Working with AI Tools
Getting useful work from AI tools
Treat a generative AI tool as a fast, fallible assistant. A strong prompt usually states the role or task, gives context, sets constraints (audience, length, tone, format), and asks for output in a usable structure such as a table, checklist, or numbered steps. Iterate: ask for a critique, request alternatives, or supply an example of the style you want. Break large jobs into smaller steps rather than one vague request.
Know the tool's limits. Language models can hallucinate-produce fluent but false facts, citations, or numbers-so verify names, dates, quotations, calculations, and legal or medical claims against reliable sources. Check for bias and missing context. Never paste confidential, personal, or client data into a tool unless your organisation has approved that use. Respect copyright and disclosure rules, and keep a human accountable for the final output.
Traps: a longer prompt is not automatically better; confident tone is not evidence; the tool may have a knowledge cutoff; and it may follow instructions embedded in pasted content (prompt injection). Do not confuse a polished draft with a verified answer.
Worked example. To draft a customer email, prompt: “Act as a support agent. Write a 120-word apology to a customer whose order arrived late. Mention the new delivery date, offer two options, and use a calm tone.” Then check the date, policy, and names before sending. Rule of thumb: specify task, context, constraints, and format-then verify every factual claim.