Prompt Engineering for Real-World Teams
Turn recurring team work into tested, reusable instructions with clear outputs, evidence requirements, review ownership, and change control. Build a prompt playbook for operations, sales, service, and management.
Your course plan
Study format
8-12 minutes per lesson, with a 60-90 minute capstone. Estimates include practice and assessment. Estimated total: 3-4 hours including the capstone and quiz.
Prerequisites and scope
AI 101 or equivalent familiarity with an approved AI tool. Introductory AI Prompt Engineering is recommended; no coding required.
Learning outcomes
- Task and output specification
- Reusable team prompt design
- Evidence and uncertainty handling
- Workflow review and evaluation
- Prompt ownership and versioning
Four modules, three lessons each
- Specify the team task
- Write and test reusable instructions
- Integrate responsible review
- Deliver a maintained team playbook
Completion and assessment
Complete all 12 lessons and the practice activities. Use the final-lesson rubric to review the capstone with an instructor or peer. The portal scores a separate 10-question final quiz; 80% or higher is required for the existing course-completion credential. Capstone work is not automatically graded by the portal. No academic credit or external license is implied.
Authoritative reference shelf
Technical and regulatory references reviewed October 8, 2026. Original teaching scenarios, assumptions, calculations, exercises, and rubrics are instructional examples. Verify current source requirements for an actual project or operation.
M1.1 | Start with an observable work outcome
Learning objective
Translate a vague AI request into a measurable team task.
Core lesson
A useful prompt starts with a job that someone needs completed. Name the user, input, desired output, quality requirement, and consequence of a mistake. Replace improve our communications with draft a customer reply using the supplied ticket and approved refund policy, then submit it for review. This makes the task testable. Choose repeated work with recognizable evidence and a responsible reviewer. Establish a baseline for time and quality before introducing AI. Avoid optimizing a task that should be eliminated or simplified. A prompt can accelerate drafting, but it cannot decide who has authority to commit the organization. Specify where the output enters the actual workflow and who accepts it.
Instructor teaching guidance
Begin by establishing what the learner already knows about m1.1 | start with an observable work outcome. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: select a recurring task from sales, service, operations, or management. Write a task contract with input, output, reviewer, acceptance criteria, and a prohibited action. Record the current time per task and the most frequent quality defect.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
The task contract makes success observable and assigns final responsibility.
Knowledge check
What would make an output unusable even if it sounds professional?
M1.2 | Write a clear instruction contract
Learning objective
Create a structured prompt that defines behavior and output acceptance.
Core lesson
Use plain sections: task, background, authorized inputs, constraints, output format, and review requirements. A role can help set perspective, but a title such as expert does not supply evidence or expertise. Tell the model which sources may support the answer and what to do when inputs are incomplete. For an original service example: draft a reply under 150 words; use only the ticket and policy; identify missing facts; do not promise a refund; return a draft plus a source note for the reviewer. Keep requirements consistent. Asking for extreme brevity while demanding a long investigation may create an avoidable conflict. Test the instruction with actual synthetic cases rather than judging the wording alone.
Instructor teaching guidance
Begin by establishing what the learner already knows about m1.2 | write a clear instruction contract. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: create a first version of the service prompt with six labeled sections. Test it on a complete ticket and a ticket with no purchase date. Record which requirement failed and revise only that requirement.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
A strong instruction is clear, internally consistent, grounded in available inputs, and evaluated through use.
Knowledge check
Why does assigning an expert persona fail to solve missing factual evidence?
M1.3 | Specify outputs people and systems can use
Learning objective
Design output contracts that support reliable handoffs.
Core lesson
Choose a format that matches the next step: a decision brief for a manager, a checklist for a technician, a draft for an editor, or structured fields for a system. Define required fields, permitted values, units, length, and how unknown values appear. Structure improves usability but does not make the content correct. For a meeting-action table, require action, owner, due date, source passage, and unresolved question. If a transcript contains no owner, return unassigned rather than inventing a name. For machine-readable output, validate syntax and values outside the model before saving. For human review, give enough source context to check the claim. Keep the output contract small enough to maintain.
Instructor teaching guidance
Begin by establishing what the learner already knows about m1.3 | specify outputs people and systems can use. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: convert one meeting summary request into a five-field action table and a short manager brief. Add explicit unknown-value rules. Test a transcript where the deadline is discussed but never agreed.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
The output is usable downstream, preserves unknowns, and can be checked against its inputs.
Knowledge check
What should appear in the due-date field when the team only mentioned a possible date?
M2.1 | Use examples to establish standards
Learning objective
Use examples to teach a repeatable standard without contaminating live input.
Core lesson
Examples can show tone, scope, classification boundaries, and required formatting more clearly than adjectives. Use representative examples and include an edge case. Explain why each example meets the standard so the team can maintain it. Remove sensitive details or use synthetic cases. Avoid examples that accidentally conflict with the written instruction. If all sample customer responses promise refunds, the model may imitate that pattern despite a vague caution elsewhere. Diverse cases reduce dependence on a single familiar scenario. Keep examples separate from the live input and test whether the prompt handles a new case rather than copying the example. More examples are not automatically better; relevance and consistency matter.
