GenAI for research, writing and analysis
Bring it all together: practical workflows for the three things people use generative AI for most — and for briefing AI agents.
Learning objectives
- Use AI as a research assistant and verify its sources
- Apply an outline-draft-edit writing workflow
- Use AI to structure and stress-test decisions
- Write effective briefs for AI agents
- Diagnose and fix common prompt problems
6.1 Research
AI tools with web search or "deep research" features can gather and synthesise sources quickly. Use them as a research assistant, not an oracle.
I'm researching [topic] for [purpose/audience].
1. Find recent, authoritative sources (government, regulators, peer-reviewed
research, major industry bodies). Prefer sources from the last 2 years.
2. For each key finding, give the source name, date and link.
3. Separate well-established findings from contested or emerging ones.
4. Note any important gaps or limitations in the evidence.Then open and check the key sources yourself — confirm they exist, are current and actually say what's claimed.
6.2 Writing
- Outline first: agree the structure before asking for a full draft.
- Feed it your voice: paste two or three examples of writing you like and ask it to match the style.
- Edit in passes: structure → clarity → tone → length → proofreading.
- Make it yours: add specifics, judgement and examples only you have. Readers notice generic AI prose.
Editing prompt
"Edit this for clarity and concision. Keep my voice. Don't add new claims. Show changes as a bulleted list, then the full revised text."
Feedback prompt
"You're a demanding editor. Without rewriting, give me the five most important improvements to this draft, in priority order."
6.3 Analysis and decision support
- Frame the decision: "We're deciding whether to X. Criteria: cost, risk, time to launch."
- Ask for structure: options, pros and cons, assumptions, risks, what evidence would change the answer.
- Stress-test: "Argue the opposite case." "What would a sceptical investor ask?"
- Keep the decision human: AI can widen and sharpen your thinking; accountability for the decision stays with you.
6.4 Delegating to agents
With agents and agentic features — which can browse, use files and take actions — your prompt becomes a brief. A good brief includes:
- The goal and what "done" looks like
- Resources and where to find them
- Boundaries: what it may and may not do (e.g. "draft emails but don't send")
- When to stop and check in with you
- How to report back (summary, sources, open questions)
6.5 Responsible prompting
Everything from the AI Literacy course still applies: approved tools, no confidential or personal data unless permitted, verify facts, disclose AI use where it matters, and stay accountable for the result.
6.6 Your prompt toolkit
| Problem | Try |
|---|---|
| Generic, bland output | Add context, audience and examples; ask for a specific angle |
| Wrong format | Show the exact format; use tags or a template |
| Made-up facts | Supply sources; "answer only from the document"; allow "I don't know"; verify |
| Weak reasoning | Ask it to think step by step, or use a reasoning model; break into a chain |
| Too agreeable | Ask neutral questions; request counter-arguments and critique |
| Inconsistent results | Turn it into a template; test with a set of inputs |
Choose a real, recurring task. Write a template using the six ingredients, add one example, and include an "if unsure" instruction. Test it on five realistic inputs including one tricky case. Revise once. Save it in your prompt library.
Key takeaways
- AI is a research assistant, not an oracle — check key sources yourself.
- Outline, draft, edit in passes, and add what only you know.
- Use AI to widen and stress-test thinking; keep decisions human.
- Agent briefs need goals, boundaries, check-in points and reporting.
- Match the fix to the problem.