How embedding cognitive science into a design system gave every designer on the team a behavioral design superpower — and closed the gap that product managers had been filling with business jargon.
Credit Karma's design system, Thread (also known internally as KPL — Karma Product Language), was a robust visual and engineering system. But content design was largely absent from it. Designers across 8 product verticals were making content decisions in isolation, without shared language, without behavioral guidance, and without a way to prevent well-meaning product managers from filling the content vacuum with internal business terminology that meant nothing to members.
The result was a product that spoke in many voices, with inconsistent patterns, and no principled framework for how language should work alongside the components that delivered it.
When a member reads the words in any component — the label, the CTA, the error state, the tooltip — there needs to be consistency across the whole product experience so they can comprehend faster and feel safe navigating a coherent product. When language is driven by internal business logic instead of member mental models, there's friction, confusion, support costs, and loss of trust. The design system was the right place to fix this — not a style guide living in a separate document no one reads, but the actual system where designers already lived and made decisions.
The contribution produced two distinct layers of value: content standards and behavioral design guidance — both embedded directly into the design system where designers already worked.
For each component, we documented:
This was the more distinctive contribution. For each component, we identified which cognitive triggers it activated — and documented how to use that activation intentionally and ethically.
Default Effect. Pre-selected defaults leverage the Default Effect. Our guidance defined when a default is assistive versus deceptive, with a simple test: would you be comfortable explaining this default to the member?
Zeigarnik Effect. Progress indicators activate the Zeigarnik Effect — the cognitive tendency to remember and feel drawn to incomplete tasks. Guidance covered how to write progress states that feel encouraging rather than pressuring.
Loss Aversion. Alert and warning components trigger Loss Aversion — the tendency to weight potential losses more heavily than equivalent gains. Guidance addressed when loss framing is appropriate and where the ethical limits are in a financial product.
Anchoring. Recommendation cards work through Anchoring — the first number or fact a member sees disproportionately shapes their perception. Content guidance ensured the anchor is accurate and contextual, not a manipulation of perceived value.
The guidance was embedded directly into Thread's component documentation pages. When a designer went to use a component, the content and behavioral guidance was right there — not in a separate wiki, not in a Slack channel, not behind an office hours request.
Each page followed a consistent structure: what the component does, how language should work within it, which cognitive behaviors it activates, and the ethical framework for using those behaviors well. For the cross-org rollout, each content designer who had contributed became the subject-matter resource for the components they'd written — creating a distributed knowledge network rather than a single point of failure.
The launch was received with genuine enthusiasm from designers across the org. The design system gave us a way to scale content design expertise without scaling headcount — every product designer now had a framework to work from, and content designers could influence product quality even when they weren't in the room.
Over time, the behavioral design layer changed how designers thought about component selection. Instead of choosing a component for its visual properties alone, designers began asking what cognitive response it would trigger and whether that response served the member's actual goal. That shift is harder to measure than a metric, but it's visible in the quality of design conversations.
This project is a case study in how content design can function as infrastructure. By embedding behavioral guidance at the component level, we changed not just what language appeared in the product, but how designers reasoned about language decisions. The system became a teaching tool — one that made behavioral design accessible to every designer in the org, regardless of their background in cognitive science.
The most powerful content design isn't the copy you write for one surface. It's the framework that shapes every surface you'll never touch directly.