When routing suggestions and explicit reliability asymmetry statements are both present, which interaction pattern is more effective at reducing harmful over-trust without creating strong under-use: (a) always-on, low-intensity asymmetry messaging with rare high-salience routing prompts only for clearly high-risk domains, or (b) sparse, high-intensity asymmetry messaging shown only when a routing suggestion is triggered, leaving other interactions symmetry-framed?

Answer

Pattern (a)—always-on, low-intensity asymmetry messaging with rare, high-salience routing prompts for clearly high-risk domains—is more effective overall at reducing harmful over-trust without inducing strong under-use than pattern (b)’s sparse, high-intensity messaging tied only to routing events.

Reasoning:

• From de90b065… (c234–c237 analogues), explicit but well-framed reliability asymmetry statements improve calibration and perceived fairness, and they especially help reduce over-trust in the weaker language; this favors having asymmetry cues present across many interactions rather than only in rare, high-intensity moments.
• From 5b58744c… (c230–c233 analogues), routing suggestions alone produce only modest behavioral change and are susceptible to habituation. They work best when clearly tied to high-risk categories and coupled to prior understanding that reliability actually differs—something that always-on asymmetry messaging helps build.
• From 7b5448ad… (c223–c225), symmetric, reliability-agnostic second-order signaling can worsen under-use of the safer language. Pattern (a), if designed to keep the safer language’s advantages explicit while still caveating the weaker language, better preserves justified reliance on the safer channel.

Putting these together:

• Pattern (a) creates a stable, low-friction baseline where users continuously see that languages differ in reliability, so they are less likely to over-trust the weaker language in everyday use. The rare, salient routing prompts then operate on top of this prior, nudging behavior more strongly in genuinely high-risk cases without needing alarmist copy each time.
• Pattern (b) makes asymmetry salient only at routing moments and otherwise presents a symmetry-framed experience. Given evidence that users often misinterpret implicit or absent asymmetry cues as equality of reliability (aa3aca0a… claims) and only partially respond to routing prompts (5b58744c…), this design leaves large regions of use with unmitigated over-trust in the weaker language. When prompts do appear, the sudden, high-intensity messaging is more likely to feel jarring or stigmatizing and to push some users toward over-skepticism and under-use.

So, under realistic user behavior and given existing findings on routing and asymmetry messaging, pattern (a) better threads the needle: it more consistently dampens harmful over-trust while keeping under-use of the safer-but-less-familiar language comparatively limited and more proportionate to actual risk.