Workshop: AI-Powered Research
A 75-minute workshop for students and professionals. Learn to frame research questions, generate AI-driven research plans, synthesize multiple perspectives, check for hallucinations, and write a polished summary — all using a repeatable 5-prompt workflow.
AI can accelerate every stage of the research process — from scoping a question to drafting the final summary. But speed without rigour is dangerous: hallucinated citations, one-sided framing, and shallow analysis are real risks. This workshop teaches a structured 5-step workflow that pairs AI speed with human critical thinking, so every output is trustworthy enough to share.
Before You Start
- ✓An active account on at least one AI chatbot (ChatGPT, Claude, Gemini, etc.)
- ✓Familiarity with basic prompting (Workshop 1 or equivalent experience)
- ✓Complete Module 2: AI Foundations — especially the section on hallucinations (/modules/ai-foundations)
Steps
Good research starts with a good question. A vague topic like 'AI in healthcare' leads to unfocused, surface-level answers. In this step you'll use AI to brainstorm sub-questions, identify the core tension, and lock in a single, scoped research question. The tighter your question, the better every subsequent prompt will perform.
I'm researching [BROAD TOPIC]. List 8–10 specific sub-questions that a researcher might investigate within this topic. For each, note whether it's empirical (data-driven), conceptual (definitional), or practical (how-to).
From the sub-questions above, I'm most interested in [CHOSEN SUB-QUESTION]. Rewrite it as a focused research question that is: (a) specific enough to answer in a 2-page summary, (b) has at least two competing perspectives, and (c) can be supported with evidence. Show three refined versions and recommend the strongest.
- •A good research question names the population, intervention/topic, and outcome — think PICO for any domain.
- •If you can answer the question in one sentence, it's too narrow. If it would take a book, it's too broad.
- •Write your final question down before moving on — it anchors every later prompt.
Before diving into content generation, build a roadmap. Ask the AI to outline the key angles, suggest real source types (journals, reports, data sets), and generate search terms you can use in Google Scholar, PubMed, or other databases. This step prevents the common trap of going deep on one angle and ignoring others.
My research question is: [YOUR RESEARCH QUESTION]. Create a research plan that includes: 1. Three to five key angles or themes I should investigate. 2. For each angle, suggest two credible source types (e.g. peer-reviewed journal, government report, industry white paper). 3. Provide 3 search queries per angle I can use in Google Scholar. 4. List any key terms or jargon I should know. 5. Flag potential biases to watch for in this topic.
- •Copy the search queries directly into Google Scholar or a database — AI-generated terms are a great starting point.
- •The bias flags are valuable: keep them visible as a checklist while you research.
- •Add your own angles if the AI missed something obvious from your domain expertise.
One of AI's best uses in research is rapid perspective-taking. In this step you'll ask the model to argue from different positions — proponent, critic, policymaker, affected community — and then synthesize across them. This produces a more balanced, nuanced view than a single prompt ever could.
My research question is: [YOUR RESEARCH QUESTION]. Present four distinct perspectives on this question: • Perspective A — a strong proponent (explain their reasoning and evidence) • Perspective B — a strong critic (explain their reasoning and evidence) • Perspective C — a policymaker or regulator (explain their priorities and concerns) • Perspective D — someone directly affected (explain their lived experience and concerns) For each perspective, cite the type of evidence they would rely on. After all four, write a 'synthesis note' identifying points of agreement, key disagreements, and unresolved questions.
- •If the AI's perspectives feel shallow, follow up: 'Steelman Perspective B — make the strongest possible case.'
- •The synthesis note is the seed of your final summary's argument structure.
- •Watch for the AI giving lip service to a perspective without real substance — push back.
AI models can fabricate citations, invent statistics, and state outdated information with complete confidence. This step teaches you a three-layer verification routine: ask the AI to fact-check itself, cross-reference claims with real sources, and flag anything unverifiable. Never skip this step — your credibility depends on it.
