not paste the original information first.

Part B Run one small task

Spend about 8 minutes on this part.

Use this new fictional source pack:

- P1: The workshop invitation draft is complete. The coordinator has not approved it.
- P2: The team has requested 20 chairs. The venue has not confirmed the quantity available.
- P3: The welcome script needs a reader. No reader has been assigned.

1. Write one sentence defining your input, output and reviewer.
2. Copy the prompt below into an approved ordinary text chat. Keep the source pack unchanged.
3. Save the response in your own practice document.
4. Read each bullet alongside its source note. Correct unsupported details yourself or request a revision and check that revision.
5. Record one fact you checked, its source label and the result of your check.

Copyable prompt:

Create exactly three short bullets for an internal workshop progress draft, using only these fictional notes. Keep unapproved, unconfirmed and unassigned details clear. Add the supporting note label to each bullet. Do not invent missing information.

P1: The workshop invitation draft is complete. The coordinator has not approved it.
P2: The team has requested 20 chairs. The venue has not confirmed the quantity available.
P3: The welcome script needs a reader. No reader has been assigned.

Check record:

Claim I checked:
Source label and relevant words:
Supported, needs correction, or unresolved:
Correction, if any:
Final decision: accept as a practice draft, or revise before accepting.

If you are practising without a live chat, audit this illustrative response instead:

“The invitation is approved. Twenty chairs have been requested, but availability is unconfirmed. The welcome script still needs a reader.”

Use the same source checks and correct it into three labelled bullets. Record that you reviewed the supplied response rather than claiming a live run.

Independent assignment Prepare a safe mini update

Suggested time: 10 minutes.

Now apply the method without copying the earlier prompt word for word.

Your new fictional input

- A1: The team has prepared the workshop sign-in sheet. It has not been reviewed.
- A2: A projector has been requested. The venue has not replied.
- A3: The team suggested a ten-minute welcome. Its length has not been approved.

Your task

Prepare a three-bullet internal progress draft from A1 to A3. Write your own short prompt, inspect the response and correct it where needed. Use only the supplied facts. Do not send the draft anywhere.

Save the following as your task decision sheet:

1. Your six decisions and reasons from guided practice.
2. One sentence identifying the independent task’s input, output and human reviewer.
3. Your input-safety decision and reason.
4. Your prompt and initial response, or a clear note that you used the offline route.
5. The final checked three-bullet output, with a source label for each bullet.
6. One specific check you performed, any correction, and your final accept-or-revise decision.

Offline route: write your own first draft directly from A1 to A3, then check it using the same method. Mark it as your draft. Complete the live prompt step later if you want practice operating the tool.

Why this assignment matters

The notes change, but the responsibility stays the same: choose a bounded task, protect the input and preserve what the evidence actually says. Your evidence sheet shows how you reached the result. Matching the example wording or making the text sound impressive is unnecessary.

Self-assessment rubric

This is formative practice. Use the criteria below to find and correct gaps; there is no separate numerical pass score for this module.

- Task decisions: All six tasks have a decision and a reason that addresses the stated risk. A different category is acceptable when the explanation leads to an equally safe next step.
- Bounded task: The input, three-bullet output and human reviewer are explicit.
- Safe input: Only the supplied fictional material is used, and no unauthorised sharing occurs.
- Faithful output: Every bullet has a supporting label; pending review, missing confirmation and unapproved duration remain clear. No invented facts remain.
- Visible review: At least one concrete claim is checked against the relevant words in the source; corrections and the final decision are recorded.

For each criterion, mark “met” or “needs correction.” Correct any privacy or accuracy gap before marking the work complete. If the issue is a missing caveat, revise that bullet and explain the change; you do not need to repeat unrelated successful work.

Self-check Before reading the answers

Suggested time: 5 minutes to answer, then 5 minutes to review the solutions below.

1. In one sentence, what is the difference between a prompt and a response?
2. An assistant labels a claim “[P2]”. Does that prove the claim is supported? What must you do next?
3. The source says “a practice session has been suggested.” The draft says “a practice session is scheduled.” What changed?
4. Your training setting is off. Does that alone authorise you to paste a confidential work document? Explain.
5. You remove a person’s name from a real complaint. Is the remaining text automatically safe to paste? Explain.
6. Which is the better first task: “plan and run the entire workshop” or “rewrite three fictional notes into three checked bullets”? Give a reason.
7. The source gives no owner for a task. What should your output say?
8. An assistant says it is certain its answer is accurate. What evidence would you use to check our workshop update?

Answers and explanations

Read this section after attempting the practice and self-check. These are reference answers. Different wording is acceptable when it keeps the meaning and safety boundary.

