A practical ChatGPT workshop for creating clearer, more useful workplace outputs with confidence, care, and good judgement.
Build prompts that give ChatGPT the right task, context, audience, format, limits, examples, and quality checks.
Apply each technique to everyday workplace scenarios, including writing, planning, summaries, meetings, and decision support.
Create prompt templates that are clear, safe, reusable, and easy to adapt when you are back at work.
Describe what makes a workplace prompt clear, complete, and useful.
Use ChatGPT safely for practical, low risk workplace tasks.
Apply zero shot, role, constraint, few shot, decomposition, iterative, and critique prompting.
Improve prompts by spotting missing context, unclear instructions, and hidden assumptions.
Review ChatGPT outputs for accuracy, tone, risk, usefulness, and next steps.
Build a reusable prompt playbook for recurring workplace tasks.
Personal data, client records, employee records, confidential strategy, legal advice, commercial secrets, passwords, or restricted files unless your organisation has clearly approved that use.
Check whether you are using a personal account, ChatGPT Business, Enterprise, or another approved workspace. The privacy settings and data controls may not be the same.
Use fictional scenarios, public information, anonymised extracts, supplied practice notes, and safe training examples.
If you would not be comfortable putting the information on a training room wall, do not paste it into a live tool during the workshop.
You will write prompts, test them, improve them, and compare the results using realistic workplace tasks.
Each technique is introduced briefly, then applied straight away to a task that could happen at work.
You will note what worked, what needs checking, and how the prompt pattern could be reused.
Draft a short update for a project sponsor explaining that the reporting deadline has changed.
Do not try to perfect it yet. This first version is here to reveal assumptions.
A vague prompt can still produce a polished answer. The real test is whether the answer is accurate, appropriate, and ready for the workplace purpose.
| Element | Why it matters | Example phrase |
|---|---|---|
| Task | The job you want completed | Summarise, rewrite, compare, draft, extract, plan |
| Context | The background, limits, and priorities | This is for an SME operations meeting |
| Audience | The right tone, detail, and language level | For a non technical senior sponsor |
| Format | The structure you need at the end | Use a table with risks, causes, and actions |
| Checks | How uncertainty and missing information should be handled | Ask questions if the notes are insufficient |
Write an email about the meeting.
Draft a warm, professional email of around 150 words for department managers. Summarise tomorrow's planning meeting, include the purpose, two preparation requests, the deadline, and one clear next step. Use only the facts below and flag anything missing.
| Technique | Best used for | Be careful with |
|---|---|---|
| Zero shot | Simple, familiar tasks | May rely on assumptions you have not given |
| Role and context | Perspective, tone, and level of detail | A role does not create real authority or expertise |
| Constraints and format | Clearer outputs and consistent structure | Too many constraints can make the answer rigid |
| Few shot | Matching a pattern, tone, or style | Examples can accidentally include sensitive details |
| Iterative refinement | Improving quality in sensible steps | Can drift if you do not track what changed |
| Critique prompts | Finding weaknesses before rewriting | Still needs human judgement |
The task is low risk, familiar, and easy for you to review yourself.
Summarise this public article into five bullet points for a busy operations manager. Keep each point clear and under 18 words.
Check whether the output is accurate, specific, and appropriate for the intended reader.
Act as a careful internal communications adviser. Help me make this update clearer for frontline managers without adding new facts.
Act as a lawyer and approve this contract.
This creates false confidence. ChatGPT can help draft questions or explain issues in plain English, but a qualified person must own legal review.
Use roles to set a useful review lens, such as communications adviser, risk reviewer, project coordinator, customer support coach, or plain English editor.
| Context type | Why it matters | Safe example |
|---|---|---|
| Business setting | Shapes tone, priorities, and level of detail | Small manufacturer with 35 staff |
| Audience | Guides depth, language, and emphasis | Senior sponsor who wants risks and decisions |
| Known facts | Reduces unsupported invention | Use only these five bullet points |
| Boundaries | Makes clear what to avoid | Do not name customers or imply approval |
| Success criteria | Gives the output a clear quality target | Clear enough for a manager to act today |
Around 150 words, five bullets, one page, or three options.
Table, checklist, decision log, agenda, RAG status, or email.
Professional, warm, concise, accountable, and plain English.
Do not add facts. Do not promise outcomes. Flag uncertainty clearly.
