AI
Prompt Engineering for Designers: A Practical Guide
Stop wrestling with vague prompts. A designer's framework for writing AI prompts that actually produce usable brand and web concepts.
Contents
Introduction
Most prompt guides are written for developers. Designers need something different: prompts that produce usable creative direction, not just interesting images.
Bad prompting for design sounds like: "modern minimalist logo for tech startup." That will get you a generic gradient mark that could be anything. Good prompting sounds like a creative brief: specific, opinionated, constrained.
At Designing Dots, we've seen teams waste weeks in prompt loops because they are trying to get AI to have taste. That's not how it works. Your taste has to be in the prompt. This guide gives you a structure to do that consistently.
The Mindset Shift: From Prompting to Briefing
Stop thinking "what words will trick the model." Start thinking "how would I brief my most talented junior to get a useful first attempt?"
A good design brief includes: objective, audience, tension, references, constraints, and what to avoid. A good AI prompt includes exactly the same things. The difference is execution speed.
If you can't explain it to a human, the model won't fix it for you.
The 6-Part Prompt Framework for Designers
We use this framework for both visual models (Midjourney, DALL·E, Firefly) and language models (ChatGPT, Claude) when doing brand work.
1. Role + Context
Tell the model who it is and what it's working on. Not "act as a designer" — that's vague. Be specific.
Example: "You are an art director for a B2B SaaS brand. We help ops teams at 50-200 person logistics companies reduce shipment errors. Our current identity feels too playful for enterprise buyers."
2. Objective + Output Format
State what you need and how it should be delivered.
Example: "Generate 8 art direction territories for homepage hero imagery. For each, give: name, 2-sentence concept, visual traits (light, composition, palette range), what to avoid."
This prevents rambling paragraphs and gets you scannable options.
3. Audience + Tension
This is the difference between generic and useful.
Example: "Our buyer is a 38-year-old ops lead, pragmatic, burned by flashy tools. Tension: they want innovation but need reliability. Visuals must signal 'we've done this before, at scale' not 'we're the next big thing'."
4. References + Taste Anchors
Models respond better to specific visual language than adjectives like "premium."
Bad: "premium minimal style"
Good: "Reference: Aritzia campaign lighting, soft overcast; Swiss grid from Josef Müller-Brockmann; color restraint like early Everlane, not like luxury fashion"
Create a taste bank for your brand: 15-20 reference pairs with notes on what you like about each. Reuse it. At Designing Dots, every brand has a taste doc we paste into prompts.
5. Constraints
Constraints are your best friend. They force ownability.
Include:
- Palette limits: "Use only muted clay, off-white, charcoal. No pure black, no bright blue."
- Typography direction: "Grotesk sans only. No serifs. No geometric circles."
- Do-not-do list: "No gradients, no isometric illustrations, no AI human hands, no techy circuit patterns"
- Production constraint: "Must work as single-color embroidery and as 16px favicon"
6. Evaluation Criteria
Tell the model how you will judge output. This improves first-pass quality.
Example: "Score each concept 1-5 on: ownability vs. logistics competitors, scalability across website and deck, believability for enterprise."
Prompt Templates You Can Steal
For brand territories (image model)
"Territory exploration for [brand], [category]. Concept: [one idea, e.g., 'precision as quiet confidence']. Visual direction: [lighting, texture, composition, lens, palette]. Inspired by [3 specific references]. Must avoid [list]. Show as [format: poster grid / texture board / minimal object study]. No text, no logos."
For brand voice (language model)
"Write 3 voice samples for [brand] in this scenario: [e.g., 404 page]. Constraints: Audience is [x]. Vocabulary to avoid: [y]. Sentence length: [short/mixed]. Personality: [2 adjectives] but not [1 anti-adjective]. Reference voice: [e.g., Stripe documentation, direct but not cold]. Give options plus one line on why each fits strategy."
For website hero exploration
"Generate 10 hero headline directions for [brand] that communicate [value]. Audience: [role]. No buzzwords [list buzzwords to avoid]. Each must follow structure [e.g., Verb + outcome in 6 words]. Rank by clarity and differentiation from competitors: [list 2 competitors]."
Iteration Method: 30-Minute Sprints, Not Infinite Tweaks
Most designers prompt like this: generate → tweak word → generate → tweak → repeat for hours.
Better:
- Sprint 1 (15 min): Generate 12 wide variations using framework.
- Curation (10 min): Kill 9, keep 3. Write why.
- Sprint 2 (15 min): Push only the 3 kept directions with tighter constraints.
If after 2 sprints you don't have something worth rebuilding in Figma, the brief is wrong, not the model.
Why this matters for design teams
Prompt engineering is not a separate role. It's a core craft skill. The teams that document prompts, share what works, and build a shared taste library ship faster and more consistently. It's the difference between everyone prompting differently and a team developing a recognizable point of view.
Common mistakes
- Prompting for the final asset vs. prompting for direction to rebuild.
- Using adjectives without references. "Premium" means nothing to a model trained on everything.
- One giant mega-prompt. Break it into brief → generate → evaluate → rebuild. You get better control.
- No negative prompts or avoidance list. What you don't want is as important as what you want.
- Saving images instead of saving prompts. The prompt is the system.
Designing Dots Approach / Framework: Prompt as Creative Direction Document
We treat every strong prompt set as a Creative Direction Document (CDD Lite) that includes:
- Inputs: brand tension, audience, references, constraints, anti-references
- Outputs: 3 territories, each with name, look/feel, language, behavior, and 1 paragraph on why it is ownable
- Decision: which territory we will rebuild humanly and what we keep from AI
No AI image ever goes to client as final. We rebuild everything. The prompt's job is to accelerate our ability to agree on direction, not to replace craft.
Key Takeaways
- Write prompts like you write briefs: specific, constrained, reference-rich.
- Use a consistent framework: Role, Objective, Audience/Tension, References, Constraints, Evaluation.
- Build and reuse a taste bank for each brand.
- Time-box exploration and force curation.
- The goal is a direction worth rebuilding, not a perfect AI render.
Next Steps / Internal Ecosystem
Build a one-page taste bank for your current brand: 10 images you love, 10 you reject, with one sentence why for each. Then use the 6-part framework to prompt 3 territories in under an hour.
Keep going with AI for Designers: How AI Is Changing the Design Workflow and How to Use AI for Brand Strategy Without Losing the Human Idea — or put it into practice with DOTS BRIEF before your next review.