Instructor teaching guidance
Begin by establishing what the learner already knows about m2.1 | use examples to establish standards. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: add three synthetic examples to a support-classification prompt: ordinary, ambiguous, and out of scope. Write the reason for each classification. Test an unseen case and inspect for inappropriate copying or invented facts.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
Examples are representative, consistent with rules, and tested on unseen work.
Knowledge check
What happens if the examples imply a policy that the written rules prohibit?
M2.2 | Require evidence and visible uncertainty
Learning objective
Design prompts that support verification and useful uncertainty handling.
Core lesson
Ask for supporting source references, material assumptions, missing inputs, and a next step where evidence is insufficient. Distinguish a factual summary from a recommendation. A recommendation may combine facts with judgment, but the judgment should be identifiable. Never ask the model to invent a citation to satisfy a format. For a vendor comparison, provide approved quotes and require each price to link to a quote line or section. Missing implementation fees should appear as unknown, not zero. Ask for a concise rationale and verifiable calculation inputs; lengthy hidden reasoning is not necessary for review. Verify high-consequence facts through the actual source and responsible specialist. A confident tone should never hide an unresolved issue.
Instructor teaching guidance
Begin by establishing what the learner already knows about m2.2 | require evidence and visible uncertainty. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: create a vendor-comparison prompt using three synthetic quotes. Remove one support fee and test whether it remains unknown. Ask a reviewer to check every price and one recommendation against the evidence.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
Facts are supported, assumptions are labeled, and missing evidence is visible to the reviewer.
Knowledge check
Why does treating an unknown fee as zero distort the recommendation?
M2.3 | Separate drafting, checking, and action
Learning objective
Design a prompt workflow with independent checks and accountable action.
Core lesson
Break a complex team workflow into stages when the stages have different responsibilities. One stage extracts facts, another drafts, a validator checks constraints, and an authorized person approves the action. Each handoff has a format and acceptance condition. Do not add stages merely to make a simple task look sophisticated. For a sales proposal, extraction uses approved customer requirements and price sheets; drafting creates an offer; review checks claims and calculations; the authorized account owner approves sending. The AI draft does not itself authorize a discount or signature. Preserve source links across stages. If validation fails, return the work to the relevant stage instead of allowing an attractive draft to bypass the control.
Instructor teaching guidance
Begin by establishing what the learner already knows about m2.3 | separate drafting, checking, and action. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: draw a four-stage proposal workflow with inputs, outputs, owners, and a rejection rule. Introduce an unsupported discount into the draft and describe how the workflow catches it before sending.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
The workflow has useful stages, verifiable handoffs, and an explicit approval boundary for consequential action.
Knowledge check
When should a single prompt be retained instead of creating a chain?
M3.1 | Build a prompt evaluation set
Learning objective
Evaluate prompt changes using evidence rather than impressions.
Core lesson
Create test cases before tuning. Include ordinary requests, incomplete data, contradictory sources, edge cases, malicious input, and a case the prompt should decline or escalate. Define the expected behavior and score dimensions such as factual support, completeness, usability, tone, and boundary compliance. Repeat representative tests because model output can vary. Keep some cases separate from tuning. Compare changes on the same inputs and record configuration. In this original instructional pilot, require zero invented prices and at least nine of ten routine cases meeting the rubric before trying a limited rollout. Real teams set thresholds according to consequences. A good demonstration is one case; reliability is a pattern supported by evidence.
Instructor teaching guidance
Begin by establishing what the learner already knows about m3.1 | build a prompt evaluation set. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: build ten cases for your task contract. Define pass/fail criteria for each, then compare prompt versions A and B. Record one improvement and one regression. Preserve the failing case for future changes.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
Prompt quality is assessed on representative, repeatable cases using predetermined acceptance criteria.
Knowledge check
Why does repeatedly adjusting the prompt to pass the final test set weaken the conclusion?
M3.2 | Protect data and resist source instructions
Learning objective
Design team prompt use around data protection and enforced boundaries.
Core lesson
Use approved tools and appropriate data-handling rules. Classify information before uploading it, minimize what is shared, and avoid placing credentials or restricted records into ordinary prompts. Confirm vendor settings and organizational requirements through the responsible owner rather than assuming all AI products handle data alike. External text may contain instructions that attempt to redirect the model. Separate input content from authority and validate downstream behavior. Formatting delimiters can improve clarity, but they are not a complete security barrier. Enforce access and action permissions outside the model. Test a customer message that asks the assistant to reveal another customers records or ignore company rules. The reviewer must recognize the attempted boundary crossing.
Instructor teaching guidance
Begin by establishing what the learner already knows about m3.2 | protect data and resist source instructions. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: create a safe synthetic test in which a ticket contains an instruction to disclose unrelated private data. State what the prompt, application permissions, and human review each contribute. Write a five-rule data checklist for team members.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
Sensitive information and action authority are protected by controls beyond prompt wording.
