Test results

Canonical URL Decision: test results

Tested 2026-10-08, skill version 1.0.0 at the time of loading this page. We run every skill on a strong and a weak model before it is listed, and publish both verdicts, including where the weak one fails.

Verdicts

Date
2026-10-08
Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")
Right on all 24 sets of URLs, read by hand: a tracked newsletter address and sorted or paged lists sent to the clean address, a syndicated copy on a partner site set to noindex rather than a cross-domain canonical, a staging host kept out of the index, an untranslated copy pointed at the original language, a two-month sale handled with a temporary redirect that keeps the category as the canonical, a product in two categories given one canonical, permanent moves as 301, removed pages as 410, and an honest unknown where the facts do not settle it. It answered as valid JSON every time.
Weak · claude-haiku-5-5 (Claude Code alias "haiku")
Right on 22 of 24 sets of URLs, read by hand, but it missed two: on the seasonal sale it described the temporary redirect in its reason but filed the decision as self-canonical, and on a product in two categories it added a notes key the answer format does not allow.

With and without the skill

Tested 2026-10-08.

Results with and without the skill, for Sonnet and Haiku
SonnetHaiku
withwithoutwithwithout
URL sets decided right (24 sets)24/2420/2422/2421/24

Same request on both sides, a fence removed first. Read by hand, Sonnet without the skill already handled tracking and sort parameters, paging, permanent moves and removed pages. It missed four: it gave a syndicated copy a cross-domain canonical where the current guidance is noindex on the partner site, put noindex on an untranslated copy instead of pointing it at the original, and on the staging host and the seasonal sale its answer was not valid JSON. Haiku without the skill missed three, among them the syndicated and untranslated copies.

Same cases and the same checks with and without the skill. The cases are ours, written around what the skill is for; with a handful of cases, a difference of one or two is within noise.

Note

Twenty-four sets of URL variants with facts about them, written by us (17 with a trap, 7 plain): tracking and sort parameters, paging, a syndicated copy, a staging host, an untranslated copy, a temporary sale move, products in several categories, permanent moves, removed pages and cases the facts cannot settle. Each answer is parsed as JSON and checked by code: the action per URL, the target address where one is needed, and the exact set of keys. The rules follow Google Search Central guidance on canonical URLs, redirects and syndicated content (the noindex advice for syndication partners), as written into the skill from its sources; they were not re-read for this test. No check was widened. One run per model and set.

What was not measured

  • Models other than the two named above were not run.
  • Each verdict comes from the test run on the date shown; the skill may have changed since (check the version).
  • Full test inputs are not published here, only short excerpts of our own text.
  • Results on your own texts, languages and domains can differ.

Back to Canonical URL Decision · Card (JSON)