Stanford built a research method called STORM that beats a single prompt by making several experts research the same topic and argue about it. Here is the founder-friendly version you can run in Claude today, plus the exact skill that does the whole thing for you in one command.
When you ask Claude one question, you get one angle back. That angle carries every blind spot baked into the way you framed the question. For a quick fact, that is fine. For a decision that shapes your offer, your pricing, or where you point your next quarter, one angle is thin, and you rarely notice what it left out.
Stanford's OVAL lab built a research method called STORM to fix exactly this. STORM stands for Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking. Instead of researching a topic from one point of view, it discovers several, has each one research the same topic and question the others, then curates the result. In Stanford's own evaluation, articles built this way came out about 25% better structured and covered more ground than the standard single-pass approach.
This guide gives you the version a busy founder can actually use. You will convene a council of six AI experts on any topic, make them disagree on purpose, then force Claude to fact-check its own sources before you trust a word of it. At the end, you can install the Her AI Systems™ version of this as a skill, so the entire process runs from one sentence every time.
Before you research anything, get specific about two things: the exact question you are trying to answer, and who the answer is for. A briefing written for a first-time founder looks nothing like one written for a seasoned operator, and naming the reader up front is what makes the final answer usable instead of generic.
Write one sentence for each. For example: "Topic: are AI voice agents worth building into my service business this year. Reader: me, a solo founder deciding whether this is worth my time and money."
This is the heart of STORM. Rather than one researcher, you spin up six, each with a different job. Every lens is built to catch what the others miss:
Paste this into Claude, swapping in your topic and reader from Step 01:
I want you to research this topic from six expert perspectives, one at a time. Topic: [your topic]. Reader: [who this is for].
Do real web research for each and cite sources with links. For each of these six lenses, give me: a two-sentence position, three to five pieces of evidence with a source link, and the one thing only this expert would point out.
1. The Practitioner (works with this daily). 2. The Researcher (peer-reviewed evidence). 3. The Skeptic (strongest honest case against). 4. The Economist (follows the money). 5. The Historian (past parallels). 6. The Newcomer (what beginners actually need explained).
Keep each lens under 250 words. Do not invent sources.
When it finishes, drop the whole thing into your note. You now have six angles instead of one.
Six opinions are only useful once you see where they collide. This step turns raw research into signal. Where two lenses claim opposite things, you have found the question worth digging into. Where all six agree, even the Skeptic, you have found the thing that is most likely true.
Now compare the six perspectives you just gave me. Tell me: where do two or more of them directly contradict each other, and name the specific claims that clash. Which lens has the strongest evidence and which has the weakest, and why. What does every lens agree on, even the Skeptic. And what did none of them address, which is our biggest blind spot.
This is the step that separates a real briefing from a confident-sounding pile of claims. AI will state a number with total confidence and be wrong. Before you act on anything, make Claude go back to the original sources and check its own work.
Go back through every statistic and citation you used across all six perspectives. For each one, find the actual primary source and confirm or correct it: the real figure, who published it, and the year. Mark each claim as CONFIRMED, CORRECTED, or COULD NOT VERIFY. Remove or clearly flag anything you cannot stand behind. Do not defend a number you cannot trace to a real source.
Watch what happens here. Some claims get confirmed, some get quietly corrected, and a few get pulled entirely. That is the system working. Now you can trust what is left.
Research you cannot act on is just reading. The last prompt collapses everything into a call, written for the reader you named in Step 01, in plain language you can use.
Using only the verified findings, write me a short briefing for [my reader from Step 01]. Include: a five-line summary, the three findings I can most rely on, the one big assumption this all rests on, three specific moves I should make now, and a one-paragraph version simple enough for a total beginner to understand. End with the single question that would change the answer if we knew it.
Running four prompts by hand is worth doing once so you understand the moving parts. After that, you will want it to run itself. That is what a skill is: a plain text file you hand to Claude once, so a whole system runs from a single sentence.
Her STORM is our packaged version of everything above. It spins up all six lenses at once, maps the contradictions, builds a clean HTML briefing, verifies every citation, and tailors the takeaways to your business. To use it, you drop the skill file into Claude's skills folder, then just say:
Run Her STORM on [your topic].
No code, no re-pasting prompts. One sentence in, one verified briefing out. You can grab the Her STORM skill from the Her AI Systems free resources, and once it lives in your Claude, this becomes a five-second ask instead of a twenty-minute process.
You just turned a single AI answer into a council of six, made them argue, checked their sources, and walked away with a decision instead of a guess. Do this on the questions that actually move your business and you will stop trusting the first thing AI tells you, which is exactly the habit that separates an AI operator from an AI user.
Once you are comfortable running research this way, the next step is wiring it into how you build content, offers, and decisions on purpose. That is exactly what we build inside Her AI Systems™.
STORM stands for Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking. It was built by Stanford's OVAL lab and presented at NAACL 2024. Instead of researching a topic from one angle, it discovers several perspectives, has each one research the topic and question the others, then curates the result. In Stanford's evaluation, articles built this way were about 25% better structured and covered more ground than standard single-pass methods.
When you send one prompt, you get one angle and you inherit every blind spot in how you framed the question. Six perspectives, the practitioner, researcher, skeptic, economist, historian, and newcomer, each find a hole the others miss. Where they agree, you have a strong hypothesis. Where they disagree, you have found the exact question worth digging into.
No. You can run the whole thing by pasting four prompts into Claude in order. If you want it to run in one command every time, you install the Her STORM skill, which is a plain text file you drop into Claude's skills folder. No code required.
Deep research spins up many agents and hands you a dense pile of stats. STORM uses a fixed council of perspectives, maps where they disagree, verifies its own citations, and ends with a decision tailored to you. It is more structured, cheaper to run, and easier to act on.
Yes, and you should. If your business has a lens the six do not cover, add it. The Her AI Systems version already adds a Newcomer lens so every briefing gets tested for whether a smart beginner could actually understand and use it.
Free Download
Your future self will thank you for saving this.
Follow along for the next guide
New free guide every Tuesday. Find them first on Instagram, TikTok, or by subscribing to the email list.
Want more?
One new free guide. Every Tuesday. No spam, no fluff, just the systems and prompts I'm building for clients, simplified for you.
Subscribe