Prompt Pattern
A prompt pattern is a four part formula that declares format purpose audience and constraints so AI tools like Figma Weave output structured layout scaffolds instead of generic wireframes. You lead with the exact canvas type then the job it must perform the people it serves and the non negotiable rules on density hierarchy spacing and content volume. This treats Weave like the eager junior with infinite patience and zero taste that it actually is. The tool cannot see your component library cannot read your design tokens and has no clue what your brand voice demands. The pattern fills those gaps with surgical structural commands that map straight to layout intent component types and content hierarchy. Designers who adopted this in 2026 stopped deleting seven bad generations and started reacting to frames worth editing. The pattern turns the blank canvas problem into a 30 second starting point that respects real product constraints instead of hallucinating another generic SaaS dashboard.
It is not vague natural language requests that leave the machine to guess everything. It is not the 2025 prompt engineering theater where you role play as a principal designer at Linear with impeccable taste and tell the model to think step by step. Those approaches generated pretty screenshots that collapsed the moment they became real Figma layers. It is not hoping Weave will magically align with your spacing scale or pull instances from your actual library because it literally cannot see them. The pattern rejects all that fluff and stays brutally concrete because concrete is the only dial Weave can turn reliably.
The Modal team deployed one in early 2026 to launch their AI inbox product. The prompt read 5 section SaaS landing for a B2B AI inbox tool targeted at startup CTOs. Hero with headline subheadline and immediate email capture form. Social proof row with logos from OpenAI Anthropic Perplexity and Replicate. Three column features grid with icon headers and two line descriptions. Pricing table with two tiers annual toggle prominent and feature comparison rows using checkmarks. Bottom CTA banner with secondary learn more link. Dense information hierarchy no full bleed hero image compact cards with 16 pixel padding and tight leading. Weave dropped a frame with correct vertical rhythm balanced card heights and logical column weights. The team swapped real components and tokens in 18 minutes instead of rebuilding layout from scratch.
Flexport analysts used a tighter pattern for internal analytics. Analytics dashboard for logistics SaaS with heavy data density. Left nav with exactly 6 icon items for shipments routes analytics alerts settings and account. Top bar containing global search notification bell and user avatar. Four KPI cards in hero row each with embedded sparkline. Main area filled by large multiline chart with four series toggleable legend and date range picker. Below it a 10 row sortable table for recent shipments with status pills and tracking links. Right rail for active filters. Monospace numbers minimal padding 14 pixel base font maximum data per square inch. The scaffold respected every count and proportion allowing engineers to map it to production code with almost no spatial rework.
Ramp fintech designers needed mobile precision so they wrote single mobile onboarding screen for business credit card signup flow. Progress stepper at top showing step 3 of 5 with completed states checked. Centered card 380 pixels wide on light gray background inside safe areas. Two text fields for legal name and tax ID. One dropdown for company size with example values prefilled. Checkbox with linked terms. Primary CTA button 48 point height locked to bottom of viewport. No bottom tab bar on this view back button top left icon only. Weave output correctly sized touch targets and a card that translated cleanly into their native component library across iOS and Android.
The Ghost content platform team ran an admin pattern that read internal admin interface for publishing platform. Top nav with logo workspace switcher search and user menu. Sidebar with 8 grouped sections under content members billing settings and analytics. Main content starts with three tabs for published drafts and scheduled. Below tabs a searchable 12 row table of posts with title author publish date status pill and quick action menu. Floating action button bottom right for new post. Data dense layout 14 pixel base font dark mode optimized subtle dividers. The generated frame correctly weighted the sidebar against the content area and suggested empty states that matched their real UI kit.
Reach for prompt patterns the instant you open a blank Figma file with only a brief in hand. They crush greenfield exploration client alignment sessions and early validation of information hierarchy. Teams at Modal Flexport Ramp and Shopify partners all reported 4x faster starts in 2026 when Weave first rolled out. The patterns also translate cleanly to v0 or Lovable when the handoff skips Figma entirely.
Stop using them once your design system ships First Draft templates for 80 percent of screens or when the work shifts to brand specific expression like the 2025 Pitch redesign that lived on custom illustration rules no AI could guess. Never use them for accessibility critical first passes or inside locked enterprise systems where every element must instantiate from the approved library on day one. In high fidelity iteration files the pattern becomes pure overhead.
Prompt patterns turn Weave from a mediocre slot machine of layouts into the best junior designer you never have to babysit.
Read the full guide
Related terms
Keep exploring
Prompt Surface
The full UI component surrounding an AI text input with empty states, suggestions, attachments, model pickers, tool toggles, streaming output, and revision controls that turns prompting into a structured, observable interaction.
Prompt Engineering
The practice of writing instructions that produce consistent, usable output from a language model. Functionally identical to writing a good creative brief.
Few-Shot Example
A few-shot example is a set of three to five real before-and-after pairs pulled from past team work and baked directly into a prompt so the model copies proven taste instead of guessing at vague rules.
Output Spec
Output Spec is the final section of a structured prompt that dictates the exact technical requirements for the deliverable including dimensions file formats code standards component variants and naming conventions.