Rewrite the update in under 120 words. Include one sentence of context, three bullet points, and one clear request. Keep the tone professional, warm, and direct.
| Workplace need | Useful format | Helpful prompt phrase |
|---|---|---|
| Compare suppliers | Weighted decision table | Use columns for cost, risk, quality, support, and recommendation |
| Run a meeting | Agenda with timings | Create a 45 minute agenda with outcomes and preparation requests |
| Improve a process | Checklist | Turn this into a step by step checklist for a new starter |
| Report progress | RAG summary | Separate green, amber, and red items, then give one action for each |
Use few shot prompting when you need a consistent tone, format, categorisation pattern, or transformation style.
Examples are especially useful for HR templates, service replies, status updates, report summaries, and learning resources.
Use the two fictional examples below as the style pattern. Convert the new fictional note into the same format. Do not copy names, figures, or unsupported details.
Ask ChatGPT what information it needs before drafting.
Ask for a structure, decision criteria, or outline first.
Create the first output using agreed constraints.
Check against criteria, risks, tone, missing facts, and next steps.
Do not draft yet. First list the questions you need answered, then propose a sensible structure.
| Revision move | Prompt example | What it improves |
|---|---|---|
| Narrow | Make this shorter and keep only decisions and actions | Focus |
| Shift audience | Rewrite for a senior sponsor who has five minutes | Relevance |
| Add quality bar | Make the next version more specific and less generic | Usefulness |
| Preserve facts | Do not add new facts. Only improve clarity. | Safety |
| Ask for options | Give three versions: concise, warmer, and more direct | Choice |
Make this better.
ChatGPT may rewrite the text without showing what actually needed to improve.
Review this draft against clarity, accuracy, audience fit, risk, and actionability. Do not rewrite yet. List the three most important improvements and explain why each one matters.
Critique first makes quality visible. It also helps learners challenge the output instead of accepting the first polished version.
List any claims, figures, dates, assumptions, or missing context in this answer that a person should verify before using it.
Check source documents, policies, subject matter owners, dates, names, figures, and organisational expectations.
Keep a short note of what was checked, what changed, and what remains uncertain.
| Test | Question | Evidence |
|---|---|---|
| Accurate | Have facts, dates, figures, and assumptions been checked? | Source notes or human verification |
| Relevant | Does it answer the actual workplace need? | Matches the audience and decision point |
| Clear | Can the reader act without extra explanation? | Plain language and visible next steps |
| Safe | Does it avoid sensitive data and overclaiming? | Boundaries and review notes |
| Owned | Who is responsible for the final use? | Named human owner or approval step |
Three customer orders are late. Two suppliers have missed delivery dates. The office manager has handwritten notes from recent calls. The owner needs a calm update and a practical action list by 3pm.
You are helping an SME operations manager. Turn these fictional notes into an action plan with owners, risks, customer messages, and questions to resolve. Keep the tone calm, practical, and clear.
Did ChatGPT turn messy notes into a usable plan?
Which facts, owners, or deadlines were missing?
Could the output promise too much to customers or suppliers?
Who reviews the plan and sends the final message?
Needs clear actions, dates, and where to get support.
Needs decision points, risks, mitigations, and confidence level.
Needs dependencies, handovers, owners, and working assumptions.
Rewrite the same fictional update for these three audiences. Keep the facts the same. Change only the level of detail, structure, and tone.
| Audience | Useful emphasis | Tone guardrail |
|---|---|---|
| Frontline team | What changes, what to do, and who to contact | Clear and reassuring, not vague |
| Senior sponsor | Impact, risk, options, and decision needed | Direct and evidence led, not defensive |
| Customer | What happened, what happens next, and the support route | Warm and accountable, not overpromising |
| Technical partner | Dependencies, assumptions, dates, and owners | Precise and practical, not overloaded |
Create an agenda for a project meeting.
Create a 45 minute agenda for a project recovery meeting. The goal is to agree the top three risks, owners, and next actions. Include timings, preparation tasks, facilitation questions, and a decision log template.
Ask ChatGPT to create the agenda, then critique whether each item supports a decision or simply creates more discussion.
| Prompt detail | Why it matters |
|---|---|
| Relationship to reader | A customer, colleague, supplier, or sponsor may need a different level of context. |
| Purpose | The email may need to inform, request, escalate, apologise, or confirm. |
| Tone | Warm, accountable, concise, firm, or neutral wording can change the effect. |
| Boundaries | Avoid unsupported promises, blame, private data, and invented facts. |
| Call to action | Make the next step visible and time bound. |
Using only the notes below, create a one page report summary with confirmed facts, likely implications, risks, open questions, and recommended next steps. Label any assumptions clearly.
