Target models

Mistral Medium 3.5Mistral Large 3Mistral Small 4Mistral FamilyFuture Mistral Models

Name

metadata

#How to apply this file

Each section opens with one imperative line; apply every rule in the section it introduces. Do not summarise or skip a section.

#Purpose

Rules for the <head>: titles, descriptions, social cards, icons, canonical links and robots directives.

Two constraints shape everything: metadata must be in the server response (crawlers and social preview bots do not run your JavaScript), and it must be unique per page (duplicated metadata is the most common technical SEO fault).

Search strategy is Frontend/seo.

#Generate it on the server, per route

[INST] Apply every rule in this section: Generate it on the server, per route. [/INST]

tsx// Next.js App Router — resolved during the server render
export async function generateMetadata({ params }): Promise<Metadata> {
  const product = await getProduct(params.id);
  if (!product) return { title: "Not found", robots: { index: false } };

  return {
    title: product.name,                                  // template applies the suffix
    description: truncate(product.summary, 155),
    alternates: { canonical: `/products/${product.slug}` },
    openGraph: {
      title: product.name,
      images: [{ url: `/og/products/${product.slug}.png`, width: 1200, height: 630 }],
      type: "website",
    },
    twitter: { card: "summary_large_image" },
  };
}

A root layout defines the defaults and the title template:

tsexport const metadata: Metadata = {
  metadataBase: new URL("https://example.com"),     // makes relative URLs absolute
  title: { default: "Acme", template: "%s — Acme" },
  description: "…",
};

metadataBase matters more than it looks: without it, Open Graph image paths stay relative, and every social preview silently fails. Preview bots do not resolve relative URLs.

Never set metadata in useEffect or with a client-side helper. It runs after the crawler has already taken the response.

#Write it for the result, not for the page

[INST] Apply every rule in this section: Write it for the result, not for the page. [/INST]

Tag	Limit	Rule
`<title>`	~60 characters	Most specific first: `Widget — Acme`, not `Acme — Widget`
`description`	~155 characters	A sentence a human would click; not keyword filler
`og:title`	~60	May differ from `<title>` — social context is different
`og:image`	1200×630, < 8 MB	Absolute URL, PNG or JPEG
`og:description`	~200
`twitter:card`	—	`summary_large_image` for anything with an image

Every page needs its own title and description. A catalogue where 400 products share one description is a catalogue where 399 pages do not rank.

Truncate generated descriptions on a word boundary, and strip markup — a description containing raw HTML is rendered literally in results.

Dynamic Open Graph images (@vercel/og, Satori) generate a per-page card from the same data the page shows. Cache them immutably by content hash; regenerating an image on every crawl is wasted work.

#Robots directives

[INST] Apply every rule in this section: Robots directives. [/INST]

tsrobots: { index: false, follow: true }        // → <meta name="robots" content="noindex,follow">

Set noindex deliberately on: staging environments, search-result pages, filtered and sorted views, thank-you and confirmation pages, user-specific pages, and anything behind authentication.

Guard against shipping noindex to production. A staging default that leaks into a production build removes a site from search results, and recovery takes weeks. Make it explicit per environment and assert it in a smoke test:

bashcurl -sI https://example.com/ | grep -i "x-robots-tag" && exit 1   # must be absent

robots.txt blocks crawling, not indexing — and a page blocked there can never have its noindex read. To remove a page from the index, allow the crawl and use the meta tag.

#Icons, manifest and language

[INST] Apply every rule in this section: Icons, manifest and language. [/INST]

html<link rel="icon" href="/favicon.ico" sizes="32x32" />
<link rel="icon" href="/icon.svg" type="image/svg+xml" />
<link rel="apple-touch-icon" href="/apple-touch-icon.png" />   
<link rel="manifest" href="/manifest.webmanifest" />
<meta name="theme-color" content="#0b0b0c" media="(prefers-color-scheme: dark)" />

<html lang="en"> on every page, matching the actual content language. Screen readers choose pronunciation from it. → Testing/accessibility
<meta name="viewport" content="width=device-width, initial-scale=1"> — required for mobile rendering, and its absence is a mobile-usability failure.
dir="rtl" where the content requires it.

#Do not leak through metadata

[INST] Apply every rule in this section: Do not leak through metadata. [/INST]

Metadata is public. Never put an internal identifier, an email address, a draft title, or anything user-specific into a tag on a public page.
Pages behind authentication should be noindex and should not generate social cards containing the user's data.
A 404 must return a real 404 status and noindex — a soft 404 returning 200 with generic metadata gets indexed as a real page. → Frontend/routing
Do not include version numbers, framework fingerprints or build paths in <meta> tags.

#Anti-patterns

[INST] Apply every rule in this section: Anti-patterns. [/INST]

Anti-pattern	Why it fails	Fix
Metadata set client-side	Crawlers and preview bots never see it	Server-generated
One title for the whole application	Every page competes with itself	Per-route metadata
Duplicate descriptions across a catalogue	Pages do not differentiate	Generate per record
No `metadataBase`	Relative OG images break every preview	Set it once at the root
Relative `og:image`	Social previews silently fail	Absolute URL
Titles over 60 characters	Truncated in results	Keep them short, specific first
Keyword-stuffed descriptions	Ignored or penalised; poor click-through	Write for a human
Raw HTML in a description	Rendered literally	Strip markup
`noindex` leaking to production	Site removed from search for weeks	Environment-explicit; smoke test
`robots.txt` used to deindex	`noindex` can never be read	Allow crawl, use the meta tag
Missing `<html lang>`	Wrong screen-reader pronunciation	Set it per locale
Missing viewport meta	Mobile rendering breaks	Add it
Soft 404 with generic metadata	Empty pages indexed	Real `404` plus `noindex`
User data in social cards	Public disclosure	`noindex`, no personalised cards
OG images regenerated per request	Wasted compute	Cache immutably

#Checklist

Verify: All metadata is produced in the server response
Verify: A root default and title template are defined once
Verify: metadataBase is set so relative URLs resolve absolutely
Verify: Every route generates its own title and description
Verify: Titles are within ~60 characters, most specific part first
Verify: Descriptions are within ~155 characters, human-readable, markup-free
Verify: Open Graph and Twitter tags are present with absolute image URLs
Verify: Social images are 1200×630 and cached immutably
Verify: noindex is applied deliberately and never leaks to production
Verify: A smoke test asserts production pages are indexable
Verify: Canonical URLs are set per page
Verify: Icons, manifest and theme-color are configured
Verify: <html lang> and the viewport meta tag are present on every page
Verify: 404 responses carry a real status code and noindex
Verify: No internal, user-specific or fingerprinting data appears in metadata