What this check tests
The audit reads the served HTML of every crawled page (the homepage plus a sample of same-site pages; the standard profile caps the crawl at 10). It collects every <meta> element whose property attribute starts with og: or article:. Then it looks for three tags: og:title, og:description and og:image. A tag counts when its content is not empty.
Each page scores the tags present out of three: 0, 33, 67 or 100. The check averages that over all crawled pages, with no page left out, and rounds.
- Pass: the average is 80 or more. A site where every page has two of the three tags scores 67 and does not pass.
- Partial: above 0 but under 80.
- Fail: 0, meaning no crawled page has any of the three tags.
Details that change results:
- Only the
propertyattribute is read. A tag written as<meta name="og:title" ...>is not counted, although some sites use it. The Open Graph protocol specifiesproperty. - If the same property appears twice on a page, the later one wins, so a second empty
og:imagecancels the first. - Tags added by JavaScript in the browser are not seen. The audit reads the HTML the server returns.
- The check does not fetch the image, check that the URL is absolute or check the value's quality.
og:typeandog:urlare not tested. The protocol lists four required properties:og:title,og:type,og:imageandog:url. It listsog:descriptionas optional but generally recommended. So this check's set overlaps the protocol's required set but is not the same.- Twitter or X card tags (
twitter:*) are not tested.
Why it matters to agents
The protocol was designed so a page can become a rich object in a social graph, with a title, a type, an image and a URL. Its main use today is link previews. The protocol says nothing about AI assistants, and this page cites no AI vendor or search engine document saying they read Open Graph tags, so do not expect a direct effect on how an assistant cites you.
The indirect argument is modest. These tags put a short title, a one or two sentence description and a representative image in one predictable place in the <head>. Any tool that fetches raw HTML, whether a link unfurler, a chat client showing a preview or an agent building a link list, can read them without parsing the body. If you want a page to look right when someone pastes the link into a chat or social app, the tags are what those previews use. Weight is 1.
How to fix it
Put these in the <head> of every page template, rendered on the server:
<meta property="og:title" content="Team plan pricing" />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://example.com/pricing" />
<meta property="og:image" content="https://example.com/images/pricing-card.png" />
<meta property="og:description" content="What the Team plan includes and how seats are billed." />
- Write a different title and description for each page. Reusing the homepage values everywhere passes the check but tells a reader nothing.
- Use an absolute
httpsURL forog:image. The protocol describes it as an image URL that represents the object. - Add
og:typeandog:urleven though this check ignores them, since the protocol requires them. - Make sure your framework's head helper emits
property, notname, and that it runs on the server. View the page source, not the inspector, to confirm. - If you set defaults in a layout, check that a page-level tag overrides the default instead of adding a second one.
Keep og:url equal to the page's canonical URL. For structured data about the same page, see Schema.org JSON-LD present.