A practical learning path
From a first prompt to work you can review.
Learn prompting in five stages: write a useful brief, understand model choices, organise evidence, apply a workflow and evaluate the result.
Updated
Completely new to AI? Start here
Begin with a plain-language tour: where to start, what you can ask, how to improve an answer and when AI needs tools or your approval to do work. Then return to the five stages below.
1. Write the smallest useful brief
Start with a task you can inspect: a rewrite, a meeting action list or a question about a document. State the result, relevant context, constraints and useful output format. Add detail when it changes the answer, not because a template has another box to fill.
Exercise: rewrite ‘help me with this report’ as a request naming your role, decision and three questions. Check whether a colleague could understand what a useful answer would contain.
2. Choose the environment and capability
Separate a quick conversation from a delegated deliverable and a code change. Then ask which model and tools suit the task. Do not pay for more reasoning to compensate for a missing file or an unavailable image generator.
Exercise: divide a poster task into message, image, layout and final checks. Identify the tools and assets each stage needs.
3. Give the evidence a clear role
Label factual sources, style examples and mandatory requirements. Ask for a trail from important claims to their evidence. Missing information should become a question, not a polished guess.
Exercise: provide a product sheet and a voice sample. Request a short draft, then check that the voice sample did not introduce a product promise absent from the sheet.
4. Complete one real workflow
Pick the guide matching the work, not the most impressive-sounding prompt. Adapt every placeholder, supply the required input and use the follow-up only when it addresses an observed weakness.
Exercise: create a report decision brief or a campaign concept set. Keep the source material and the raw output so you can inspect the result.
5. Review, repair and reuse
Check facts, source coverage, instruction adherence and whether the output supports the next action. Repair a specific failure while preserving what worked. Try a second input before treating the brief as reusable.
Exercise: compare two drafts against the same requirements. Record a factual error separately from a preference. Our test protocol explains what we still need before publishing a measured model comparison.
Keep exploring
About this workflow
This is an original AI-assisted editorial workflow. Illustrative scenarios are not customer case studies, and the prompts have not been systematically evaluated on live models. Check consequential outputs against your sources and use qualified review where needed.