A person runs every audit, and this is the person.

Why this exists

Advice about AI answer engines grew faster than the evidence for it. Business owners are told to "do AEO", quoted a monthly retainer, and given a score out of a hundred that nobody explains.

Some conditions on a website make it easier for a machine to retrieve, understand, attribute and quote it: a stable page, a claim that says where its number came from, an author who exists, a passage that still makes sense when lifted out. Those conditions can be inspected. Whether that produces a citation is a separate question nobody can honestly answer for you.

So this product measures the conditions and makes no promise about citations. The rubric, the checks, the refund rule and the limits are all published.

Chris Brock

Chris has worked on the operations side of technology: security and compliance programs, technology integration after acquisitions, ecommerce platform consolidation, and making systems produce evidence that holds up under audit.

That is why every finding here names the URL it came from, when it was observed, and the captured response it was read from.

He wrote the rubric, wrote the tool, and signs the reports. Because the instrument and the operator are the same person, an audit from Cite Unseen is not an independent second opinion. It is a documented first one, detailed enough that you can disagree with it specifically.

How the work is done

A bounded crawler reads public pages, records every response as evidence with a hash, and runs the checks that can be settled by reading bytes: whether a canonical is declared, whether structured data parses, whether a heading has an id. Those checks set their own rating.

Checks that need a reader (does a passage answer the question it sits under, is a claim material, does a credential mean what it appears to) collect the same evidence and stop. They stay inconclusive until a person reads the evidence and signs a rating, with the reason recorded.

A check that cannot be settled either way stays inconclusive and is left out of the score, not counted against the site. A tooling failure is never scored as a customer failure.

The standards the checks use

The rubric is ours. The specifications it checks against, and that a finding will point you to, are these:

• Schema.org, the vocabulary the structured-data checks validate against, including which types are subtypes of which. We expect it serialized as JSON-LD 1.1.
• RFC 9309, the Robots Exclusion Protocol. Our crawler obeys it and the crawler-directive checks read it.
• The Sitemaps protocol, which defines what the sitemap-quality check looks for.
• WCAG 2.2, the source of the contrast and structure expectations. Answer engines and screen readers want many of the same things from a page.
• The HTML Living Standard, for what a heading, a canonical link and a language attribute mean.

llms.txt and the ai.txt proposals are voluntary and unratified. We check whether a site serves them and say that no engine is known to require either.

When we are wrong

The sample report on this site carries a correction notice: a parser defect made a check report a fault the site did not have, and the score changed when it was fixed. The notice stays on the page.

If you show us a finding that is wrong, the audit is refunded in full and the check is corrected for the next site.

Contact

The request form reaches a person directly and costs nothing to use. Or write:

• hello@citeunseen.io for an audit, an invoice, or a finding you think is wrong.
• privacy@citeunseen.io for data questions and deletion requests. See what we collect.
• security@citeunseen.io for anything security-relevant, including our crawler's behaviour against your site. Please write before disclosing publicly.

Chris Brock LLC dba Cite Unseen · Atlanta, Georgia, United States