The rule. Show the shortest useful explanation inside the response itself, and keep the sources, the method, and the caveats one deliberate click away.
Why. The guideline is about access, not volume: the eleventh human-AI interaction guideline reads "Make clear why the system did what it did. Enable the user to access an explanation of why the AI system behaved as it did" (Amershi et al., 2019), and the HAX Toolkit expands it into seven interchangeable explanation patterns, from local and global explanations to example based and "what if" ones, with the instruction to "mix and match explanation patterns as needed" (Microsoft, n.d.). Layering is the mechanism that makes that choice tractable: progressive disclosure means you "initially, show users only a few of the most important options" and "offer a larger set of specialized options upon request," which Nielsen credits with improving "3 of usability's 5 components: learnability, efficiency of use, and error rate" (Nielsen, 2006). The catch is that the first layer has to survive alone, because the deeper ones go mostly unread: Nielsen Norman Group's 2025 study of explainability in chat interfaces reports that people "rarely click citation links" and that while users could verify each source, "they seldom do so in practice" (Chan, 2025).
Seen in the wild. Microsoft Copilot cites its sources inline in the answer and adds a "See all" button that opens the complete reference list, so provenance is glanceable first and exhaustive on request (Chan, 2025).