Short answer: audit in gate order — reachable, readable, extractable, corroborated. Most audits in this category start at the fourth gate, which is why they produce long reports and no movement.
Gate 1 — Can each named agent reach the page?
Not "does robots.txt allow it" — request the page as each agent and compare status codes and served bytes. A CDN rule or a WAF that blocks by user agent will fail this while robots.txt reads as permissive. Check GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Claude-SearchBot, Applebot at minimum, and note that a failure to fetch is not the same as a page not existing.
Gate 2 — Is the content there without JavaScript?
Fetch as plain HTML with JS off. If the substance appears only after hydration, most retrieval agents see an empty shell. This gate catches more failures than any other on modern stacks, and it is invisible in a browser.
Gate 3 — Can an answer be lifted out?
Pick the heading that matches a real question. Read the first two sentences under it, out of context. If they are a complete, correct answer that names its own subject, the page passes. If you had to scroll up to understand them, it fails — and link building will not fix that.
While you are there: are facts in tables with units and dates, or scattered through prose? Is anything important behind an accordion or tab? Does the page hedge where a competitor states a number?
Gate 4 — Is the claim corroborated anywhere you don't own?
List the claims you want an engine to repeat about you — who you're for, what you cost, what you cover. For each one, find a source that is not your own domain. Where you cannot, that is the gap, and it is the slowest one to close.
Then measure, and keep the baseline
Preserve the original baseline before you change anything, and rerun the same question set after. An audit without a before-and-after is an opinion with a checklist attached.
What an audit cannot tell you
Whether the questions you are optimising for are the ones buyers ask. That comes from your own Search Console queries and from real prompt data, not from a page-level audit — and it is the most common reason a technically perfect site stays absent.