19/04/2026
Method selection is where most product research fails. Teams rarely struggle with executing a chosen method — they struggle with picking the right method for the question in front of them.The answer depends on five or six factors, and none of them are obvious to anyone starting out. So I built a reference.
I just published a Periodic Table of Research Methods at thehuman2ai.com/research/methods. It maps 40 canonical methods — drawn from MeasuringU, NN/g, the UserInterviews Field Guide, and the four JTBD schools — across five groups (Understand, Generate, Validate, Measure, Analyze). You filter by business task, product stage, AI compatibility, and expertise level required.
Nine pre-built research-design chains cover the most common product decisions:
• Launching a new product
• Entering a new market
• Redesigning navigation / IA
• Redesigning an interface
• Measuring satisfaction
• Retention diagnostics
• Conversion-drop investigation
• Feature discovery
• Pricing and packaging
Each method card gives you a description, the question the method answers, participants and timing, AI compatibility, full workflow, common beginner mistakes, and a ready-to-adapt AI prompt.
The build itself was also an experiment. I wanted to test how close a well-instrumented LLM workflow could get to methodology expert-review. Before any method was written, I built and verified a source library — primary texts from the methodologists themselves (Christensen, Moesta, Ulwick, Kalbach on JTBD; Sauro and Lewis on measurement; NN/g's method corpus; and around 40 more). Dubious sources got dropped. Every method description was written against only the verified library — no drift into internet-common-wisdom.
Claude Code and Codex handled the drafting and methodological review. It was slower than doing the work myself, but I wanted to see the ceiling of LLM-assisted methodology work. The ceiling is higher than I expected.
Two notes for anyone using the prompts:
First — none of them work flawlessly on copy-paste. Run them through your own LLM grounded in your product's context. Five extra minutes, noticeably better output.
Second — there's no hidden AI layer on the site. This is intentional. Every research project needs to be grounded in your business's sources of truth, loaded into your own workspace, inside your own setup. Running a research pipeline through someone else's model layer — unknown reasoning levels, unseen system prompts — produces generic output that rarely applies to
your specific product.
Beta stage for now — feedback welcome.
→ https://thehuman2ai.com/research/methods
Pick a research method by business task, product phase, or AI compatibility. 40 methods with ready-to-use AI prompts.
