Baudelaire
v0.0.18

Meta tags & social

Every page gets a description, Open Graph, Twitter Card and canonical tags in its <head>, drawn from its frontmatter and the site config.

html {
  meta #true
  jsonld #true
}

meta is on by default. jsonld is not.

What the site states

Two of these tags say something no page can. Both live in the meta block, which is the flag above with settings behind it:

html {
  meta {
    twitter "@example"
    image "/og.png"
  }
}
Key Default Does
twitter – The site’s account, as twitter:site. Without it a card credits whoever posted the link and nobody else.
image – The preview image for a page that names none and gets no generated card.

image is a floor, not an override: a page’s own image wins, and so does a generated card. It is resolved and made absolute the same way an authored one is, so /og.png becomes a full URL under the site’s url. It carries no og:image:alt, because one picture standing in for every page is not described by any one page’s title. A page that wants an alt names its own image and its own alt.

NOTE

typst-html owns the document <head>, so these tags can’t be written in a layout. baudelaire appends them to the parsed DOM instead, which is why they are config rather than markup.

What the tags read

#let frontmatter = (
  title: "Launch day",
  summary: "Everything new in this release.",
  image: "/assets/launch.png",
  alt: "The release notes, open in a browser",
  author: "Ada",
  date: datetime(year: 2026, month: 3, day: 1),
)
Key Falls back to Fills
description summary <meta name="description">, og:description, twitter:description.
image a generated card, then html { meta { image } } og:image, twitter:image.
alt the page title, for a generated card og:image:alt.
author the site-wide author <meta name="author">, article:author.
date – og:type, article:published_time.
updated – article:modified_time, emitted only when it moved.

alt describes the image for a reader who can’t see it. A generated card needs none: it draws the page title, so that’s what its og:image:alt says.

What gets emitted

Group Tags
Document description and author.
Open Graph og:type, og:title, og:description, og:url, og:site_name, og:locale, og:image, og:image:alt.
Article article:published_time, article:modified_time, article:author, one article:tag per taxonomy term. Dated pages only.
Twitter Card twitter:card, twitter:title, twitter:description, twitter:image, and twitter:site when the site names an account.
Link relations canonical, one alternate per translation plus x-default, one alternate per feed format, the page’s PDF when it has one, and the web app manifest when the site writes one.

A dated page is an article, everything else a website. Several unfurlers read article:published_time for the dateline. twitter:card is summary_large_image when the page has an image, summary when it doesn’t.

Only the tags whose data exists are emitted, so a minimal page still gets a clean, valid set.

URLs

og:url and the canonical link appear only when you set a base url in the config, the same precondition as feeds and the sitemap.

An image naming a local asset is resolved to its fingerprinted URL and then made absolute against url, so a crawler fetches the cache-busted file. An already-absolute http value is left as authored. See Images for what the pipeline does to the file itself.

Locale

og:locale comes from lang, in Open Graph’s spelling: fr-CA becomes fr_CA. A bare fr is passed through rather than given an invented territory, since fr_FR would be wrong for a Belgian site. Declare the region in lang to get the full form.

Structured data

html {
  jsonld #true
}

Adds a JSON-LD island to every page’s <head>: an Article where the page is dated, a WebPage otherwise, with headline, description, image, url, datePublished, dateModified, author and keywords from the taxonomy terms.

It’s built from the same facts as the meta tags, so the two can never claim different things about one page.

NOTE

Unlike its neighbors this is off by default. The meta tags restate what the page already says; structured data is a claim made to a search engine about what the page is, and that’s yours to make.