Knowledge check
Why is a sentence saying keep everything private insufficient protection?
M3.3 | Assign human review that can change the outcome
Learning objective
Design review responsibilities that materially reduce errors.
Core lesson
A reviewer needs the source, output, acceptance criteria, time, and authority to reject or revise. Assign review intensity according to consequence: a public marketing draft and a contractual commitment require different checks. Specify what the reviewer actually verifies rather than writing human in the loop as a label. Give reviewers a short checklist and an escalation route. Track defects and correction time, because nominal time savings can disappear in rework. For a policy response, verify applicability, source date, exceptions, and wording. For a financial table, verify inputs, arithmetic, and authority. Avoid automating an approval simply because previous drafts looked good. The responsible team owner decides whether controls are adequate.
Instructor teaching guidance
Begin by establishing what the learner already knows about m3.3 | assign human review that can change the outcome. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: make a reviewer checklist for a proposal with factual, commercial, and communication checks. Run it on a deliberately flawed synthetic draft. Identify a defect the reviewer cannot resolve and name who receives the escalation.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
Review is meaningful when the reviewer can inspect evidence and stop or correct the output.
Knowledge check
How do you know whether review is improving quality or merely adding a ceremonial click?
M4.1 | Maintain a prompt library as an operating asset
Learning objective
Maintain reusable team prompts with governance and change evidence.
Core lesson
Store each approved prompt with purpose, owner, version, inputs, output contract, examples, evaluation results, permitted tools, data rules, and a review date. Give users one current version and archive retired versions. Copying undocumented prompts through chat threads creates conflicting standards. A change record should explain the defect addressed, affected behavior, test evidence, and rollback. Re-evaluate after model or source changes, because a prompt can remain identical while system behavior changes. Retire prompts that no longer serve a useful process. Include a brief user guide that shows the needed inputs and how to handle failure. Ownership makes the asset maintainable beyond its original author.
Instructor teaching guidance
Begin by establishing what the learner already knows about m4.1 | maintain a prompt library as an operating asset. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: create a one-page record for your prompt, label it version 1.0, and write a change proposal for version 1.1. Define the test gate and how the team will return to the prior version if quality declines.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
A usable library has current versions, clear owners, recorded tests, and a practical rollback path.
Knowledge check
What must be checked when the AI vendor changes the model behind an existing workflow?
M4.2 | Pilot across real team roles
Learning objective
Evaluate team adoption and operational value in a bounded pilot.
Core lesson
Select a small pilot with a bounded task, willing users, a process owner, and a defined period. Train through actual work using synthetic or approved information. Record baseline time, accepted-output quality, review time, adoption barriers, and incidents. Give users a quick route to report unclear instructions or wrong outputs. Measure net improvement, not just generated volume. Ten drafts requiring substantial correction can be worse than five accepted drafts. Interview people affected by the workflow and check whether work shifts unfairly to reviewers. End the pilot with expand, revise, or stop criteria. A successful prompt is useful only when the team can apply it consistently in its working conditions.
Instructor teaching guidance
Begin by establishing what the learner already knows about m4.2 | pilot across real team roles. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Practice: design a two-week pilot for three team members. Define a task sample, baseline, success threshold, reviewer workload limit, stop condition, and feedback meeting. Write the evidence needed before expanding use.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
The pilot measures accepted work, total effort, risk, and team usability against a baseline.
Knowledge check
Why might generated-output volume increase while team productivity declines?
M4.3 | Capstone: deliver the team prompt playbook
Learning objective
Deliver a tested prompt playbook connected to accountable team work.
Core lesson
Create three reusable prompts for one team: a fact extraction task, a draft task, and a quality review task. Use synthetic or approved sources and assign each prompt an owner. Include output contracts, unknown-value rules, examples, boundaries, and a ten-case evaluation. Show one complete handoff through the team workflow. The playbook should be readable by a colleague who did not attend the course. Include a pilot plan, reviewer checklist, version record, and rollback. Describe one case requiring escalation and demonstrate that no prompt grants itself authority to send, sign, spend, or modify a restricted record. The goal is a practical work system that can be measured and maintained.
Instructor teaching guidance
Begin by establishing what the learner already knows about m4.3 | capstone: deliver the team prompt playbook. Teach the principle in plain language, define unfamiliar terms, then connect it to the course goal. Ask the learner to explain the idea back in their own words before moving to application.
Applied exercise
Spend 60-90 minutes assembling the playbook. Rubric: task and output design 25, examples and evidence 20, evaluation 25, review and boundaries 20, maintenance and pilot 10. Target 80/100 with no invented material facts. Complete the separate scored portal final quiz.
Facilitation sequence
Guide the learner through the application in five steps: describe the present situation, identify the lesson principle, choose an action, name the evidence to review, and define how success will be measured. Use follow-up questions when an answer is vague or unsupported.
Mastery standard
The playbook can be used, checked, maintained, and improved by an actual team.
Knowledge check
Which prompt failed most often, why, and what evidence supports the revision?