NN/g study guide: how to use artificial intelligence in UX work

The article, published by Nielsen Norman Group (NN/g) and written by Tanner Kohler, is not a traditional essay but a "study guide": a curated collection of links to NN/g's own articles, videos and podcast episodes about the use of artificial intelligence in user experience (UX) work. That is why this summary is of moderate length and not artificially inflated: it faithfully reproduces the structure of a thematic index of resources, with brief descriptions of each, rather than developing a single extended argument.

The starting premise is clear and framed as a question and answer: should a UX person use AI in their work? Yes, according to NN/g, because the relevant question is no longer whether AI can help, but when and how to do so. But they immediately qualify: does AI guarantee better or more efficient work? No. The material insists that you have to think critically and use AI strategically, not hand the work over to it unsupervised. The phrase that sums up the approach is telling: if you don't want to work for the robot, make the robot work for you.

The guide is organized into several thematic blocks. The first, "Integrating AI into UX workflows," starts from the idea that handing the entire workflow over to AI is not, for now, the best strategy; instead, the field's most advanced practitioners have found specific ways to let AI assist their existing work, acting as strategists who put on AI "gloves" to expand what they can achieve, while the hands —that is, human judgment— continue to steer the outcome. This block cites resources on how to start with small tasks while watching for hallucinations, a prompting structure called CARE (context, specific ask, rules and examples), the use of generative AI as an assistant for content, research, ideation and design, the "7 deadly sins" that erode UX skills if you give in to the temptation of relying too heavily on AI, the analogy of working with AI as if it were a talented intern who still needs supervision, the phenomenon of people without technical training building agentic AI systems through self-taught experimentation (so-called "vibe architects"), the idea that the design process has not died but has been compressed —experienced designers still follow the same steps, only faster— and an analysis of why AI-generated Christmas ads fail, as an example that human judgment is still necessary in creative work.

The second block, on career strategy and the evolution of UX roles, openly acknowledges that UX jobs are changing because of AI, something uncomfortable but not unprecedented in the discipline's history. The recommendation is not to mourn how things used to be, but to adapt skills and mindset. The linked resources argue that tools change constantly but AI will not eliminate the need for fundamental skills such as understanding people and knowing how to ask the right questions; that designers retain value by focusing on strategy, narrative, outcomes that matter and judgment about data; that the UX "generalist" is making a comeback, able to move across disciplines with strategic thinking, as AI expands what a single person can cover; that there are five basic principles for redefining design skills in the face of AI; that small experiments and a moderate news diet (both optimistic and skeptical) are enough to keep from falling behind; that "good taste" and judgment will still be needed to produce superior designs even if AI lets more people create content; that the virtual reality hype cycle offers relevant lessons, because the disillusionment it produced could repeat itself with AI; that the single label of "designing with AI" has already split into four distinct types of work; and that UX consulting clients still expect sound judgment, research rigor and respect for real-world constraints, despite the greater use of AI.

In the design section proper, the central thesis is that valuable skills are shifting from "UI" (moving pixels in a design tool) toward pure "UX": defining user needs, specifying intent and overseeing the creation of good journeys, as well as curating the context given to AI. Resources are mentioned on using AI to create prototypes (with promise but also risks), "promptframes" —an evolution of the wireframe that uses AI to increase fidelity and get better feedback from users and stakeholders—, using AI to generate realistic tables and charts in prototype testing, so-called "designing with UX context" (where the output of research and design work stops being documents for humans and becomes curated context that guides AI), critique as a core design skill in the AI era (encoding user needs and design judgment into well-defined evaluation criteria), "outcome-oriented design" (adaptive frameworks that respond to individual goals instead of optimizing a single interface for the average user), and the importance of precise prompts —with visual keywords, references, mock data and code snippets— as opposed to vague prompts when asking AI for interfaces.

The research block argues that AI-generated information still cannot replace real data from real people, with a blunt line: AI doesn't control the customer's wallet, people do. Even so, AI can help in the planning and analysis phases. It cites pieces on how to speed up research with AI (most useful in planning and analysis), how to plan studies with generative AI through careful prompting, a research agenda on generative AI in UX, "digital twins" that simulate human behavior with increasing usefulness, "synthetic users" as a complement —never a substitute— for research with real users, a contextual inquiry methodology applied to AI image generation tools, a case study on using ChatGPT to plan research, the idea that the team learning that comes from observing real users cannot be outsourced even if AI matches the quality of the output, the responsible use of AI as a "thought partner" in analysis (without letting it lead the interpretation), the claim that AI speeds up survey writing but does not replace human expertise in spotting subtle design flaws that undermine data quality, AI-moderated interviews as a faster, more scalable option but not equivalent to human-led semi-structured interviews, the methodological blind spots that become more consequential when AI plans and analyzes research, and the risk of outsourcing qualitative analysis to AI, both for the quality of the findings and for the researcher's credibility.

On writing, the argument is that AI has an exceptional ability to write, but leaning on it too heavily can erode your own style, tone and message; it saves time in the short term but can produce, over the long run, a "monotony in cacophony". The advice is not to use tone words in prompts (it usually fails) but to supply existing texts and ask for multiple alternatives, along with concrete practices to make AI edit better.

In service design, AI can improve the experience of internal employees (which indirectly improves the customer experience) and also interact directly with customers. It anticipates that AI agents will increasingly act on behalf of users and organizations, changing how services are delivered and received, and that new metrics will be needed to evaluate performance across AI systems, human-AI collaboration, data quality and user trust.

On ideation and workshops, the guide argues that AI can act as a "cybernetic teammate": teams that use AI to boost ideation outperform individuals and teams without AI, and even individuals using AI on their own. It offers tips for preparing and facilitating AI-powered workshops.

The "State of AI for UX work" section gathers dated studies portraying AI's impact at different moments, warning that some findings still hold and others have probably changed. Among the concrete data points that do appear in the text: in 2025, narrowly scoped AI design tools were marginally useful but not ready to replace designers; UX activities were among the top types of requests made to Claude, many of them focused on writing tasks; in April 2024, most AI tools for UX design failed to meaningfully support core workflows; programmers using GitHub Copilot increased their output by 126%, with larger gains among the less experienced (2023); customer support agents using AI handled 13.8% more inquiries per hour while slightly improving resolution quality, benefiting less-skilled agents in particular (2023); generative AI raised employee productivity by 66% on average, with the greatest benefit for less-skilled workers and complex tasks (2023); and business professionals using ChatGPT wrote faster and produced higher-quality output (2023). It also notes that in 2020, before large language models were widely available, Jakob Nielsen already predicted many of the ways AI would change UX research and design, and that the state of the discipline in 2026 points to a phase of stabilization after the instability of 2025, where differentiation and demonstrated business impact remain vital.

Finally, the guide links to seven episodes of NN/g's own podcast featuring, among others, Jakob Nielsen, Don Norman and NN/g leaders, plus outside guests from companies such as Condens.io and Great Question, covering topics such as the evolution of the discipline, UX research with AI, the responsible use of AI in analysis, the future of service design and AI's accuracy in UX analysis.

Taken as a whole, the document is not a one-off news item but a reference resource NN/g offers to UX professionals, with a consistent through-line: AI expands capabilities and speeds up specific tasks —planning, prototyping, editing, ideation— but human judgment, research with real users and the fundamental skills of understanding people remain irreplaceable, and those who work in UX should adapt their skills rather than resist change or hand the work over entirely to the machine.