The whole point of sycophancy is that it’s invisible to the person on the receiving end … you don’t request a skeptical mode when you don’t realize you need one.
So what if certain kinds of skepticism were the default? Ask “what evidence supports this claim?” before endorsing policy or factual assertions. Generate one strong counterargument for non-trivial claims. Mark confidence bands clearly, e.g., low, medium, high, rather than presenting everything with the same tone of assurance.
In reasoning-heavy contexts, these defaults improve epistemic quality even if they introduce mild friction (Interface friction and critical thinking).
Valid claims can get over-questioned, which creates false negatives, and users may perceive neutral skepticism as hostility. There is also latency and cost overhead from the extra reasoning steps.