Research Like a PhD
- Categories TUTORIALS
- Date July 20, 2026
Research Like a PhD:
The STORM Method for M&E
A step-by-step method — adapted from Stanford’s STORM research system — for turning a one-shot AI answer into a grounded, multi-perspective evidence brief you can defend in front of a donor, a technical committee, or your own evaluation team.
5 guided steps
Free · No login required
Why this matters for evaluation work: a single AI prompt on a contested topic — “should we adopt AI-assisted coding for our qualitative data,” “is cash transfer more cost-effective than in-kind aid here” — tends to return an averaged, defensible-sounding paragraph that argues nothing. STORM’s method fixes this by researching the question from several named perspectives first, grounding every claim in a source, and keeping disagreements visible instead of smoothing them away. This app walks you through it and builds your working files.
Is the STORM method the right tool for this question?
Answer three questions about the research task in front of you. This isn’t a formal instrument — it’s a quick gut-check before you spend 20 minutes on a method built for contested, multi-angle questions.
Build your research panel
Fill in your question, choose who belongs on the panel, then generate a ready-to-paste working file. Editing the panel before any research runs is the single highest-value step in the whole method.
The three-prompt method
This is the method behind Stanford’s STORM system (NAACL 2024), which generates long, cited articles by discovering perspectives first, then running grounded interviews from each one before writing anything. Run these three prompts in order in Claude, editing between each step.
Discover perspectives — then edit them yourself
Ask for the panel before any research runs. Read the list it returns and cut, merge, or add a perspective based on what you actually know about your program and stakeholders.
Interview each perspective, grounded in sources
Run this once per perspective. Require web search, a follow-up on any thin answer, and a named source for every claim — including where sources disagree.
Synthesize into a brief a stakeholder can act on
Pull every interview into one outline for a reader who knows nothing about the topic yet. Ask it to flag any claim resting on a single source — that’s exactly what the checklist in Tab 4 checks.
Know the limits before you rely on it
- Output can still read flat and encyclopedic — it needs your editorial voice on the way out, not just on the way in.
- The method doesn’t critique itself. It won’t tell you which of its own claims is the weakest — that’s your job in the checklist tab.
- It can interview six perspectives thoroughly on the wrong question. Get the research question right first.
Verify before you rely on it
Run this checklist against the synthesized brief before it goes into a report, a decision memo, or a donor conversation.
Frequently asked questions
What is STORM, and do I need the original Stanford system to use this method?
STORM is a research system published by a Stanford team in 2024 (NAACL) that writes long, cited articles by discovering perspectives, running grounded interviews, then outlining and writing. You don’t need the original system — this app teaches the method so you can run it directly in Claude with the three prompts in the Method Guide tab.
How is this different from just asking Claude one detailed question?
A single prompt tends to collapse a contested topic into one averaged, smooth-sounding answer. This method asks the same question from several named perspectives, grounds each answer in a source, and keeps disagreements visible — which is usually where the real finding is.
Do I have to use all the perspectives suggested in the Panel Builder?
No — editing the panel before research runs is the most valuable step in the whole process. Cut anything that doesn’t fit your program context and add a perspective the default list missed, such as a local partner or a specific donor.
Can I use this for a formal evaluation report, not just an internal brief?
Yes, with the verification checklist completed first. Treat the generated brief as a strong first draft grounded in named sources, not a finished deliverable — every claim still needs the same source-checking you’d apply to any evidence going into a report.
What are the known weaknesses of this method?
The output can read flat and encyclopedic, it does not critique its own weakest claims, and it will happily research the wrong question thoroughly if your framing is off. The Method Guide and Verification Checklist tabs are built specifically to catch these.
Is any of my program or beneficiary data sent anywhere by this app?
No. This app runs entirely in your browser to build your quiz result, panel selection, and downloadable files. Nothing is transmitted or stored on EvalCommunity’s servers. Keep beneficiary-identifying details out of any prompt you run in Claude directly, per your organization’s data policy.
Where can I read the full tutorial this app supports?
See the full step-by-step article on EvalCommunity Academy: Research Like a PhD — The STORM Method for M&E.
The courses and articles have been developed by an experienced team of evaluators and software developers under the guidance of Fation Luli. The EvalCommunity Academy combines practical expertise in Monitoring & Evaluation with cutting-edge AI technologies to provide high-quality, accessible learning experiences for professionals around the world.