UX-context design: how to make research a context for AI

The Nielsen Norman Group has unveiled a UX-context design approach: user research, interface rules, and design standards should become not only material for humans, but also a constant context for AI tools that create layouts, texts, prototypes, and code.

The problem is familiar to teams working with generative tools: without precise input, the model assembles a “medium” interface. She knows common patterns, but she doesn’t understand the audience of a particular product, the terminology, the limitations of scripts, and the solutions that have been proven. As a result, AI can quickly make a persuasive screen that doesn’t respond well to users’ real-world tasks.

## From handoff to context

NN/g suggests changing the format of UX artifacts. Instead of presentations and reports that need to be interpreted manually, a team can maintain a machine-readable context close to the codebase.

The authors refer to DESIGN.md, a format for visual rules, and suggest a broader, as yet hypothetical UX.md file. It can be recorded:

• synthesis of user research and confirmed insights;
• principles of interaction: when to confirm an action, when to give a cancellation, how to formulate errors;
• product glossary and user terms;
• user models: their experiences, goals, concerns and typical barriers;
• context of use, including stressful, mobile or regulated scenarios.

The idea is not to give the model the entire archive of research. A key role is played by the curatorial work of the designer or researcher: highlight the facts that should change the AI decision.

For example, knowing that the user is working on the system for hours and making complex decisions should lead the agent to a dense expert interface rather than a simplified screen with minimal information. If a product is opened a few minutes a month, the context, on the contrary, should tilt AI toward a more understandable and directable scenario.

## Why is it important

The approach changes the very concept of handoff. Context is not transmitted once, but is constantly updated with the product and participates in each generation. Its quality can be judged not by the beauty of the document, but by how much better AI creates interfaces, texts, and working scenarios.

For product teams, this is especially important because design decisions are no longer just made by designers. The manager can ask the AI to assemble a fast layout, the engineer to add a feature with the interface, and the agent to prepare a working pull request. If all of these tools don’t have a common context, the product quickly gets a set of locally logical but inconsistent solutions.

## Where to start

There is no need to wait for the formal UX.md standard. The practical first step is to create a short Markdown file in the repository with proven UX insights, links to the design system, terminology rules, and a few examples of successful solutions.

Next, this context should be connected to AI prototyping and agency development tasks, and then check which elements actually improve the outcome. If the rule can be tested deterministically, it is better to supplement it with a linter, test or checklist review.

UX.md is not yet a finished product or an approved standard, but a methodological hypothesis called NN/g. But its logic is already applicable: the value of the designer in the AI process is increasingly shifting from preparing individual layouts to creating and maintaining a quality context for the entire system.

Source: Nielsen Norman Group