Review everything you have written in this conversation so far. For every factual claim, statistic, date, or named source: 1. Rate your confidence (High / Medium / Low) that it is accurate. 2. Flag any claim you are uncertain about with ⚠️. 3. For each flagged claim, suggest a specific way I can verify it (e.g. 'search Google Scholar for…', 'check the WHO website for…'). 4. If you cited any papers or reports by name, confirm whether they actually exist or might be fabricated.
I'm going to paste a list of claims from my research draft. For each one, tell me: • Is this a widely accepted fact, a contested claim, or something you may have fabricated? • What is a reliable source I could use to verify or replace it? Claims: 1. [CLAIM 1] 2. [CLAIM 2] 3. [CLAIM 3]
- •Never trust an AI-generated citation without checking — even well-known models fabricate paper titles and authors.
- •If the AI rates its confidence as 'Low', treat that claim as unverified until you check it manually.
- •Use the Hallucination Checker take-home asset (below) for every future research project.
With a focused question, a multi-perspective evidence base, and verified facts, you're ready to write. Use AI to draft the summary, then refine it for clarity, balance, and proper attribution. The goal is a standalone document you'd be comfortable sharing with a professor, manager, or client.
Using the research question, perspectives, and verified claims from our conversation, write a structured research summary (600–800 words) with these sections: 1. **Research Question** — state the question clearly. 2. **Background** — brief context for why this matters. 3. **Key Findings** — synthesize the main arguments and evidence from multiple perspectives. 4. **Points of Consensus & Debate** — what do experts agree on, and where do they disagree? 5. **Limitations** — what couldn't be fully answered and why? 6. **Conclusion** — a balanced takeaway with suggested next steps. Use a professional, neutral tone. Do not include any claim we flagged as unverified. Cite source types (not specific papers unless verified).
Review the summary you just wrote. Check for: • Balance — does any perspective get disproportionate space? • Clarity — would an intelligent non-expert understand every paragraph? • Attribution — is every major claim linked to a source type or perspective? • Gaps — is there anything important from our research that didn't make it in? Revise the summary to fix any issues you find.
- •Read the summary as if you're a skeptical reviewer — does every claim feel grounded?
- •Replace vague attributions ('studies show…') with specific source types ('a 2023 WHO report indicates…').
- •Save both the prompt chain and the final summary — this workflow is reusable for any research topic.
Take-Home Assets
A systematic checklist for verifying AI-generated claims. Use this on every research project before sharing your findings.
- 1.Ask the AI to rate its confidence (High/Medium/Low) on every factual claim.
- 2.Flag all Medium and Low confidence claims with ⚠️ for manual verification.
- 3.For any named citation (paper, report, book), search for it in Google Scholar or the publisher's site — confirm it exists.
- 4.Cross-reference statistics against primary sources (WHO, government data, peer-reviewed journals).
- 5.Check dates — is the information current or potentially outdated?
- 6.Look for fabricated specificity: precise percentages or dates that seem too convenient are red flags.
- 7.Ask a second AI tool the same question and compare answers — divergence signals uncertainty.
- 8.If a claim can't be verified after 2 minutes of searching, remove it or label it as 'unverified'.
- 9.Run the final draft through the self-check prompt one last time before sharing.
The complete 5-prompt sequence from this workshop. Copy each prompt in order, fill in your topic, and run the full research workflow in any AI tool.
- 1.Prompt 1 — Frame: "I'm researching [TOPIC]. List 8–10 specific sub-questions a researcher might investigate. For each, note whether it's empirical, conceptual, or practical."
- 2.Prompt 2 — Plan: "My research question is: [QUESTION]. Create a research plan with key angles, source types, Google Scholar queries, key terms, and potential biases."
- 3.Prompt 3 — Synthesize: "Present four perspectives on [QUESTION]: a proponent, a critic, a policymaker, and someone directly affected. Include evidence types and a synthesis note."
- 4.Prompt 4 — Verify: "Review all claims so far. Rate confidence (High/Medium/Low), flag uncertain claims with ⚠️, suggest verification methods, and confirm whether named sources exist."
- 5.Prompt 5 — Summarize: "Write a 600–800 word structured summary with: Research Question, Background, Key Findings, Consensus & Debate, Limitations, Conclusion. Exclude unverified claims."