Help articles are written to be read start to finish. An assistant retrieves fragments, so an answer buried in the middle of a long article is often missed entirely.
The content is not wrong. It is structured for a different consumption pattern, and retrieval rewards a different shape.
One question per section, answered in the first sentence, with context after. Explicit headings that state the question rather than a clever title.
This also produces better human documentation, which is a rare case where optimising for a machine improves the thing for people.
Yes, and results improve markedly if articles are restructured one-question-per-section.
Often two articles contradict and it retrieved the older one. Pruning matters more than adding.
Yes. Undocumented limits are where confident wrong answers come from.
Half an hour with your own data usually saves reading three of these. The guides will still be here afterwards.