Case study — Content standards and AI
Standards that outlast the team
Northbeam · Reconstruction · Fictional brand
I authored the criteria, prompted the model, and built the Evaluative Content Design Checklist — 94 criteria, 26+ UI components, 6 design patterns.
The Challenge
Associate-facing teams and consumer-facing teams were writing without a shared global content design standard. Alerts, modals, text fields, and toasts behaved differently by team. Chat bots had no single standard, which created legal risk. Engineering paid for the drift in rework.
The gap was not only punctuation. It was how content behaves inside the components used to build products.
Two systems had to come together: Storefront (consumer) and Floor (associate). I owned Floor content. Consumer digital already through review and legal stayed out of scope.
Then the content design function was eliminated. A Figma page with no owner does not scale. Teams still shipping UI needed a way to align to the standard without a content designer on the squad.
The Approach
I led bringing Storefront and Floor under one content standard in Figma. The standard covered word usage and how content behaves in components: button labels name the action (verb + noun); one action per button; title case on buttons and labels; sentence case on body, helper text, errors, and badges; alerts inform and recover; they do not alarm.
Leave-a-PDF-and-hope was killed by the org change. Two separate standards were killed — that was the original drift.
What shipped on top of Figma was the checklist I built. It scores real UI against Storefront content design: error messages and alerts; clear language; capitalization and punctuation; component-specific content; voice and tone; accessibility and inclusion; research and testing; content business strategy.
Each item returns Pass, Warning, Fail, or Indeterminate. Indeterminate means the screenshot cannot prove the claim (alt text, research, KPIs). The tool does not fake a pass. It also returns a rewrite.
"Multiple model numbers are associated to this item number. Scan item barcode to add item to cart."
→ "This item has multiple model numbers. Scan the barcode to add it to your cart."
Item Entry
→ Add Item. Same pattern: Customer Lookup → Look Up Customer.
A run on raw system-exception copy scored 3 pass / 18 warning / 57 fail / 25 indeterminate.
The Result
One Figma source of truth across Storefront and Floor. A workflow product teams can run before launch after content design was no longer staffed on the squad.
No separate business KPI on file for this case. Do not cite pass-rate chips from the reconstruction as results — those chips are illustrative.