January 18, 2024

Prompting for Designers: A New Skill to Master

AIPromptingSkillsWorkflow

The New Design Skill

Prompt engineering has become an essential design skill. Not "nice to know" -essential.

Whether you're generating concept art in Midjourney, exploring UI variations in AI tools, or using ChatGPT to accelerate content creation, your ability to communicate effectively with AI determines the quality of results you get.

This isn't about replacing design skills. It's about adding a new tool that amplifies them.

What Makes a Good Prompt

Specificity

AI models work best with specific instructions. Vague prompts produce vague results.

Weak: "Design a login page" Strong: "A minimalist login page for a B2B SaaS product, centered form with email and password fields, primary blue CTA button, light gray background, subtle shadows, inspired by Linear and Notion's design language"

The second prompt gives the AI constraints to work within. Constraints improve output.

Structure

Different AI tools respond to different prompt structures:

For image generation: [Subject], [Style], [Composition], [Lighting], [Details], [Quality modifiers]

"A productivity app dashboard, clean minimalist UI design, centered layout, soft ambient lighting, glassmorphism effects, 4k, ultra-detailed, Dribbble trending"

For text generation: [Role], [Context], [Task], [Format], [Constraints]

"You are a UX writer specializing in B2B SaaS. Write 5 variations of error message copy for a failed file upload. Keep each under 15 words. Use encouraging, not apologetic tone."

Iteration

Rarely does the first prompt produce ideal results. Prompting is conversational:

  1. Start with a general prompt
  2. Evaluate the result
  3. Identify what's working and what isn't
  4. Refine the prompt to address gaps
  5. Repeat until satisfactory

This iterative loop is the actual skill -not writing the perfect prompt, but converging toward good results through refinement.

Image Generation Techniques

Style References

Don't describe style abstractly -reference known styles:

  • "In the style of Dieter Rams"
  • "Bauhaus-inspired"
  • "Similar to Apple's product photography"
  • "Rendered like a Studio Ghibli background"

AI models have seen countless images with these labels. Leveraging that training produces more consistent results than describing style from scratch.

Composition Guidance

Direct the image structure:

  • "Rule of thirds composition"
  • "Centered subject with negative space"
  • "Shot from above, flat lay"
  • "Close-up detail shot"
  • "Wide establishing shot"

Without composition guidance, AI makes arbitrary choices.

Technical Specifications

Include technical parameters:

  • "4K resolution, high detail"
  • "Studio lighting, soft shadows"
  • "Depth of field, bokeh background"
  • "Matte finish, no reflections"
  • "Isometric perspective"

These technical terms significantly affect output quality.

Negative Prompts

Many tools support negative prompts -what to exclude:

  • "No text, no watermarks"
  • "Avoid cluttered backgrounds"
  • "No people, no hands"
  • "Exclude cartoonish elements"

Negative prompts prevent common unwanted artifacts.

Text-to-Design Workflows

Concept Exploration

Use AI to explore solution spaces quickly:

"Generate 10 different navigation patterns for a mobile app with 5 main sections. Include both bottom tabs and hamburger menu variations. Describe the pros and cons of each."

You're not asking AI to design -you're asking it to enumerate options for your consideration.

Content Generation

AI excels at volume:

"Write 20 different tagline options for a project management tool. Target audience is startup founders. Emphasize simplicity and speed. Maximum 6 words each."

You review and select. AI generates candidates.

Variation Generation

Once you have a direction, generate variations:

"Here's my hero section copy: [paste]. Create 5 alternative versions that: 1) Make it shorter, 2) Make it more emotional, 3) Add urgency, 4) Target enterprise users, 5) Incorporate social proof."

Exploring variations manually is slow. AI makes it fast.

Documentation Assistance

AI accelerates documentation:

"Based on this component specification [paste], generate user-facing documentation that explains when to use this component and when to avoid it. Include 2 good examples and 2 anti-patterns."

You refine and validate. AI creates the first draft.

Iterative Refinement Patterns

The Funnel Pattern

Start broad, narrow progressively:

  1. "Generate concepts for a fitness app onboarding flow"
  2. "Expand on concept #3 with more detail"
  3. "Refine the goal-setting step specifically"
  4. "Write the final copy for this screen"

Each iteration focuses tighter until you reach specificity.

