Start with an output someone needs
“Using AI” is not a job description or an outcome. Begin with a piece of work: a meeting summary, an interview outline, a first draft of an event FAQ or a comparison of public documents. Define the audience, the permitted inputs and what a correct output should contain. This makes it possible to judge whether a tool helps, rather than treating a fluent answer as evidence that the work is finished.
A useful first exercise is to give a tool a short public event brief and ask it to turn the brief into a checklist. Keep the source beside the result. Mark which items came directly from the brief, which are suggestions and which details are still missing. Do not let the tool silently fill missing dates, prices, venues or contact details. A visible question is better than a confident invention.
Make checking part of the task
For research, verify a cited source actually exists and supports the sentence attached to it. For content, compare names, numbers and dates against the approved brief. For translation, review tone and meaning with someone who understands the audience. A second AI answer is not independent confirmation. Keep a small review log so another person can see what was checked and what remains uncertain.
Treat permissions as an input constraint. Use public, fictional or explicitly approved material for practice. Do not paste participant contact lists, children’s information, private applications or unpublished commercial plans into a tool merely because it is convenient. If you are unsure what a project permits, ask the responsible person before uploading the material. The tool’s output does not remove your responsibility for how the input was handled.
Show the judgement, not only the prompt
In a portfolio, explain the original task, the tool’s role, the errors you caught and the final review. Show a brief before-and-after example if permissions allow. Distinguish generated material from your own analysis. If you claim time savings, record a comparable baseline and include the time spent checking and correcting the result. Without that record, describe the workflow rather than inventing a productivity percentage.
The career question is therefore practical: can you produce accurate, useful work and explain how you checked it? AI can be one part of that process, while briefing, editorial judgement, communication and accountability remain visible responsibilities. Choose a small task, test it carefully and keep the human review step explicit. Do not delegate publishing, outreach or consequential decisions to a tool simply because it can generate the next action.