In chat-native product discovery flows where the agent can re-run the same comparison table under alternative, labeled “freshness vs. relevance” policies and show a compact before/after diff (e.g., which rows moved or disappeared), how does repeated exposure to these contrastive re-rankings affect users’ decision confidence, coverage confidence, and over-trust in any single ranking—especially once the novelty of the feature wears off?

Answer

Repeated, clearly labeled freshness-vs-relevance re-rankings with compact diffs tend to (a) raise decision and coverage confidence a bit by making the contingency of rankings visible, (b) reduce blind over-trust in any single ranking but (c) create a new anchor on the family of views the agent surfaces, especially after novelty fades.

Net effects (steady state, after novelty)

• Decision confidence: usually slightly higher; users feel they’ve “kicked the tires” by seeing at least two policies. Gains plateau once users internalize that policies mostly reshuffle the same top cluster.
• Coverage confidence: often increases more than warranted; seeing a few contrastive views makes many users feel they’ve explored the space, even if all policies operate on a narrowed candidate set.
• Over-trust: shifts from “this ranking is the truth” to “these 2–3 views bracket the truth.” Over-trust in any one view drops; over-trust in the system’s option set and its policies rises.

Key dependencies

• Label clarity: short, concrete labels (“favor newer reviews” vs “ignore age unless very old”) and row-level move/disappear badges help users grasp the trade-off and maintain some skepticism.
• Diff design: compact, local diffs (up/down badges, “dropped due to being older”) support calibrated trust; large, unexplained jumps can erode confidence.
• Candidate pool: if all policies draw from a hidden, pruned pool, contrastive views risk overstating coverage and quietly reinforce trust in that pool.

Directional summary

• Slight + on decision confidence.
• Mixed: modestly improved calibration about policies, but residual over-trust that the agent’s explored space is complete.
• Strongest benefit for engaged users who try at least one alternative policy; light users may see one re-rank once, then treat it as a fixed “best mix.”