By Dan Donovan, @askdandonovan
A Stanford-inspired prompt that makes Claude investigate your decision from five angles, show you exactly where those angles disagree, and hand back a sourced decision brief before you commit money or time.
STORM is a research system built by Stanford's OVAL lab. The name stands for Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking, and it was published at NAACL 2024. The core idea: instead of asking one generic set of questions about a topic, it discovers several distinct perspectives and lets each one interrogate the topic separately, then synthesises the results with citations.
Stanford's own evaluation reported their articles scored better on organisation by 25 percentage points and on breadth of coverage by 10 points against a baseline retrieval system. You can read the project page, the paper, the open source code, or try their free demo at storm.genie.stanford.edu.
What's on this page is not the Stanford system. It's the method borrowed and rebuilt as a single prompt you can paste into a normal Claude chat, pointed at a business decision instead of a Wikipedia article.
Before you touch the full prompt, prove the idea works on something you're actually deciding this month. Open claude.ai, paste this, and fill in the two brackets.
I'm deciding: [YOUR DECISION IN ONE SENTENCE]
Context: [BUDGET, TIMEFRAME, TEAM SIZE, WHAT YOU'VE ALREADY TRIED]
Answer this three times, separately and independently:
1. As a practitioner who has actually implemented this. What breaks in week three?
2. As a skeptic. Give me the strongest argument that I'm about to make a mistake.
3. As an economist. Follow the money, the incentives and the second-order effects.
Then show me one table: where do these three directly contradict each other,
what exact claim is in dispute, and which side currently has better evidence?
Do not blend them into one safe answer. If you can't verify something, say so.If the table at the end tells you something you hadn't thought of, the full five-perspective version below is worth the extra few minutes.
Four steps. The prompt does the heavy lifting, but the two minutes you spend on steps 1 and 2 decide whether the output is useful or generic.
This is where most people lose. A topic gets you an essay. A decision with real constraints in it gets you a recommendation. The difference is money, timeframe and the alternative you'd spend it on instead.
Your question is ready when it contains all four of these:
Five experts arguing about a business they know nothing about is theatre. Give them the operating reality and the argument gets specific. Fill this in and keep it, you'll reuse it on every future run.
Business: [WHAT YOU SELL, TO WHOM, HOW LONG YOU'VE BEEN DOING IT]
Size: [HEADCOUNT, ROUGH REVENUE BAND]
Money on the table: [THE AMOUNT THIS DECISION COMMITS]
Timeframe: [WHEN THE DECISION HAS TO BE MADE, AND OVER WHAT PERIOD IT PLAYS OUT]
Already tried: [WHAT YOU'VE ATTEMPTED AND WHAT HAPPENED]
Constraints: [ANYTHING NON-NEGOTIABLE: CASH, COMPLIANCE, A CONTRACT, A PERSON]
What good looks like: [THE OUTCOME THAT WOULD MAKE THIS A GOOD DECISION]
What I'm afraid of: [THE FAILURE MODE THAT KEEPS YOU UP]This is the whole thing. Drop your question into the first bracket, your context block from step 2 into the second, and send it. Give it a few minutes if web search is running.
Act as a multi-perspective research panel helping me investigate this question:
[INSERT YOUR QUESTION]
My context:
[ADD RELEVANT BACKGROUND, CONSTRAINTS, FILES AND DESIRED DECISION]
Use any source documents I provide. If reliable web research is available, use it and
include working links. Never invent a source, quotation, statistic or citation. Clearly
label anything you cannot verify.
PERSPECTIVE ANALYSIS
Analyze the question independently through these five perspectives:
1. Practitioner
Focus on what works in real situations, operational constraints and implementation.
2. Skeptic
Challenge the assumptions, identify failure modes and argue the strongest opposing case.
3. Economist
Examine costs, incentives, trade-offs, second-order effects and who benefits.
4. Historian
Identify genuinely comparable precedents, what happened and where the comparison breaks down.
5. Academic
Assess the quality of the available evidence, competing explanations and remaining uncertainty.
For each perspective, provide:
- Its central position
- Its strongest supporting evidence
- Its strongest objection
- Assumptions it is making
- Source links for factual claims
- A confidence rating from 1 to 10 with a brief explanation
CONTRADICTION MAP
Create a table showing:
- Where two or more perspectives directly disagree
- The exact claim being disputed
- Evidence supporting each side
- Which side currently has stronger evidence and why
- What additional evidence would resolve the disagreement
Then identify:
- What all five perspectives agree on
- What none of them adequately addresses
- The most important hidden connection
- The weakest claim in the complete analysis
DECISION BRIEF
Synthesize the research into:
- A one-paragraph executive summary
- The most defensible conclusion
- The three strongest supporting findings
- The three most important risks or uncertainties
- The strongest credible alternative conclusion
- One immediate, low-risk action
- One larger action to consider after gathering more evidence
- A list of unresolved questions
VERIFICATION PASS
Before finishing:
- Check every factual claim against its cited source
- Remove or label claims that cannot be verified
- Check whether one perspective dominated unfairly
- Separate sourced facts from interpretation
- Confirm that every link supports the specific claim attached to it
- Revise the decision brief if the verification changes the conclusionWhat each lens is actually doing for you:
The contradiction map is the point. Five opinions blended together is just a longer average answer. The table is what stops that happening:
| Disputed claim | Side A | Side B | Stronger right now |
|---|---|---|---|
| Automation can replace the sales hire | Economist: no ramp time, no fixed salary, cost is known up front | Practitioner: the pipeline runs on relationships a workflow can't hold | Practitioner, until inbound converts without a human touching it |
| The 12-month payback is realistic | Historian: comparable rollouts took closer to two years | Academic: the comparison set is too small to conclude much | Neither. This is the claim to go and test first |
Illustrative shape only. Your table will be filled with your own decision.
A confident brief is not a correct brief. Three follow-ups, sent one at a time in the same chat, do most of the work of catching a bad one. This takes two minutes and it's the step people skip.
Which single claim in this brief, if it turned out to be false, would most change
your recommendation? Show me the source you used for it, quote the exact line that
supports it, and tell me how confident you are that the source says what you claim
it says.Now argue the alternative conclusion as if you believed it. What would have to be
true about my business for the alternative to be the right call? Then tell me which
of those things I can check this week.List every link you gave me. For each one: what specific claim is it attached to,
and does the page actually support that claim or did you infer it? Mark anything
you inferred. If you can't open a link, say so instead of describing it.Then do the two-minute human check before you act on any of it:
Four decisions where being wrong is expensive and slow to reverse. Paste any of these into the question slot, add your context block, and send.
Should I hire a second [ROLE] at [SALARY], or spend the same money automating
[THE PROCESS THAT ROLE WOULD OWN], over the next 12 months?Should I raise prices by [X]% across the board in [MONTH], grandfather existing
clients for 6 months, or hold prices and cut [COST] instead? I care most about
[MARGIN / RETENTION / CASH THIS QUARTER].Should I shut down [SERVICE LINE], which is [X]% of revenue and [Y]% of my team's
hours, and move that capacity to [THE OTHER THING], in the next two quarters?Should I pay [$X/month] for [SOFTWARE], build the same thing internally, or keep
running it manually? Migration would take [TIME] and touch [WHO / WHAT SYSTEMS].Worth being straight about, because the version of this floating around online usually isn't.