What are the technical checks?
The first check is crawlability without JavaScript: fetch key pages as a crawler that executes no scripts would, and confirm the main content, headings, and JSON-LD are present in the initial HTML. The pass condition is that everything an engine needs is server-rendered. On ftes.ai this is a build requirement.
The second is schema: Organization on the root, Article or BlogPosting with author and dates on content pages, FAQPage only where a visible FAQ exists, BreadcrumbList, and Product or Service where relevant, all validating in the engines' testing tools and describing only what is on the page. The third is llms.txt: present at the root, current, and written as an honest map of the site's important pages.
- 1. Content and schema present without JavaScript
- 2. Schema honest, complete, and validating
- 3. llms.txt present, current, and useful
What are the structure checks?
The fourth check is answer-first: each key page opens with a direct 40 to 60 word answer to its primary question, starting with a definition or verdict, not with an introduction. The fifth is chunk self-containment: read a sample of sections with the rest of the page hidden and confirm each names its entity, answers a question, and contains no cross-references.
Both are judged page by page. A site that passes the technical checks and fails these is findable and not citable, which is the most common audit result.
- 4. Key pages open with a direct 40 to 60 word answer
- 5. Sampled sections stand alone: entity named, question answered, no cross-references
What are the trust checks?
The sixth check is entity consistency: the brand, products, and key concepts have one name each, used identically across pages, with a canonical definition on the site and the same facts everywhere, including listings you control. The seventh is verifiability: sample the claims on key pages and confirm each is a product mechanism, shown arithmetic, or a named public source, with no invented statistics, studies, customers, or quotes.
A fail on the seventh is the most serious finding in the audit. Unverifiable claims do not just fail to earn citation; they mark the page as a risk and lower trust in the domain.
- 6. Entities named consistently with canonical definitions
- 7. Every sampled claim verifiable; zero invented figures
What are the coverage and measurement checks?
The eighth check is fan-out coverage: for each primary question the site wants to be cited for, list the five to nine likely sub-queries and confirm each is answered in a self-contained section or FAQ somewhere in the cluster. Gaps here are the content plan.
The ninth is a measurement baseline: a fixed prompt set exists, has been sampled across the relevant engines at least once with results recorded and dated, and a schedule is set to repeat it. Without this, no later change can be judged, and the audit has no way to close its own loop.
- 8. Sub-queries listed and each answered somewhere in the cluster
- 9. Prompt set sampled, recorded, dated, and scheduled
How do you act on the results?
Act on the results in order. Technical fails are fast and unblock everything else. Structure fails are fixed page by page, starting with the pages that already rank for the target sub-queries, because those are closest to citation. Trust fails are fixed before anything else is published, because publishing more onto an untrustworthy domain compounds the problem.
Coverage gaps become the content plan, and the measurement baseline becomes the way the plan is judged. A Search Visibility Digital FTE on ftes.ai runs this loop as its function: audit, propose with a worth-it verdict, pass the fact gate, publish on approval, measure, and report what moved and what did not.
What does a completed audit look like for a real site?
As a labelled example of the output for one site: checks one to three pass, with the note that the schema validated and the llms.txt was last updated on a stated date. Check four fails on a third of the key pages, listed, because their openings are introductions rather than answers. Check five fails on sampled sections across the same pages, with the cross-references quoted. Check six passes for the brand and fails for one product with two names in use, both listed. Check seven passes, with the sampled claims and their sources recorded. Check eight lists the sub-queries of the three primary questions and marks nine of them uncovered. Check nine fails because no prompt set has been sampled yet.
The action plan follows the order: the prompt set is written and sampled first so the baseline exists, the product's name is unified, the failing openings and sections are rewritten on the pages that already rank, and the nine uncovered sub-queries become the content plan. The whole audit is a document of a few pages, dated, and the Search Visibility FTE repeats it each quarter against the previous one, so the site's progress is a record rather than a feeling.