What this check tests
Only pages that have at least one parseable JSON-LD block are looked at (see Schema.org JSON-LD present). Pages with none are left out of this average, so a site with markup on one page can score well here and badly on the presence check.
For each such page the audit takes the first block only and checks whether these seven top-level keys exist: name, description, author, datePublished, dateModified, image and url. A key counts if it is present with any value, including an empty string. Nothing is checked about the value. The page's score is the fields present divided by seven; the check averages that over the pages with markup and rounds.
- Pass: the average is 70 or more. That means five of seven fields on a typical page (5 of 7 is 71).
- Partial: above 0 but under 70.
- Fail: 0. This includes the case where no page has JSON-LD, with the message "No Schema.org data to evaluate".
The list is the same for every type. Two things follow. Fields are read only at the top level of the first block, so a block that wraps its nodes in @graph scores 0 for that page. And the list mixes fields from different schema.org types: name, description, image and url are defined on Thing, while author, datePublished and dateModified come from CreativeWork. The list is oriented to articles and pages, so blocks of other types such as Organization or Product typically fill fewer of the seven. That is not an enforced cap: the audit counts the listed keys on any type, so a single block that has five of them passes. Article-like pages are what this check is built for.
Why it matters to agents
Descriptive fields are what a parser gets out of a block. A name, a description, an author and a date let a tool say what a page is, who wrote it and how fresh it is without reading the whole body. That is general good practice for structured data.
Google's guidelines make a related but different point: items that are missing required properties are not eligible for rich results, and the more recommended properties you give, the higher quality the result is for users. Required and recommended properties are set per feature in Google's documentation, not by the fixed list used here. Google also says its AI features need no special schema.org markup, so do not expect a direct effect on chat answers from raising this number.
How to fix it
- On article-like pages (posts, guides, docs), make the first block carry all seven keys: a
headlineorname, adescription, anauthorobject,datePublished,dateModified, animageURL and the pageurl. Note that this check wantsname, notheadline; schema.org has both, so includenameas well. - Put the page's main block first in the HTML. Only the first block is read, so a small
WebSiteorBreadcrumbListblock placed above it will be the one that is scored. - Fill the values from your CMS fields, not by hand, so dates stay current. Update
dateModifiedwhen the page changes. - For
Organization,Productand similar pages, expect a partial. Fill the four fields that exist for the type and do not invent anauthoror a publication date. Schema.org definesauthoranddatePublishedon CreativeWork, so adding them to a company profile would be false markup. - Keep the values true to the visible page. Google's guidelines say not to mark up content that readers cannot see.
Choosing types is covered by Appropriate Schema.org type, and whether @context and @type are well formed by JSON-LD valid.