Usability Testing Still Beats AI Guesswork, NN/g Says

Nielsen Norman Group's ongoing research keeps reinforcing that structured usability testing beats AI-generated assumptions about users. AI can speed up synthesis, but it can't replace watching real people struggle with your interface.

Nielsen Norman Group's research output continues to back a simple message: talking to and observing real users, even in small unmoderated sessions, outperforms guessing at user needs — including guesses an AI model makes about 'typical' user behaviour. AI tools can help write discussion guides or cluster open-ended feedback faster, but the underlying method (recruit, task, observe, iterate) hasn't changed. For a junior designer building a portfolio, this is good news: you don't need fancy AI research tooling to run a credible study — you need five users and a clear task. Label: **established** — decades of usability research back this; AI hasn't displaced it.

Related in UCD methods

Established

Still true: 5 users find most of your usability problems

NN/g's classic finding — that testing with five users uncovers most usability issues — still holds and still gets ignored by teams who think they need a 'proper' sample size.

nngroup.com19 Mar

Established

Heuristic evaluation is still your fastest fix

With AI tooling dominating design chatter, it's easy to forget the cheapest, fastest usability method still works: heuristic evaluation. It costs nothing but time and catches problems before you ever test with real users.

uxdesign.cc1 Jan

Established

Heuristic Evaluation Still Earns Its Keep in 2025

Nielsen's usability heuristics remain a go-to lightweight method for catching obvious UX problems before user testing, decades after they were written. It's boring, reliable, and still worth learning properly.

interaction-design.org1 Aug