Open this when you have a topic to research and want a result you can actually trust. What follows is a four-step method, not a wall of theory: convene a panel of perspectives, map where they clash, verify every claim at its source, then aim the whole thing at your own decision. Work the steps in order — one prompt sees one angle, and a single angle leaves blind spots.
Step 1 — Convene the panel
Instead of firing off one prompt and taking one view of the topic, I have several perspectives research it separately, each from its own seat. Five default lenses, and what each one tends to catch:
- Practitioner — what actually works in the field, and where the theory breaks on contact.
- Academic — the underlying research, the definitions, and what the evidence really supports.
- Skeptic — the shaky claims, the hype, and the thing everyone is quietly taking for granted.
- Economist — the costs, the incentives, and the second-order effects nobody priced in.
- Historian — what has happened before, the pattern this rhymes with, and how it played out.
Then add one or two roles that fit your topic — a customer, a frontline employee, a regulator. Each seat finds a hole the others miss, so choose the seats that cover the parts you can't see yourself.
Paste this:
Research [topic] from five perspectives — a practitioner, an academic,
a skeptic, an economist, and a historian. Have each one work separately
and list what it sees from its seat and where the gaps are.Step 2 — Map the contradictions
Once the panel reports back, don't average it into mush. The disagreements are the signal. Turn the perspectives on each other and make three things explicit: where they clash, which conclusions stand on solid ground versus thin, and the assumption every one of them made without saying it. That shared silent assumption is the useful part — it points straight at the perspective still missing from your panel.
Paste this:
Now compare the five perspectives against each other.
Where do they disagree? For each conclusion, mark whether it rests on
strong evidence or weak evidence. And name the one assumption all five
share without stating it — the blind spot none of them questioned.
That shared assumption points to the perspective still missing here.Step 3 — Verify at the source
A first pass will carry claims that simply aren't right. So before you trust any of it, send every claim back to where it came from and give it a verdict: confirmed, corrected, or demoted. Then ask for a plain-language read on how much weight to put on the whole: which findings several independent lenses backed, and which hang on a single weak source. Keep that read qualitative — you want to know what to lean on, not a number to argue with.
Paste this:
Take every factual claim and trace it back to its source. Give each one
a verdict: confirmed, corrected, or demoted. Then, without scoring, tell
me which findings several independent perspectives backed, and which
rest on a single weak source.Step 4 — Tailor it to you
A generic report hands you a brain dump of everything it found. That isn't the point. Tell the panel up front who you are, what you do, and what decision this research feeds — and it can end with takeaways that are actually yours to act on. The framing at the start is what turns raw data into a call you can make.
Paste this before Step 1, or drop it in at the end:
Before you start: here is what I do and what I want out of this research
— [your work, the decision you're facing]. End with practical takeaways
for me, not a raw data dump.Decision — sub-agents or an agent team
Both let several perspectives work at once. The difference is whether they can talk to each other, and that decides which one you want.
- Sub-agents. They each work for one main session and report back to it; they don't talk to each other. Reach for this when you want separate views gathered and compared from one place — cheaper and faster, and enough for most research.
- An agent team. The perspectives can argue directly with each other and work toward a consensus. Reach for this when the debate itself is the point — you want the experts to push back on one another, not just file separate reports. More powerful, and more expensive to run.
Default to sub-agents. Move up to a team only when you genuinely need the perspectives to fight it out.
The principle that stays
Skills and tools will change; this doesn't. More independent perspectives that are willing to disagree give you more complete research than any one prompt — the idea has roots in a research approach called STORM, and it holds well beyond that origin. So when a topic sits outside your expertise, don't fake the knowledge: borrow it. Convene the roles that kill your blind spots, let them contradict each other, and keep only what survives the source.