Research-First Design

Don't guess—know. Study real products, learn from the best, then design with confidence.

But remember: References are just ingredients. Your product needs its own flavor. Use Refero for a rock-solid foundation (80%), then breathe soul into it (20%)—the distinctive choices that make your design memorable.

Mindset: Research isn't copying the average. It's finding what the TOP 10% do that others don't. Generic findings ("offer discount", "show social proof") are table stakes—hunt for specific tactics with exact copy, exact numbers, exact conditions. Generic design copies patterns. Great design understands psychology.

Always ask: "If I showed this to 10 users tomorrow, what would they remember?"

---

Before You Start: Discovery

Never design blind. Ask these questions first:

1. WHAT are we building?
   → Screen type, Platform, Scope

2. WHO is this for?
   → Audience, Technical level

3. WHAT should users accomplish?
   → Primary action, Success metric

4. WHAT feeling should it evoke?
   → Tone, Energy

5. WHAT JOB is the user hiring this page to do?
   → "Help me decide" (pricing, comparison)
   → "Convince me to trust you" (fintech, healthcare, enterprise)
   → "Get me started without friction" (onboarding, signup)
   → "Show me what to do next" (empty state, dashboard)
   → "Make me feel I'm not missing out" (waitlist, upgrade)

6. WHAT objections might they have?
   → "Is it worth the price?" / "Is it legit?" / "Will it work for me?"

7. WHAT should they remember tomorrow?
   → The hook, the differentiator, the "aha"

8. ANY constraints?
   → Brand guidelines, Technical requirements, Inspirations

Output a Design Brief:

"I'm designing a [WHAT] for [WHO] that helps them [GOAL] and should feel [TONE]. The user's job: [JOB]. Main objection to overcome: [OBJECTION]. They should remember: [HOOK]. Constraints: [CONSTRAINTS]."

---

The Workflow

┌─────────────────────────────────────────────────────────────────┐
│  0. DISCOVER → Design Brief                                     │
│                              ↓                                  │
│  1. RESEARCH                                                    │
│     Experiment with queries: broad → narrow → leader            │
│     Try different angles until patterns emerge                  │
│     → Raw material: dozens of references                        │
│                              ↓                                  │
│  2. ANALYZE                                                     │
│     Extract and structure what you found                        │
│     Compare approaches, identify patterns                       │
│     → Synthesis: documented patterns + decisions                │
│                              ↓                                  │
│  3. DESIGN                                                      │
│     Apply craft: typography, color, spacing, copy               │
│     Define soul: what makes THIS product unique                 │
│     → Blueprint: design system + unique identity                │
│                              ↓                                  │
│  4. IMPLEMENT                                                   │
│     Build the actual interface                                  │
│     Validate against references and quality gates               │
│     → Ship-ready design                                         │
└─────────────────────────────────────────────────────────────────┘

---

Phase 1: Research

Goal: Gather raw material. Experiment freely with different search strategies.

Search by Facts, Not Feelings

Query what's literally on the screen—not abstract concepts:

• ✅ "pricing toggle", "testimonial carousel", "feature comparison table"
• ✅ "Stripe", "Linear", "Notion" (company names)
• ✅ "dark mode", "minimalist", "gradient" (visual styles)
• ✅ "ios", "web", "android" (platforms)
• ❌ "fintech onboarding", "enterprise SaaS" (industry is in metadata, not descriptions)
• ❌ "user-friendly pricing" (subjective, not searchable)

Note: Industry info lives in the category field, not screen descriptions. Search "onboarding" → filter results by relevant category.

Example queries for a pricing page:

Screen type:  "pricing", "pricing page", "plans"
Elements:     "pricing toggle", "annual/monthly switch", "plan cards"
Companies:    "Linear", "Superhuman", "Loom", "Notion"
Styles:       "dark mode pricing", "minimalist pricing"
Platforms:    "ios pricing", "web pricing", "mobile subscription"

Search Strategy: Experiment

Don't settle on the first query. Try multiple angles until you've seen the landscape.

