Designing Verification Frameworks for Trusted AI

• Role: End-to-End Design Lead

• Scope: Enterprise-wide pattern across multiple clouds, platforms, and Agentforce.

• Objective: Define a scalable AI verification pattern in the absence of a single PM or PRD.

The Scaling Problem

• Citations were originally built for a single platform (Lightning) and failed to scale for multi-cloud AI.

• No shared interaction contract existed across AI products.

• Teams were building fragmented, product-specific solutions.

• Legal requirements for source transparency lacked a system-level solution.

Friction in Trust & Verification

• Reading Flow: Inline citations disrupted the user's natural reading path.

• Density: High citation density reduced overall comprehension.

• Displacement: Expanding source lists pushed main AI content off-screen.

• Evidence: Low engagement with full source disclosure confirmed a usability breakdown.

The Human + AI Mental Model

• Core Principle: Trust must be inspectable, not assumed.

• Explainability: Using tooltips and clear function indicators.

• Provenance: Providing clear links, timestamps, and author metadata.

• Predictability: Ensuring consistent behavior during model failure or errors.

Strategy & User-Driven Tradeoffs

• Tested: Evaluated pagination vs. scrolling.

• Insight: Scrolling significantly reduced friction compared to pagination.

• Insight: Citation markers (inline) are the most critical signals for reinforcing trust.

• Decision: Moved to a "Popover-first" model to preserve user context.

Flexible, Container-Agnostic Design

• Primary Interaction: Scrollable popover for immediate verification.

• Secondary Interaction: Deep-disclosure panel for exhaustive source auditing.

• Adoption Paths: Supports "Popover only" or "Popover + Panel" depending on product maturity.

Enterprise-Wide Adoption

• Adoption: 100% of Salesforce AI text output now follows this pattern.

• Reach: Successfully scaled across 14 distinct product teams.

• Efficiency: Eliminated ad-hoc design work through a centralized Figma library.

• Strategic Shift: Moved the org from feature negotiation to goal-based alignment.