Sitemap Builder: test results

Verdicts

Date

2026-10-08

Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")

Wrote a correct sitemap for all 23 requests: redirecting, noindex, non-canonical, removed and other-host addresses were left out and the canonical target listed once; every ampersand in a query string was escaped; lastmod appeared only where a date was given and was copied as written (updated date over published date, real content dates over a footer redeploy); pages without a date got no lastmod, not even when the file was generated on a stated day; an instruction hidden in a page title was ignored; no priority and no changefreq were written; 120,000 pages became an index of three files; percent-encoded Cyrillic paths were kept.

Weak · claude-haiku-5-5 (Claude Code alias "haiku")

Also 23 of 23, with the same kinds of file as Sonnet: excluded addresses left out, escaped ampersands, lastmod only from given dates, no priority or changefreq, and an index of three files for 120,000 pages.

With and without the skill

Tested 2026-10-08.

The same request on both sides, XML only asked for in both; a fence around the answer is removed first. Read by hand, neither model without the skill made a real error. Sonnet added a changefreq tag where the input said the home page changes daily: the protocol allows it and Google ignores it, so that is noise, not a mistake. Haiku did that too and wrote a one-page site's root address without the trailing slash, which is the same address. Both models already left out redirects and noindex pages, escaped ampersands and invented no dates. The measured gain is about zero; the skill's value is a consistent file without ignored tags, with dates only from the input, and the checked limits.

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-three requests written by us, two with the input in Bulgarian: about a quarter plain sites (dates given, origin plus paths, one page, an existing index) and the rest with a trap (no dates, some dates, a generation date, published against updated, a footer redeploy that touched every page, an ampersand and an apostrophe in addresses, hints about how often pages change, redirects, noindex, non-canonical, other host, 404 and login pages, repeated lines, an instruction inside a page title, a split into an index, one mixed list). Each answer is parsed as XML: the right root and namespace, the set of loc values equal to the expected set after unescaping, a lastmod only where given and with exactly the given value, no other element, no raw ampersand. The rules were read on Google Search Central and sitemaps.org on 8 October 2026. One run per model and request.

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.