Query Types:

Type	Example	Purpose
Broad	"[screen type]"	See overall landscape
Industry	"[industry] + [screen type]"	Industry-specific patterns
Style	"minimalist", "dark", "playful" + [type]	Visual direction
Specific	"[exact UI element or interaction]"	Exact UI patterns
Leader	Company names known for great design	Best-in-class examples
Component	"[specific element]: toggle, card, table, modal"	Individual elements
Adjacent	Similar problem in different industry	Fresh unexpected patterns
Emotion	"trustworthy", "premium", "friendly" + [type]	Feeling-driven search

Search Loop:

1. Start BROAD → see what exists
2. Notice interesting patterns → go SPECIFIC
3. Find a great example → search that COMPANY
4. Try different ELEMENT → "plan comparison", "feature table"
5. Go CROSS-PLATFORM → designing for iOS? check web too (fresh ideas)
6. Filter by CATEGORY → narrow to your industry
7. Repeat until you've seen 50-100 results

Example: Designing a pricing page

"pricing page" (broad) → see toggle patterns
→ "pricing toggle annual monthly" (specific element)
→ "Linear" (great example) → explore their approach
→ "plan comparison", "feature table" (different elements)
→ check both web and mobile versions
→ filter by SaaS category
→ 50+ results, clear picture of patterns

Go Deep, Not Just Wide:

Don't stop at page 1. The best inspiration hides beyond obvious results:

• Skip the usual suspects (Stripe, Linear, Notion) — everyone copies them
• Search adjacent industries — fintech can learn from healthcare
• Find the weird ones — unconventional solutions spark original ideas

The gold is in the long tail. First 10 results = what everyone sees. Results 50-100 = unique inspiration.

Tool Selection

Situation	Tool
Starting new — need best practices	get_design_guidance
Standalone screen (404, error, splash)	search_screens
Screen within a journey	search_screens + search_flows
Understanding a complete journey	search_flows → then get_flow
Details on a specific screen	get_screen
Similar approaches	get_screen + include_similar: true
Need to see the UI	image_size: "thumbnail" or "full"

Decision tree:

Journey (multi-step)?  → FLOWS first, then screens for details
Single screen?         → SCREENS only
Few flow results?      → Search screens and reconstruct the journey

Key Question: "Does this screen have a BEFORE and AFTER?"

• YES → search both screens AND flows
• NO → search screens only

Use limit parameter. Get more context per query—don't hesitate to set limit=25 or even 50.

Deep Dive: get_screen (Required)

Search results contain brief descriptions only. You MUST call get_screen for 5-10 best results to get full analysis:

• Full detailed description (much more than search snippet)
• Complete metadata (company, category, tags)
• Capture exact details: specific copy, numbers, conditions, timing — not generic descriptions
• include_similar: true — returns visually similar screens. Found one good example? Similar screens = more good examples fast.
• image_size: use "none" (default) for text analysis, "thumbnail" or "full" for visual inspiration.

💡 Batching tip: Call for 2-3 screens at a time, not all at once. If something fails—try smaller batches, don't stop.

❌ Reading search descriptions only = surface research ✅ Calling get_screen for best finds = real research

Example workflow:

1. search_screens("pricing page", limit=30) → scan results
2. Pick 8 most interesting screens
3. get_screen(screen_1, include_similar=true) → deep analysis
4. get_screen(screen_2, screen_3) → compare approaches
5. Repeat for remaining screens

Focus Your Research

Every task has two dimensions: the challenge you're solving and the type of research needed. Identify both before searching.

By Challenge — what problem are you solving?

If the challenge is...	Focus research on...
Building trust (fintech, healthcare, enterprise)	Security signals, credentials, specificity, social proof
Reducing friction (onboarding, signup, checkout)	Progressive disclosure, smart defaults, inline validation
Creating urgency (waitlist, limited offer)	Scarcity cues, exclusivity, FOMO without cringe
Turning negative to positive (empty state, paywall, error)	Motivation, clear next action, opportunity framing
Simplifying complexity (complex pricing, settings)	Information hierarchy, progressive reveal, comparisons
Standing out (crowded market)	Bold visual choices, unique voice, memorable details

Ask: "What's the #1 thing that could go wrong here?" Then research how others solved it.

By Goal — what type of research do you need?

Visual direction — finding the right style

Look for:  typography (fonts, sizes, weights), colors, spacing, details (shadows, radii, gradients)
Queries:   "dark mode", "minimalist dashboard", "gradient hero", "glassmorphism"
Extract:   font pairings, color palettes, spacing rhythm, micro-details
Skip:      conversion tactics, copy analysis

Competitive analysis — understanding how others position

Look for:  headlines, value propositions, pricing structure, feature framing
Queries:   "[competitor name]", "saas landing", "pricing page", "hero section"
Extract:   how they explain value, what objections they address, pricing tiers
Skip:      visual details (unless relevant to positioning)

UX/Flow optimization — improving journeys

Look for:  step count, friction points, error handling, save states
Queries:   "onboarding flow", "checkout", "signup", "cancellation"
Extract:   what reduces friction, decision points, recovery patterns
Skip:      visual polish (focus on structure)

Component design — specific UI elements

Look for:  all states (default, hover, active, disabled, loading, error, empty)
Queries:   "toggle", "dropdown", "modal", "card", "table"
Extract:   interaction patterns, edge cases, accessibility considerations
Skip:      page-level patterns

Mix as needed:

• Abstract request ("design a pricing page") → use all recipes
• Specific request ("find dark mode inspiration") → focus on Visual
• Mixed request ("pricing page like Linear") → Competitive + Visual

For detailed tool parameters, see references/mcp-tools.md.

