Start from the real page.
Answer engines and auditors both have to get the page first. Status codes, JavaScript, robots rules, and template chrome decide whether the body a person sees is actually there to read. If that first pass is thin or wrong, every finding after it is a guess.
A technical scrape gets you the real page: rendered when it has to be, raw HTML when it doesn't, with its fields and links attached. Get that right and your audits, maps, and agent answers all stand on solid ground.