Prompt engineering course for NZ business teams

Who should learn prompt engineering for business?

Staff who already use an AI chat tool but get inconsistent drafts: administrators, operations coordinators, marketers and team leaders. This is workplace prompt writing for non-developers. The exercises focus on Claude, with a checklist you can also apply when comparing other approved tools.

A vague request leaves the model to guess your audience and the information it should use. In this workshop you turn a normal business job into a prompt someone else can follow. You test the request on an ordinary case and an awkward one, then revise it against a written checklist.

The worked case is a fictional supplier proposal. Staff extract the proposed work, exclusions and missing information, then draft a decision brief. The source contains an ambiguity on purpose. The exercise succeeds when the draft keeps that uncertainty visible and the reviewer can trace every claim.

Participants bring a cleared task of their own for the final lab. We agree its scope beforehand and use public or invented substitutes where confidentiality prevents working with the original. You keep the resulting prompt and the test cases that checked it.

Before you start: Basic familiarity with Claude or another approved AI chat account. You should be able to send a prompt and review a response. Confirm the accounts your group can use with us before booking; coding is not required.

How do you build and test a useful business prompt?

1. Define the job before writing instructionsChoose the reader and the decision the output will support. Write acceptance criteria, such as a summary with a source reference for each claim. Decide what a reviewer needs to see and which parts the model must leave unresolved.
2. Supply context without exposing private informationSeparate task instructions from the source material. Replace identifying detail in the supplier brief and check whether the remaining context is safe to share. State which documents are permitted evidence and ask the model to flag missing inputs.
3. Specify a format and show a useful exampleRequest a decision brief with separate proposed work, exclusions and open questions. Compare free-form text with the requested structure. Provide an invented example of the desired output and check whether the model copies its facts into the new case.
4. Test a prompt against an awkward caseRun the same prompt on an incomplete proposal with conflicting dates. Ask for uncertainty to remain explicit. Compare the result with the acceptance criteria and revise instructions that let unsupported promises slip through.
5. Check sources and resist instructions inside a documentAsk for the passage behind each claim and read it yourself. Include a harmless instruction embedded in the sample proposal, then check whether the output follows the user task. Treat document text as evidence and stop if it asks for unrelated actions.
6. Build a reusable prompt with a colleagueWrite inputs, instructions and a review checklist for your approved task. Give the prompt to a colleague with a new sample and no spoken explanation. Use their questions to improve the template, then record where it should not be used.

We adjust the content to your team, and use your own documents and tasks for the exercises.