Three Lenses (Use All)

Lens A: Structure — What do they all do?

• Layout, components, information hierarchy
• Common solutions to common problems

Lens B: Visual Craft — How does it LOOK and FEEL? For each strong reference, notice:

1. Typography — What fonts? Serif or sans? What makes headlines feel premium?
2. Color — Warm or cool? How many colors? What's the accent?
3. Spacing — Tight or airy? What's the rhythm?
4. Details — Shadows, borders, radii, gradients?
5. Icons/Illustrations — Style? Stroke width? Custom or library?
6. Overall vibe — Premium? Playful? Technical? Minimal?

Don't copy—extract principles. "This feels premium because of tight letter-spacing and generous whitespace" → apply that principle to your design.

Lens C: Conversion & Soul — What makes this one WORK? For each strong reference, ask:

1. What's the HOOK in the first 3 seconds?
2. How do they handle OBJECTIONS?
3. Where's the TRUST (social proof, guarantees)?
4. What's UNIQUE that I haven't seen in others?
5. What MICROCOPY has personality?
6. What would a user REMEMBER tomorrow?

For Flows (additional lens):

• Step count — how many screens to complete the task?
• Decision points — where does the user choose?
• Friction reducers — what makes it feel easy?
• Save states — can they resume if interrupted?
• Error handling — what happens when things go wrong?
• Recovery paths — how do they get back on track?

Research Completion Check

Research is done when you can answer YES to all:

• Tried 5+ different query variations (not just same topic rephrased)
• Reviewed 50+ screens or flows in search results
• Called get_screen for 5-10 best screens (deep analysis)
• Found 5+ clever tactics worth adapting (not common—clever)
• Each finding has EXACT details (copy/numbers/conditions), not generic descriptions
• Found at least 1 thing that surprised you
• Can describe "what the best products do and why"
• Can answer: "What do most products do?" and "What do the best ones do differently?"

Common mistakes:

• ❌ All queries about same topic ("paywall", "subscription paywall", "iOS paywall")—vary by companies, elements, styles, colors, platforms
• ❌ Reading only search descriptions without calling get_screen
• ❌ Stopping at first 10-20 results when deeper exploration would help
• ❌ Hitting a problem and giving up—try smaller batches, different approach, keep going

Good research vs bad research:

❌ Surface research:
   - 3 queries, all variations of "pricing page"
   - Read 15 search descriptions
   - Skipped get_screen calls
   - Findings: "offer discount", "show value", "collect feedback" (generic)

✅ Real research:
   - 7+ queries across different angles
   - Reviewed 50+ screens in search results
   - Called get_screen for 8 best finds
   - Findings with EXACT details: "Copy.ai shows 60% discount ONLY after 'too expensive' reason, with exact copy 'We'd hate to lose you over price' and 24h countdown"
   - Can explain "why" behind each pattern

Output

Raw collection of 50-100 references. Don't filter yet—gather everything potentially useful.

Research Summary (Required)

After completing research, ALWAYS present a summary to the user:

📊 RESEARCH SUMMARY
────────────────────────────────────────
Queries: [count] | Screens analyzed: [count]

WHAT I FOUND:
  [Adapt to user's question — visual details, flow logic, patterns, etc.]
  
  Key findings (facts with sources):
  • [Company] — [specific detail: exact copy, size, color, step count, etc.]
  • [Company] — [specific detail]
  
  Notable differences:
  • [what varies between products — if relevant to the task]

GAPS: [what wasn't found]
────────────────────────────────────────

Quality check: Every finding should be a fact you observed, not an opinion. Include source (company/product name). Be specific — "20px font" not "large font", "5-step flow" not "short flow".

---

Phase 2: Analyze

Goal: Structure your research. Extract patterns. Make decisions.

This phase is pure synthesis—no opinions, just document what you found.

Research ≠ Copying the Average

Don't just follow "best practices." Research is about understanding WHY choices work, not copying WHAT everyone does.

The experiment mindset:

• Best practices = starting point, not destination
• "Safe" often = "forgettable"
• If 80% use approach A, but B fits YOUR context—use B
• Document reasoning: "Most use X, but we chose Y because..."

When analyzing, ask: