Nielsen Norman Group: testing complex interactions earlier with AI prototyping
Published on Nielsen Norman Group in September 2026, Megan Chan’s article addresses a practical gap in how designers approach user testing for complex interfaces. AI-powered prototyping tools have changed what is feasible at the early stages of a project, and the article describes how to use them to move user testing earlier in the design process rather than waiting for high-fidelity builds.
What the article is about
The central claim is that AI tools now make it practical to build fully interactive prototypes of interfaces involving complex interactions — multi-filter dashboards, conversational AI, data-heavy screens — in a fraction of the time traditional methods require. This shifts the window for user testing: instead of testing with real-coded prototypes late in development, teams can test with AI-generated interactive prototypes early enough that findings still shape the architecture.
Context: the problem with complex interaction prototyping
Static wireframes and click-through prototypes reveal some usability issues but systematically miss others. Edge cases, transition logic, and the interaction between components only become apparent when something is genuinely interactive. Before AI prototyping tools, creating that level of interactivity early enough to be useful required developer involvement — time most teams do not have at the concept stage.
Key method
Chan describes a five-step process: defining what the prototype needs to do, making explicit design decisions before prompting rather than letting the AI make them by default, providing relevant data for the prototype to display, writing detailed prompts that reflect those decisions, and iterating on the output. Each step addresses a failure mode the article identifies — underdefined requirements, implicit design choices, output that cannot be mistaken for the real product.
The article is direct about what AI prototyping cannot do: strategic design choices remain the designer’s. An AI tool realizes decisions faster; it does not make them. Chan names this explicitly as the reason why the method starts with design decisions rather than prompts.
She also names a specific risk: a high-fidelity AI prototype can look finished. Teams and stakeholders may treat it as production-ready and skip necessary refinement cycles. She recommends framing AI prototypes explicitly as research instruments to counter this tendency.
What changes with earlier testing
The practical value of testing earlier is not just speed — it is the quality of decisions that can still be made. Findings from testing at concept stage can change how a feature is scoped, how a filter system is organized, or whether a conversational interface is the right approach. The same findings surfaced after engineering has started change far less. The article’s argument is that AI prototyping tools have moved the threshold for what early testing can plausibly include.
Who it is useful for
UX designers and product designers who need to test complex interaction patterns early but cannot build interactive prototypes from scratch at concept stage. The article is especially relevant for teams working on filters, conversational interfaces, or data visualizations where static prototypes fall short. The method applies across the current generation of AI prototyping tools; no specific platforms are named.