Generic research prompts produce Wikipedia-level summaries. Here is how to get outputs that actually inform decisions.
Do not say "research X for me." Give the model a brief like a manager would give a junior analyst:
<aside> 🧾
Research Brief Template
TOPIC: [Specific subject] PURPOSE: [Why I need this / what decision it informs] AUDIENCE: [Who will read or use this] DEPTH: [Surface overview / working knowledge / expert depth] FOCUS: [What specific angle matters most for my purpose] OUTPUT: [Comparison table / executive summary / bullet framework / narrative] EXCLUDE: [What I already know / what is not relevant to my purpose]
</aside>
Do not say "compare X and Y." Say: "Compare X and Y on these 5 specific dimensions: [list]. For each, give a verdict: who wins and why. End with: under what conditions would you choose Y over X, and vice versa?"
Always close a research session with: "For each major claim you made, indicate whether this is well-established consensus, emerging consensus, or contested and uncertain." This prevents you from building decisions on AI-generated speculation.
After any analysis, ask: "What would change about this analysis if [key assumption] were false?" This reveals how fragile the conclusions are and what you should verify independently.