A canonical URL is the address a search engine treats as the main version of a page when the same content is reachable at several URLs. Site owners suggest it with a rel="canonical" link, a redirect or a sitemap, but the search engine makes the final choice; Google calls a canonical preference a hint, not a rule.
How canonicalization works
Some duplication is normal. Google lists common causes: regional versions of the same page, separate mobile and desktop URLs, HTTP and HTTPS versions, sorting and filtering on category pages, and demo sites left open to crawlers. Bing adds URL parameters, letter case and trailing slashes. Google's own example is a green dress product page that is also reachable through a URL carrying a ?gclid= parameter.
When Google finds pages with the same or very similar main content, it clusters them and marks as canonical the one it judges most complete and useful for searchers. That page is crawled most regularly, used as the main source for evaluating the content, and usually the one shown in results.
Google ranks the signals by strength: redirects and rel="canonical" annotations are strong, sitemap inclusion is weak, and they work better combined. It also says what not to use for the job: robots.txt, the URL removal tool and noindex. For JavaScript sites, Google's preferred approach is to put the canonical in the HTML source and make sure scripts do not change it, one reason server-rendered HTML matters for crawlers.
The link relation itself is defined in RFC 6596, published by the IETF in April 2012. It tells authors to expect search engines to index content only from the preferred URL, consolidate signals such as link popularity onto it, and display it as the representative address.
Why canonical URLs matter for AI answers
An AI answer cites a URL, so duplicates can split or misdirect citations. Two Bing product managers wrote in December 2025 that large language models group near-duplicate URLs into a cluster and choose one page to represent it, and may pick an outdated or unintended version when the differences are small. Their fixes are familiar SEO ones: canonical tags, consistent metadata, 301 redirects and IndexNow.
Google says a page must be indexed and eligible for a snippet to be a supporting link in AI Overviews or AI Mode, and its Search Console report for generative AI features assigns most data to the canonical URL rather than to duplicates. In that report, the canonical is usually the address that gets counted.
An example from our scans
In our scans of AI answers about AI visibility software, Perplexity cited one cyberpress.org article, a ranking of MCP SEO tools, under two addresses across 11 citations: the plain article URL and its AMP version ending in ?amp=1. The AMP page declares the plain URL as its canonical, which is how a crawler learns the two are one article. A report that counted citations by exact URL would show two pages where there is one.
Common misconceptions
- "rel=canonical is an order." Google treats it as a strong signal that it can override.
- "Duplicate content gets penalized." Google says some duplicate content is normal and not a spam violation, and Bing says it does not trigger penalties on its own. The cost is diluted signals and wasted crawling.
- "Block duplicates in robots.txt." Google advises against it, because a blocked URL can still be indexed without its content.