This newsletter is brought to you by the AI UXR Bundle, my full set of Claude skills, agents and prompts for you. Efficiency + rigor + AI = happy user researchers! Take a look here.
in september, i was running a live workshop on Claude skills, with 30 researchers watching.
i opened a fresh chat incognito with my personal email and asked Claude to write a research insight. It came back immediately with this beauty:
wabam, insight acquired! it’s pretty, huh? and in this little card format that’s kinda intriguing. it’s the kind of thing you’d drop into a readout for your PM.
there was just one problem. none of it was based on data.
Claude had made the whole thing up:
i know that you, as a researcher, wouldn’t believe it or use AI in this way, but stakeholders are. they’re like, wow how great! look at this “insight.” research is so easy!
without guardrails or constant 👀 👀 👀 (that we don’t have time for), how can we ensure the quality of our research?
i ran the exact same request through my insight writer skill. This time, before Claude wrote a single word, it asked me four questions:
how many participants said this?
what’s causing it?
what happens if the team doesn’t act?
what format do you need?
when i gave it thin data on purpose, it highlighted the missing info instead of making things up.
here’s why i’m telling you this.
if you’ve tried AI on your research and thought “why didn’t i just do this myself?”, i thought that for a long time too. My early outputs were absolute garbage. What i was missing was the context.
you can build these files yourself, for free. Here’s how i built mine:
write out how you do the task, step by step, the way you’d explain it to a new researcher on your team or the way you’d democratize it to a stakeholder
gather 5 examples you’re proud of and 3 weak ones, with a note on why each one works or doesn’t
grab a template or two of the ideal output
give it all to Claude and have it ask you questions before it drafts anything
test it by feeding it a bad input, then rerun a finished project through it and compare the output to what you did by hand
this works, and it’s how every skill i use started.
It’s slow, though. my skills (i have about 140 now) took me the past year to build. some skills, on average, take days to build, some weeks, depending on how much content i have on hand.
or you can take the shortcut and not stay up all night wondering what garbage your stakeholders might send you the next time they “analyze research”
I’ve already done this for every stage of my research and put it all into the UXR AI Bundle. it works with all LLMs (yes, that includes: GPT, Gemini, Claude, and even Copilot).
That’s 76 skills, 9 Cowork agents that string them together for each research stage, and 109 prompts built on 13 years of my real work (and a whole lotta my time).
it’s $249, one payment, with lifetime access ($315 if you bought the three separately) to updates. I already have two new bonuses in progress: dedicated stakeholder-facing skills and an LLM grader.
if…
your company has a learning budget, purchase the skills, and then send me an email for a custom invoice
you’ve already purchased one of these on their own and want access to the bundle, hit me up and i’ll give you a credit
you’d rather your LLM asked how many participants said it before it wrote anything, grab the bundle!
have the best day!
nikki
ps. email me with any questions and i’m happy to answer