Do FAQ pages earn their place, and what keeps them from rotting?

Open question

Two published positions contradict each other. The GOV.UK style guide (cited) says not to use FAQs on GOV.UK, because content written by starting with user needs will not need them. Nielsen Norman Group (cited) argues that FAQs, done well, deliver value to visitors and to the organisation, that search is rarely enough as a substitute because users' vocabulary differs from the site's, and that an FAQ should be part of a knowledge-management process fed by real questions. Both positions are stated for websites in general; neither reports outcomes for technical documentation of software, where the alternative to an FAQ is usually a how-to guide or a reference entry, and where the readers now include agents extracting answers.

Sub-questions:

• Outcome: has any team measured support-ticket volume, search-exit rate or time-to-answer before and after adding, restructuring or removing an FAQ page, and what did the numbers show?
• Placement: when the same answer exists in a topic page and in an FAQ, which copy do readers find, and which one goes stale first? Do teams that keep FAQs as pointers only (question plus link to the canonical section) see less drift than teams that duplicate the answer?
• Sourcing: which teams generate the question list from real support requests, search logs or issue trackers, at what cadence, and how do they retire questions that stop being asked?
• Ownership: what ownership and review rules have kept an FAQ accurate over years, and what happened to FAQs without an owner?
• Agents: do question-shaped headings improve or worsen extraction of correct answers by language-model agents compared with task-shaped headings, for the same content?

What a useful answer contains

The kind of product and documentation set; the size of the FAQ and how questions were sourced; the metric and its baseline; how staleness was detected (dates, link checks, reader reports); and whether the FAQ duplicated or pointed to canonical content. Opinions restating either cited position without new observations do not answer the question; single anecdotes should say so.