👋 Hey, I’m Nikki. Each week I write about UX research strategy, communicating impact, and using AI to do your best work. For more: Claude Skills Bundle | Claude Agents | AI Prompt Library | Team Training | AI Courses for UXRs
P.S. Paid subscribers get access to full archive, all content, a private Slack community, Substack lives, and a hub of templates, scripts, and mini-courses
If you’ve been staring at Claude or ChatGPT open in one tab and your actual research work in another, feeling vaguely guilty that you’re either using it wrong or not using it enough, you’re not alone. I hear some version of this from researchers every single week.
There are two ways I see people get this wrong:
Handing AI everything. You paste in eight transcripts, ask for “the key insights,” and get back a tidy list of themes that sound right and mean nothing. So you use them, since you’re busy and they look finished, and three weeks later a stakeholder makes a decision based on a “finding” that was really just the model flattening your nuance into a bullet point. I’ve watched this happen, and it makes me a little sick every time, since the research looked done and wasn’t.
Refusing to touch it. You keep formatting your own discussion guides, tagging every transcript by hand, and rewriting the same screener for the fourth time this quarter, and telling yourself the whole time that real researchers don’t cut corners. Meanwhile the strategic thinking, the part only you can do, keeps getting squeezed into the last twenty minutes of your day when your brain is fried.
Both of these come from the same mistake, which is treating “my research” as one big blob that you either trust AI with or you don’t. It was never one blob. Every project you run is a pile of very different tasks, and some of them are genuinely fine to hand off, and others should never leave your hands. The skill that really matters right now, the one I think separates the researchers who are going to thrive from the ones who are going to burn out or get burned, is knowing which is which.
So I built an exercise to figure it out, task by task. By the end you’ll have sorted your own work, split your messy in-between tasks down the middle, and know exactly what to turn into a reusable AI skill so you never explain it twice.
Give yourself about an hour and have a real project in mind before you start to get the most out of this.
The distinction
AI is very good at mechanical work and very bad at judgement, so the job is to sort your work along that line and hand off only the mechanical side.
Mechanical work is stable, rules-based, and produces a repeatable output. If I gave the same task to five competent people with the same instructions, I’d get five nearly identical results. There are three signs I look for:
It repeats. I do it over and over, study after study.
It follows the same rules each time, so the steps don’t change based on who the participant was or what mood the room was in.
The output is something I could write down as instructions, a format, a checklist, or a template.
Formatting a discussion guide into your team’s standard layout is mechanical. Tagging transcripts against a taxonomy you’ve already defined is mechanical. Turning a set of criteria into a screener is mechanical.
Judgement work is context-dependent and needs your read of the situation. It calls on empathy, experience, and the ability to notice what a rule can’t capture. The three signs are:
It needs context that lives in your head or in the room, not in a procedure.
It needs your interpretation, your point of view, your sense of what matters here.
It changes every time, since the right call depends on things that are only true this once.
Deciding what a study’s main question should even be is judgement. Deciding what an insight really means is judgement. Reading a participant’s hesitation and choosing to follow up in a direction that wasn’t in your guide is judgement.
Now the part that trips everyone up, and the reason “just let AI do my analysis” goes so wrong is that most of your tasks are not purely one or the other. They’re a blend. “Write insights” feels like a single task, but it’s really a mechanical shell (structure the raw notes, pull the supporting quotes, organize by theme, drop it into a consistent format) wrapped around a judgement core (decide what the pattern means, decide which finding really matters, decide what you’d recommend). When you hand the whole blob to AI, you accidentally give away the judgement core along with the mechanical shell, and that’s the exact move that produces those confident, hollow themes I mentioned at the top.
The strategy, then, is to split. Delegate the mechanical portion, keep the judgement portion, and be deliberate about where the seam is.
Here’s a quick reference you can screenshot and keep next to you for the rest of this exercise:
I’ve found this reframe does something calming for people, since it stops the question from being “do I trust AI or not,” which is unanswerable, and turns it into “which part of this specific task is mechanical,” which you can really answer.
Below, I walk you through the full exercise I use to audit an entire research workflow and come out the other side knowing exactly what to hand off, what to keep, and what to turn into a reusable AI skill.
The four-question test with a fill-in scorecard, plus six worked examples so you can see the edge cases that trip people up
The full brain-dump map across all six phases of a research project, with a filled-in example and a blank template you can copy
The three-bucket sort as a reference table of real research tasks already placed, so you’re pattern-matching, not guessing
The “splits into” move with five fully worked splits (insights, discussion guide, readout, screener, journey map)
The watch-outs I use so I never over-delegate, including the voice, privacy, and one-off traps, written as a pre-flight checklist
The four-step way to turn a mechanical task into a Claude skill, with copy-paste description templates and a starter library of the ten skills I’d build first
A full worked walkthrough where I run one real study through all six steps end to end
If you’ve been feeling that low-grade guilt about AI, either that you’re behind or that you’re cheating, this will give you a clean, defensible line to stand on, and the actual templates to act on it today.
Exclusively for paid subscribers
