Couch to AI Researcher

Most AI tutorials are written for generic knowledge workers. This one is built for UX researchers, doing UX research work. Every exercise uses real research artifacts, discussion guides, screeners, transcripts, journey maps, stakeholder updates.

By Day 30, you've got a setup that knows what research is, what your role is, and what your team needs from you. No translating from generic prompts to your actual work.

I'm Nikki Anderson, a UX researcher who's been writing The User Research Strategist on Substack and training researchers for years. I built this since I've seen the gap between "AI is going to change UX research" think pieces and "actually, here's what to do on Monday morning." This program is the Monday morning answer.

Yes, with a small caveat. The exercises assume you're doing some kind of research work, scoping studies, drafting guides, synthesizing findings, talking to stakeholders. If you're earlier in your career and not running studies yet, you can follow along with hypothetical examples on every day, and the leverage will hit harder once you're doing the work for real.

Yes. Claude Pro at $20 a month is enough. The free tier has daily message limits that will stop you partway through these exercises. Cowork (Week 2 onward) needs a paid plan too.

For Week 1 (basic prompting techniques), any frontier model works. Week 2 onward uses Claude Cowork, which only runs on Claude. If you'd rather stay loyal to another tool, you'll get the most value from Week 1's technique-building exercises and you'll need to translate the Cowork weeks to your tool's equivalent.

Most days run 5 to 15 minutes. The Week 2 setup days run a bit longer (20 to 30 minutes for installing Cowork and connecting tools). The Week 4 race-week deliverable days vary based on what you're shipping. The point is daily, applied, on real work, not a 2-hour weekly grind.

Pick up where you left off. The program is sequenced so each day builds on the last, and missing a day or three doesn't break anything. The "30 days" framing is for accountability, not punishment. Your race day is the date that matters.

claude.ai is Claude in a browser. Every conversation is its own thing, no memory of yesterday. Cowork is Claude with persistent memory, tool integrations, and skills (configured mini-assistants). Where claude.ai is great for one-off tasks, Cowork is the version that becomes your actual research assistant.

A project is a workspace inside Cowork that holds one piece of work, the brief, the participants, the protocol, the past findings. Each study gets its own so the context doesn't bleed between projects. When you sit down to work on Study A, Cowork pulls from Study A's project, not from every study you've ever set up.

A saved prompt is text. A skill is a configured agent. The skill has its own instructions, can read from your project documents automatically, can save outputs back to your tools, and improves over time as you give it feedback. Where a prompt sits in your clipboard, a skill sits in your toolkit and shows up when you call its name.

A researcher profile is what you teach Cowork on Day 9: who you are, what you research, how you write, what you'd never do. Once it's saved, every output you ask Cowork for gets shaped by it. Without a profile, you're talking to default Claude every time, and default Claude writes like default Claude.

Anthropic doesn't train on your conversations or data when you're on a paid plan. Your Cowork project documents and integrations stay scoped to your account. If your company has data policies, the integrations let you scope what Cowork can access (specific folders, channels, workspaces).

Check with your security or IT team before connecting work tools. You can run all of Week 1 (claude.ai with hypothetical examples) without connecting anything sensitive. Many companies have either an enterprise Claude plan or a vetted set of approved AI tools. If yours doesn't, the program still works on personal projects, side studies, or hypothetical scenarios as you build the case internally.

Use Opus as you're learning the techniques. Opus gives you the cleanest signal of whether a prompt move actually worked. Once you're confident, you can drop to Sonnet for routine work and save Opus for the hard stuff (synthesis, strategic POVs, anything where nuance matters).

The skills you built are durable. The projects keep working. Race day is one milestone, not the end. Most people leave the program with one or two skills they use weekly, a Cowork setup that's saving them hours, and a habit of reaching for AI on research work they used to do alone. Pick the next deliverable to transform, the next skill to build, and keep going.