01
Start with the person's situation
Name the audience, their current context and the action they need to complete. 'A mobile dashboard' defines a format. 'A field technician checking the next visit on a phone with unreliable reception' exposes decisions about priority, connectivity and density.
Keep assumptions visible. If you do not know which task is most frequent, mark it as an assumption to validate. An AI-generated persona is not a substitute for evidence from customers or product usage.
02
Supply content and constraints before decoration
List the information users need to make a decision: names, dates, statuses, prices or permissions. Include representative examples with difficult lengths and missing values. Avoid using real personal data in a demonstration brief.
State existing constraints such as the component library, brand requirements, supported layouts and data source. If the application already has a design system, tell the agent where to find it. Decide whether this task is a visual prototype or a connected feature.
- Audience and context: who is acting, where and under what conditions.
- Primary task: the decision or action the screen should support.
- Content: available fields, realistic examples and missing-data cases.
- Constraints: existing components, technical scope and brand requirements.
- Acceptance: observable behavior required for a usable result.
03
Turn adjectives into acceptance criteria
Replace 'intuitive' with a condition you can inspect. For example: the next appointment and its location are visible before secondary reports; an unsuccessful save retains entered values; the primary task can be completed with a keyboard.
Avoid prescribing every pixel before understanding the content. Ask the agent to explain the hierarchy and unresolved questions first when a brief contains competing goals. Document the chosen tradeoff so later revisions do not silently reverse it.
04
Review with a separate scenario
After generation, give someone a task that describes a goal without telling them the control names or the intended route. Watch what they try. Record confusion and incorrect outcomes rather than asking only whether they like the appearance.
Use the findings to revise the brief or interface. If the person cannot find the location of a visit, inspect the hierarchy before adding decorative emphasis everywhere. One session can reveal a problem; it cannot establish a population-wide success rate.
Worked example
A brief for a field-service schedule
Reusable fictional example. Replace the fields and assumptions with your own product facts.
Starting point
Create a beautiful responsive dashboard for our service business.
More actionable
Audience: field technicians reviewing today's visits on a phone. Goal: identify the next visit and open its details. Content: start time, customer placeholder, address, status and contact action. Constraints: reuse existing components; mock data only; do not claim offline support. Include loading, no visits and failed loading. Acceptance: the next visit is prioritized, keyboard navigation works, and a failed refresh does not erase visible details.
The brief gives the agent a product decision to implement and explicitly distinguishes mock data from capabilities that would require engineering work.
Take into your next review
Review checklist
- The brief identifies an audience and realistic context.
- There is a clear primary task.
- Representative content and edge cases are supplied.
- Technical scope and unresolved assumptions are explicit.
- Acceptance criteria describe observable outcomes.
- The review task does not tell the participant where to click.
Common questions
Do I need design expertise to write a UI brief?
You can start with product knowledge: who uses the screen, what they need to do and what information they need. Design expertise can help resolve tradeoffs, but detailed style terminology is not a prerequisite for a useful brief.
Should I ask the AI to make all missing decisions?
For low-impact exploration, suggested assumptions can help. Keep them labeled. Confirm business facts, critical workflows and technical capabilities before treating a draft as an implementation specification.
What should I include in a UI prompt for Cursor, Claude Code or Codex?
Include the audience, task, representative content, existing components, technical scope and acceptance conditions. Point to relevant project files. This brief is tool-neutral; confirm each host's current instruction mechanisms separately rather than assuming identical behavior.
How do I brief an AI to redesign an app without changing behavior?
List the flows and states that must remain, identify permitted visual changes and supply representative screens. Ask for a review plan before editing. Verify validation, navigation and task completion afterward; visual similarity does not establish behavioral equivalence.
Sources & scope
Primary references checked on November 15, 2026. These sources support the principles cited beside them. The workflow, briefs and checklists are Skillux editorial recommendations; the examples are illustrative, with no claimed performance results.
W3C Understanding documents explain individual success criteria; this article is not a complete WCAG conformance assessment. References do not imply endorsement of Skillux. Read our editorial approach.