In practice
Start from customer language — support tickets, sales calls, reviews — then validate volume in a tool rather than the other way round.
Group results into clusters by intent before writing anything. The output of research is a page plan, not a spreadsheet of terms.
Why it matters
It decides what gets written, so a weak research phase wastes every hour of production that follows.
It is also the cheapest way to find demand nobody in the category is serving yet.