Guided practice Part A

1. Proceed. The input is supplied fictional material, the tool is approved, the task is small and you will review the output.
2. Change the input. Replace the entire real list with invented entries, or use the supplied fictional notes. Real contact details and access needs are unnecessary for formatting practice. Do not enter them first. “Stop and ask” also earns credit if you explain that no real data should be shared until permission is clear.
3. Stop and ask. Both permission to share a confidential contract and the authority to make a signing decision are unresolved. Use the organisation’s approved process and an appropriate reviewer. A privacy toggle does not resolve either issue.
4. Proceed. The sentence is fictional and the task is a small rewrite with a source to check. “The handout draft is finished and still needs review” preserves its meaning.
5. Change the input. Write a wholly fictional scenario without using the real complaint. The exercise does not need identifiable material. “Stop and ask” is also acceptable if no real complaint is entered and you identify the permission problem.
6. Stop and ask. Obtain confirmation from the responsible source through the authorised process before reporting a booking as confirmed. You may instead change the scope to a draft that explicitly says “confirmation pending.” AI cannot supply the missing confirmation merely by asserting it.

Guided practice Part B

Example task sentence:

“I will use fictional notes P1 to P3 to create a three-bullet progress draft, and I will check each bullet against those notes.”

Illustrative acceptable output:

- The invitation draft is complete and still awaiting coordinator approval. [P1]
- Twenty chairs have been requested; the available quantity is unconfirmed. [P2]
- A reader is needed for the welcome script; no one has been assigned. [P3]

Example review record:

“Claim checked: twenty chairs have been requested, but availability is unconfirmed. P2 says ‘requested 20 chairs’ and ‘has not confirmed the quantity available.’ Supported. I have checked the other two bullets against P1 and P3. Accept as a practice draft.”

For the offline sample, “The invitation is approved” is incorrect. P1 says the coordinator has not approved it. Correct the first sentence and separate the output into three labelled bullets. No other invented facts need to be introduced.

Independent assignment

Example safe-input decision:

“Proceed. A1 to A3 are supplied fictional notes. The task is a short practice draft, the source evidence is available and I will review it.”

Illustrative acceptable final output:

- The sign-in sheet is prepared and awaiting review. [A1]
- The projector has been requested; the venue has not replied. [A2]
- A ten-minute welcome has been suggested, but its length is not approved. [A3]

Example check:

“I checked whether ten minutes was agreed. A3 says it was suggested and not approved. I kept both qualifications. Accept as a practice draft after checking A1 and A2 as well.”

Common corrections:

- “Sign-in sheet ready for use” skips the outstanding review. Say it is prepared but unreviewed.
- “Projector secured” turns a request into a confirmed resource. Preserve the lack of a reply.
- “The welcome will take ten minutes” turns a proposal into a commitment. Keep the duration unapproved.

A response that asks for the missing confirmation rather than inventing it is sensible, but the final assignment still needs three source-faithful bullets. No real outreach is part of this exercise.

Self-check answers

1. A prompt is the instruction or question you give the assistant; a response is the content it returns. Source material can be included in the prompt.
2. No. Read P2 and compare its actual words with the claim. Labels help you locate evidence; incorrect claims can still have labels.
3. A suggestion became a commitment. “Scheduled” implies an agreed arrangement that the source did not supply.
4. No. Training controls do not establish permission, eliminate all retention or guarantee confidentiality. Workplace policy and the data’s sensitivity still matter.
5. No. Other details can identify the person, and the complaint may remain confidential. Use a wholly fictional example for this exercise.
6. The three-bullet task is better bounded. Its input, output and review evidence are clear, and mistakes can be corrected before use.
7. If the source merely omits the owner, say “owner not provided.” Say “no owner has been assigned” only if the source explicitly states that. Do not guess a person from context.
8. Compare each statement with the supplied original notes. The assistant’s assurance is not an independent source of evidence.

Recap and next step

You have practised choosing a useful task before prompting, using safe input, and checking a small result. Keep your task decision sheet: it shows both what you produced and why you accepted it.

Before any similar task, ask: “May I share this? Is the task bounded? What evidence will I use? Who will review the result?”

In Module 2, Give AI a clear brief, you will build on this foundation. You will specify the task, audience, source, constraints and output format, then improve one weakness at a time. Bring the short prompt you wrote for the independent assignment.

Further reading

The lesson is self-contained. These official sources support the product-specific privacy and accuracy points and were checked on 12 October 2026. Controls and policies can change.

- OpenAI — Does ChatGPT tell the truth (https://help.openai.com/en/articles/8313428-does-chatgpt-tell-the-truth)
- OpenAI — Data controls in ChatGPT (https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt)
- OpenAI — Temporary chat in ChatGPT (https://help.openai.com/en/articles/8914046-temporary-chat-in-chatgpt)
