Canonical Tag for AI Search

TL;DR

Declare a canonical URL on every page using either an HTML or an HTTP Link header. Use absolute URLs, self-canonicalize each variant, and never let parameterized, AMP, or mobile variants point to mismatched targets. AI engines respect canonical signals when picking which URL to cite, but only when the signals are consistent across HTML, sitemap, and HTTP layers.

Definition

rel="canonical" is a link relation defined by the HTML specification and documented by Google Search Central. It tells crawlers that the linked URL is the preferred representation of the current page's content. Crawlers consolidate ranking, indexing, and citation signals onto the canonical URL.

A canonical signal is a hint, not a directive. Search engines and AI crawlers may override the declared canonical when other signals (sitemaps, internal links, HTTP redirects) strongly disagree.

How AI Engines Use Canonical Signals

AI engines see many copies of the same content: with and without query parameters, with www and without, on http and https, on the main domain and on AMP, on syndicated partner domains. Without canonical consolidation, citation share fragments across variants and the publisher receives less attribution per query.

Google AI Overviews inherit Google Search canonicalization. Perplexity, ChatGPT, and Claude consult canonical signals during ingestion and when generating citations: when a page is reachable through multiple URLs, they typically cite the canonical. When canonicals disagree across delivery layers (HTML says one URL, HTTP header says another), AI engines may fall back to the URL they fetched, which is rarely the intended canonical.

Delivery Methods

HTML Element

<link rel="canonical" href="https://example.com/articles/canonical-tag" />

Place inside . Use absolute URLs only; relative URLs work but are more error-prone behind proxies.

HTTP Link Header

For non-HTML resources (PDFs, images, JSON):

Link: ; rel="canonical"

NGINX Example

location ~ \.pdf$ {
  add_header Link '<https://example.com${request_uri}>; rel="canonical"';
}

Cloudflare Workers Example

addEventListener('fetch', event => {
  event.respondWith(handle(event.request));
});
async function handle(request) {
  const response = await fetch(request);
  const newHeaders = new Headers(response.headers);
  newHeaders.set('Link', <${request.url}>; rel="canonical");
  return new Response(response.body, {
    status: response.status,
    headers: newHeaders,
  });
}

AI-Specific Edge Cases

Parameterized URLs

Query parameters (?utm_source=..., ?ref=..., sort and filter parameters) create distinct URLs that crawlers may fetch separately. Canonicalize all parameterized variants to the parameter-free URL. AI engines that follow campaign links back to your site treat the canonical URL as the source of truth.

www vs Non-www

Choose one host and 301-redirect the other. Both versions should declare the same canonical. Inconsistency causes AI engines to fragment citation share between www.example.com and example.com.

Trailing Slash

Decide once: with or without. Apply consistently. The canonical must match the served URL exactly (byte-identical), or some parsers treat it as a soft mismatch and ignore the signal.

AMP

The AMP page's canonical should point to the non-AMP page. The non-AMP page declares to the AMP variant. AI engines that ingest AMP follow the canonical back to the rich page when citing.

Mobile Variants on Separate URLs

If you serve m.example.com separately, the mobile page canonicalizes to the desktop URL, and the desktop page declares rel="alternate" with media to the mobile URL. Responsive designs avoid this by serving one URL.

HTTPS

Never canonicalize from HTTPS to HTTP. Always upgrade to HTTPS. AI engines deprioritize and may suppress citations from URLs that downgrade.

Required Properties

Pre-Publish QA Checklist

1. Every page declares exactly one canonical.
2. Canonical URL is absolute and uses HTTPS.
3. Canonical URL is reachable with 200 OK (not 301, 404, or 5xx).
4. Canonical URL appears in the sitemap.
5. Canonical URL matches the served URL byte-for-byte (case, trailing slash, parameters).
6. Internal links to the page use the canonical URL.
7. HTML and HTTP Link header canonicals agree if both are present.

Debugging Mismatched Canonicals

Use the URL Inspection tool in Search Console for Google's view of canonicalization. For AI engines, fetch the page with each AI bot's User-Agent and inspect the response headers and HTML. Common mismatches:

• HTTP redirect chain that ends at a different URL than the declared canonical.
• Trailing slash mismatch between declared canonical and served URL.
• Sitemap listing a non-canonical URL.
• JavaScript that rewrites the canonical client-side; many AI bots do not execute JavaScript.

Common Mistakes

• Canonicalizing all variants of an international site to the English version, which collapses hreflang signals.
• Canonicalizing HTTPS pages to HTTP versions.
• Using relative URLs in canonical tags, which break behind proxies.
• Conflicting HTML and HTTP-header canonicals.
• Setting canonical on a page that should be noindex instead.

FAQ

Q: Do AI engines treat canonical as a directive?

No. Canonical is a signal. AI engines consider it alongside redirects, sitemaps, and internal links, and may override when signals disagree. Make all signals consistent.

Q: Should canonical URLs include query parameters?

Usually no. Strip tracking and session parameters from canonicals. Keep parameters that genuinely change the content (for example, ?id=123 on a database-backed page).

Q: Can I canonicalize across domains?

Yes for syndicated content where the original publisher's domain is the canonical. Both pages must agree, or AI engines may cite either.

Q: Does AMP still matter for AI citations?

Less than it once did. AI engines prefer the non-AMP canonical when it is well-formed. Maintain AMP only where measurement shows continued value.