This format helps ChatGPT separate what is known from what is inferred. It also helps the reviewer spot gaps before the report is shared.
| Useful request | Example | Where human judgement is needed |
|---|---|---|
| Generate criteria | What criteria should we use to compare these options? | Confirm the criteria match business priorities |
| Compare options | Use a weighted table and show trade offs | Check the weights and evidence |
| Identify risks | What could go wrong with each option? | Add local context and accountability |
| Prepare a recommendation | Draft a recommendation with caveats | Approve, reject, or revise before use |
A reusable structure with placeholders for task, context, audience, format, constraints, and checks.
When to use it, when not to use it, and what safe material it requires.
How the output should be checked before it is used at work.
| Template field | Example content |
|---|---|
| Use case | Turn meeting notes into actions and decisions |
| Prompt | Use the notes below to create a decision log, action list, risks, and open questions |
| Inputs | Fictional or approved notes, meeting aim, audience, and date |
| Output format | Table with decision, owner, deadline, evidence, and status |
| Review checks | Verify names, dates, owners, commitments, and missing context |
Before using this project, restate the standing instructions you are applying and list any files or context you relied on.
Availability and sharing controls may vary by plan and workspace.
Longer documents, drafts, code, review cycles, and work where selecting a section for targeted edits is helpful.
Use canvas for this draft. After creating it, suggest edits for clarity, tone, and missing evidence before rewriting.
Use only approved or safe content. Workspace admins may control canvas features and network access.
| Instruction type | Good example | Caution |
|---|---|---|
| Voice | Use professional, warm, plain English workplace language | Do not let tone hide uncertainty |
| Formatting | Prefer tables for comparisons and checklists for actions | Change format when the task needs it |
| Risk | Flag assumptions, missing facts, and privacy concerns | Still verify independently |
| Role | Act as a careful workplace writing and planning assistant | Do not imply formal approval |
Add audience, context, examples, and success criteria.
Set length, format, and priority rules.
Name the reader, relationship, and desired voice.
Say "use only provided facts" and ask it to flag assumptions.
| Failure mode | What it looks like | Prompt protection |
|---|---|---|
| Hallucination | Invented facts, sources, names, or figures | Use only supplied facts. Flag uncertainty. |
| Overconfidence | Strong claims without evidence | Separate facts, assumptions, and recommendations |
| Bias | Unfair assumptions about people or groups | Ask for fairness risks and alternative interpretations |
| Context loss | Output ignores key constraints | Restate constraints and ask for a compliance check |
| Scope creep | Answer grows beyond the task | Define output length and decision boundary |
Choose one recurring workplace task. Create a first prompt, test it in ChatGPT, revise it twice, and record what changed in the output.
Finish with one reusable template and one review checklist.
Can another person see exactly what ChatGPT has been asked to do?
Does the prompt avoid sensitive data and risky overreach?
Does the result help with a real workplace decision, action, or communication?
"The strongest part is... The biggest risk is... One change I would make is..."
| Evidence item | What it should show |
|---|---|
| Three prompt templates | Recurring workplace tasks with placeholders and boundaries |
| Test notes | How the prompt changed after live ChatGPT testing |
| Quality checks | Accuracy, tone, safety, actionability, and ownership review |
| Reflection | What you will use, what you will avoid, and what support you need |
Ask clarifying questions and identify missing facts.
Create a first output using clear audience and format rules.
Review against a rubric before revising.
List checks, decisions, and the human owner before workplace use.
Design a repeatable ChatGPT workflow for preparing a weekly management update. Include intake questions, drafting prompts, critique prompts, verification checks, and stop conditions.
Be clear about the task, context, reader, format, and risk.
Compare outputs, ask for critique, and verify facts before use.
Turn strong prompts into templates, playbooks, and team habits.
Use one prompt template on a low risk task. Compare the output with your normal approach and record what changed.
Share one prompt pattern and one safety rule with a colleague. Ask what they would add before using it in your organisation.
ChatGPT can help you draft, structure, compare, and improve. A person still owns the decision, the quality, and the consequences.