What gets checked
Well-formed XML (an unescaped & in a URL is the classic one), UTF-8, the exact namespace http://www.sitemaps.org/schemas/sitemap/0.9 (https or a trailing slash is wrong), one absolute <loc> per entry under 2,048 characters, W3C Datetime lastmod values with a timezone when a time is given, valid changefreq words, priority between 0.0 and 1.0, duplicates, mixed hosts or protocols, and the 50,000-URL / 50 MB limits.
Extensions are checked too: image:loc and the 1,000-images-per-page cap; the required video fields, duration and rating ranges; news publication name, language, date and title, plus the two-day freshness rule; and hreflang codes, self-references and reciprocal return links within the file.
Errors, warnings and info
Errors are things that can make Google or Bing reject the file or skip entries. Warnings are likely mistakes, such as listing both http and https URLs, a lastmod in the future, or every URL sharing the same date. Info covers harmless notes like changefreq being ignored by Google.
Fetching by URL also reports redirects, the HTTP status and the content type the server sends — sitemaps served as text/html are a common reason for “Couldn’t fetch” in Search Console.
Questions
Search Console says “Sitemap could not be read”. Why?
Usually the URL redirects to a login or HTML error page, the XML is malformed (often an unescaped &), the namespace is wrong, or the file is served with an HTML content type. Fetch it here by URL to see which.
Is a sitemap with warnings still valid?
Yes. Warnings point at likely mistakes that reduce the sitemap’s usefulness but do not break it. Errors are what you must fix.
Can it validate .xml.gz files?
Yes — uploads and fetched URLs ending in .gz are decompressed in the browser (uploads) or by the fetcher before checking.
Does it check whether each URL returns 200?
Not yet; the validator checks the file itself. To find redirects, 404s and noindex pages, crawl the site in the creator, which reports a status for every URL.