Does consistent AI labeling actually reduce automation bias in practice?

IBM's Carbon Design System's "Carbon for AI" guideline defines a single, reusable pattern for marking AI presence across IBM products: a styling language plus one component, the AI label. The guideline states the goal directly — "Transparency of AI presence is key to user trust. Having a consistent identity helps build awareness and anticipation of AI presence across our experiences" — and assigns that job to "Carbon for AI styling," which "builds on the core Carbon library principles, using light as a metaphor to 'illuminate' AI-generated content," applying "light-inspired effects like brightness, glow, and gradients to emphasize AI instances across the experience" so that "AI-generated or AI-recommended content" is made "distinctive and stands out for users." The guideline also sets an explicit restraint on that styling: integration "requires you to minimize distractions and show explanations only when needed or requested by the user."

The mechanism is layered disclosure through one component rather than scattered, ad hoc cues. The AI label is described as "the primary indicator to communicate that AI is present, and it's how users can access explainability content" — a single recognizable marker "identifiable across any IBM product" that doubles as "the trigger for the explainability popover which serves as the first layer of explainability." That popover gives "a consistent, up-front way to access explainability with the option to dig into more details if needed," so the label functions as both an announcement (AI is involved here) and an entry point into successively deeper explanation, rather than exposing full explainability content by default.

Against the library's other human-AI interaction guideline, How should AI systems be designed across different interaction phases?, this is a narrower and more concrete artifact: Amershi and colleagues distill behavioral recommendations organized by when in an interaction they apply, while Carbon for AI specifies a single visual-and-interaction component (consistent styling plus one label-and-popover pattern) meant to operationalize the make clear what the system can do and make clear why the system did what it did class of guidelines as a shippable UI element. The pairing matters more against Does explaining AI decisions actually reduce user overreliance on it?: that review finds XAI and transparency mechanisms frequently fail to curb automation bias... and can instead deepen it, particularly when explanations are available but not actively engaged with. Carbon for AI's popover is exactly the kind of up-front... access to explainability that review warns can function as a trust heuristic rather than a verification prompt — the style guide specifies the disclosure mechanism's shape but not whether users engage with it critically once they open it.

The excerpt is a design-system style guide, not an evaluation: it asserts that consistent AI labeling and layered explainability build trust and support recognition, but it reports no usability testing, adoption data, or outcome measurement for the AI label component, and several referenced details (the "Design for AI: Style elements," "Explainability," and "Chat pattern" pages) are marked "For IBMers only" and are not part of the excerpt. The implication the excerpt supports is narrower than this pattern improves trust or explainability outcomes — it is that IBM standardized the trigger and visual grammar for AI disclosure across its products, leaving open, per the automation-bias review above, whether a popover accessed "when needed or requested" actually prompts the critical engagement that distinguishes effective disclosure from a reassurance cue.