Why the order matters more than the length
INFERRED An audit that examines everything at once produces a list. One that examines things in dependency order produces a plan, because a finding in the first area changes the priority of everything after it. A disallowed retrieval agent makes every content recommendation premature. This page describes the shape of a credible audit so a Perth business can judge one before paying for it. The publisher of this site sells audits; that conflict is stated on the Perth cost page and applies here.
The five areas
1. Technical access
Whether search and AI agents can reach and read the pages: robots.txt per named agent, firewall and CDN rules, raw HTML compared with rendered HTML on each major template, index status, canonicals and internal links. DOCUMENTED Google’s own fundamentals list starts here, and it names the CDN layer explicitly. A credible audit fetches pages as the agents do rather than inferring status from one line of robots.txt. The crawler access guide is the self-check version.
2. Entity clarity
Whether the business resolves to one thing: an inventory of every public statement of its name, address, founding date, services and people, across the site, the registries, the Business Profile and directories, plus the structured data read against the visible page and every sameAs target followed. The consistency audit is the self-check version.
3. Content retrievability
Whether pages answer their questions in a form a system can lift: where the answer sits, how many words precede it, whether headings describe their sections, and whether the site covers the comparison, cost and suitability questions or only the pages that sell.
4. Evidence and corroboration
Whether the site publishes anything that does not already exist elsewhere, and whether independent sources agree with its facts. Original method, first-hand observation, real figures with sources; registries, associations, publications, reputable directories.
5. Prompt coverage
Which questions shape decisions in the category, and how assistants answer them today. A credible audit runs real prompts more than once per platform and reports the variance between runs, not a single screenshot. It records which answers name no business at all, because those are the open questions.
What an audit cannot tell you
- Whether any model will cite the business tomorrow. Prompt testing is a dated sample; a source appearing in three runs of five says something different from one appearing in every run, and neither is a forecast.
- What a competitor did to get where they are. Only what is observably different about them now.
- A traffic number. The relationship between visibility and sessions runs through user behaviour that varies by query type, as the traffic guide sets out.
- A share of category queries. The denominator is not published.
DOCUMENTED Google’s caveat applies to every recommendation an audit makes: meeting every requirement “doesn’t mean that Google will crawl, index, or serve its content”.
From findings to a plan
Findings are placed on two axes: how much they unblock, and how much effort they take. Structural blockers with low effort go first (a disallowed agent, a wrong canonical, a mismatched name), because they are hours of work and gate everything above them. Then clarification work: structured data, answer restructuring, relationship modelling. Then the slow, durable things: original evidence, coverage, corroboration. A plan that fits on a few pages is more useful than a ninety-page document; length is a poor proxy for value.
Questions to ask any provider
- What is the deliverable, and can I see a redacted example?
- Is crawler access verified per agent by fetching, or inferred from robots.txt?
- Is prompt testing included, how many runs per prompt, and is variance reported?
- What does the audit not determine?
- Which findings would you expect to change my priorities, and why?
INFERRED A provider who cannot name a limitation is either inexperienced or selling something. It is the cheapest filter available and it works on the first call.
What is documented and what is inferred
- Documented: Google’s fundamentals and its no-guarantee caveat.
- Inferred: the five-area structure, the dependency order, the prioritisation method.
- Conflict of interest: the publisher of this site sells audits. This page describes what any credible audit should contain, and says so.