Question 1

Do AI crawlers like GPTBot and ClaudeBot respect rel=canonical tags?

Accepted Answer

Partially, and not in the way Google does. OpenAI's GPTBot follows canonical tags as one signal among several but will frequently index both the canonical and the variant URL if the variant has its own inbound citations, particularly from Reddit or Wikipedia. Anthropic's ClaudeBot is more conservative and treats rel=canonical as a strong hint, collapsing duplicates more aggressively in line with Google's behavior. PerplexityBot ignores canonical tags entirely for citation selection and picks whichever URL has more cross-domain references in its real-time index. Common Crawl, which feeds the training corpora behind most foundation models, records both URLs and lets downstream consumers decide. The practical implication is that canonical tags still matter, but they no longer guarantee deduplication across the AI search surface. Operators need a layered defense — canonical, plus noindex on truly redundant variants, plus syndication agreements that specify which URL the publisher prefers.

Question 2

Should I use cross-domain canonical or noindex when syndicating content to Medium or LinkedIn?

Accepted Answer

Cross-domain canonical is the better default for AI search visibility, but only if the syndicating platform actually emits the canonical tag pointing to your original. Medium honors cross-domain canonical when you publish through the Import Story tool or when your CMS uses the Medium API to push posts. LinkedIn does not support cross-domain canonical for newsletter or article posts — there is no way to tell LinkedIn that your blog is the original. For LinkedIn republishes, the safer pattern is to delay the syndicated version by two to four weeks, rewrite the lede, and let Google and AI crawlers index the original first. Noindex on the syndicated version is a third option but usually wastes the audience-reach value of publishing on a high-authority surface. Most operators land on cross-domain canonical for Medium, modified republish with delay for LinkedIn, and original-only for Substack.

Question 3

What is the right canonical strategy for paginated content like blog archives and product listing pages?

Accepted Answer

The 2026 best practice is self-referential canonicals on each paginated page rather than a single canonical pointing to page one. Google's Search Central documentation deprecated the rel=prev/next signal in 2019 and now treats each paginated page as its own indexable URL. The same logic applies to AI crawlers. Pointing every paginated page back to page one with rel=canonical tells crawlers to ignore the deeper pages entirely, which means products or articles only reachable through pagination get discovered slower or not at all by GPTBot and ClaudeBot. The exception is filtered or sorted versions of the same listing — a category page sorted by price-low-to-high should canonical to the unsorted version because the underlying content set is identical and the URL variation is functionally a parameter. The distinction is duplicate content (canonical to one URL) versus different content slices (self-canonical on each page).

Question 4

How does Google-Extended handle canonical tags differently from regular Googlebot?

Accepted Answer

Google-Extended uses the same crawling infrastructure as Googlebot and therefore reads canonical tags the same way at the fetch layer, but the indexing decisions diverge downstream. Googlebot feeds the traditional search index where canonical tags drive URL consolidation. Google-Extended feeds the Gemini training corpus and the AI Overviews answer layer, where the deduplication logic is different — Google-Extended will sometimes include both the canonical and variant URLs in the training set because diverse text expressions of the same idea improve model robustness. The publisher control mechanism is robots.txt directives that allow or block Google-Extended specifically, which is independent of how Googlebot treats the same URLs. Sites that want their canonical tags to flow through to AI Overviews need to verify that Google-Extended is allowed in robots.txt and that the canonical URL is also crawlable by Google-Extended, not just Googlebot.

Question 5

Are AMP canonical tags still causing problems for AI crawlers in 2026?

Accepted Answer

Yes, and the technical debt is larger than most teams realize. AMP officially lost preferred-treatment status in Google search in 2021, and the AMP project itself went dormant by 2024, but a meaningful number of news and publisher sites still ship AMP variants with the corresponding rel=amphtml and rel=canonical pair. AI crawlers handle AMP inconsistently. GPTBot will sometimes fetch the AMP variant first because it loads faster, then attribute the citation to the AMP URL instead of the canonical. ClaudeBot follows the canonical reliably. PerplexityBot picks whichever loaded first. The cleanup is to either remove the AMP variants entirely or to add aggressive HTTP redirects from the AMP URLs to the canonical, plus noindex headers on the AMP responses. Publishers that have not done this work are still leaking citation attribution to amp.html URLs that no longer surface in any user-facing experience.