Pages that look finished to a visitor and score badly for the machines that decide whether you get recommended. A person scrolling on a phone sees a complete layout; an AI crawler downloads the HTML and reads something else — missing metadata, a heading stuffed with CSS, a canonical that points at the wrong URL, or a shell that only fills in after JavaScript runs.
Open any example below, then paste the URL into Site Audit. Each page is a known defect, so the score and the findings should match the description on the card. These URLs stay out of the search index on purpose; they exist so the rubric can be checked against real markup instead of a screenshot.
Featured
Our own homepage, built so an AI crawler can barely read it. A visitor sees the finished page; the HTML a crawler downloads is a shell with the defects Site Audit is meant to catch.
Open the main page counter-exampleEach of these is a single, known defect. They stay out of the search index on purpose — Site Audit fetches the URL directly, so you can see the finding without teaching Google a broken page. Use them the way a test fixture is used: one URL, one expected code, no ambiguity about whether the rubric drifted. The control pages that must stay clean live under the same fixtures directory; they are not listed here because they are not counter-examples.
HEADING_CONTAINS_STYLE, and the H1 extracts as its actual text
/example/fixtures/style-in-headingTITLE_NOT_IN_HEAD, META_DESCRIPTION_NOT_IN_HEAD, CANONICAL_NOT_IN_HEAD
/example/fixtures/metadata-in-bodyNOINDEX_DETECTED, naming the directive that caused it
/example/fixtures/robots-noneNOINDEX_DETECTED
/example/fixtures/googlebot-noindexCANONICAL_POINTS_ELSEWHERE, showing the target URL
/example/fixtures/canonical-elsewhereINVALID_JSON_LD, which is a different fix from adding markup
/example/fixtures/malformed-json-ld