The Comparison Pattern

Generate multiple approaches, then compare:

  1. "Create approach A that emphasizes gamification"
  2. "Create approach B that emphasizes simplicity"
  3. "Compare these approaches for [target user type]"

AI can articulate trade-offs you might miss.

The Critique Pattern

Use AI to critique AI output:

  1. "Generate a landing page structure"
  2. "Now critique this structure from a UX perspective"
  3. "Based on that critique, generate an improved version"

Self-critique often produces improvements.

The Expert Pattern

Assign expertise for better results:

"As an accessibility expert, review this component design and identify potential issues for screen reader users."

Expert framing accesses different training patterns.

Building Your Prompt Library

Save What Works

When a prompt produces excellent results, save it. Build a personal library organized by use case:

  • Concept exploration prompts
  • Copy generation prompts
  • Critique and review prompts
  • Documentation prompts
  • Image generation prompts by style

A library turns hard-won discoveries into reusable assets.

Create Templates

Build fill-in-the-blank templates for common tasks:

"Generate [number] variations of [content type] for [product/feature]. Target audience is [audience]. Tone should be [tone descriptors]. Each should be [constraints]."

Templates accelerate new projects.

Version Your Prompts

When you refine a prompt, keep both versions. Note what the refinement improved. Your prompt library becomes a learning resource.

Share and Learn

Prompt engineering has a community. Share prompts that work well. Learn from what others share. The field moves quickly; community learning matters.

Platform-Specific Tips

ChatGPT / Claude

  • Use system prompts to establish persistent context
  • Break complex tasks into steps
  • Ask AI to confirm understanding before proceeding
  • Use temperature settings (when available) for creativity vs. precision

Midjourney

  • Master the parameter syntax (--ar, --style, --chaos, etc.)
  • Use image prompts for style consistency
  • /describe uploaded images to learn prompt vocabulary
  • Remix mode enables iterative refinement

Figma AI (and similar tools)

  • Reference your existing design system
  • Be explicit about component names and variants
  • Use screenshot references when available
  • Accept that these tools are early and improving

Specialized Tools (Galileo, Uizard, etc.)

  • Understand each tool's strengths and limitations
  • Provide context about your brand and constraints
  • Expect to heavily edit output -these are starting points
  • Combine multiple tools for best results

The Designer's Role

Prompting doesn't replace design judgment. It changes where judgment applies:

Before: Judgment in creation Now: Judgment in direction-setting and curation

You still need to know what good design looks like. You still need to understand user needs. You still need to apply strategic thinking.

What changes is the execution path. Instead of manually creating every option, you describe options and curate results. The design expertise remains essential; the hands-on execution becomes collaborative.

Avoiding Prompt Pitfalls

Over-reliance

AI output needs human judgment. Don't accept results just because they're generated. Evaluate everything against your design standards.

Inconsistent Results

AI has randomness. Running the same prompt twice produces different results. Build review processes that catch quality variations.

Confidentiality

Don't put confidential information in prompts to external AI services. Understand data policies before sharing sensitive content.

Stale Prompts

AI tools evolve. Prompts that worked last month may work differently now. Review and update your prompt library periodically.

The Learning Curve

Prompting skills develop through practice:

Beginner: Following others' prompts verbatim Intermediate: Adapting prompts to specific needs Advanced: Creating prompts from scratch for novel situations Expert: Understanding why prompts work and teaching others

Don't expect expertise immediately. Like any skill, prompt engineering requires deliberate practice.

Starting Your Practice

Begin with low-stakes experiments:

  1. Pick one AI tool and learn it well
  2. Start with tasks that don't have deadline pressure
  3. Save every prompt (successful or not) with notes
  4. Compare results across different prompt approaches
  5. Join communities and learn from others

Within a few weeks of regular practice, you'll see significant improvement. Within months, prompting will feel like a natural extension of your design process.

The skill is real. The investment pays off. Start now.

What prompting techniques have worked well for you? I'm always collecting new approaches.