sitemap-models.xml on our appliance site is 1.4 MB. I parsed it, saw 8,400 <loc> entries, and wrote "8,400 model pages" into a note. Then I built half a chart out of that number before checking what the URLs actually pointed at.
817 brand-category pages. 6,756 detail sub-pages. 827 model pages.
The number I wanted was three percent of the file.
Why I checked at all
We had a second sitemap — sitemap-core.xml, 1,668 URLs. Two files, one site, and the obvious question is how much they overlap. The answer is zero. Not "a little" — the intersection of the two URL sets is the empty set.
That's allowed. A sitemap index can point at disjoint shards. But disjoint shards with names that describe neither shard is how you end up quoting 8,400 model pages in a meeting.
What the two files hold
| File | URLs | Model pages | Brand-category | Detail sub-pages |
|---|---|---|---|---|
| sitemap-core.xml | 1,668 | 1,655 | 0 | 0 |
| sitemap-models.xml | 8,400 | 827 | 817 | 6,756 |
| Overlap | 0 | 0 | — | — |
core is 16 brands. models is 111 brands, 101 of which appear nowhere in core. So the split isn't curated-versus-generated, it's big-brands-in-one-file and everything-else-in-the-other, with a filename that claims otherwise.
Grouping the detail pages
| Suffix | Count |
|---|---|
| /manual/ | 1,689 |
| /specs/ | 1,689 |
| /parts/ | 1,688 |
| /problems/ | 1,688 |
| /error-codes/ | 2 |
Four suffixes cover 1,689 models. Total model root pages: 2,482. Which leaves 793 models declared with nothing behind them.
We fetched /manual/ on four of those 793 at random. All four returned 200.
Three things I didn't want to find
Roughly a third of the site is undeclared. 793 × 4 = 3,172 pages that answer 200 and appear in neither sitemap file. If you count our pages from the sitemaps you get 6,756 detail pages. There are probably 9,928.
87% of the gap is one category. 693 of the 793 undeclared models are refrigerators, the second-biggest category we have. Manual, specs, parts and problems run across eight categories and skip the one with the most models. Nobody wrote that down as a decision.
Two page types, split 1,688 to 2. /problems/ and /error-codes/ are the same idea with two names. /problems/ has 1,688 URLs. /error-codes/ has two: a Maytag MVWB965H and a Whirlpool WTW8000D, both washers. Our homepage title tag says "Appliance Manuals, Specs, Parts & Error Codes," which is a promise backed by two URLs.
So I opened four /problems/ pages. Each one states there's no public code table for that model. None of the four contains a code.
Where the note went wrong
I counted a sub-page as a model page because the parent file said models. That's a naming bug in the artifact, not a parsing bug in my code — and it's the more dangerous kind, because the parse looked correct. Renaming the file wouldn't fix the site, but it would have cost me nothing and saved the hour.
What this doesn't tell you
Whether any of the 3,172 undeclared pages got indexed. Sitemaps are a hint. Plenty of unlisted URLs get crawled through internal links and plenty of listed ones never get looked at. That's Search Console coverage data, and I didn't pull it.
And four out of 1,688 is enough to say "these pages exist and say this." It's not enough to say "all 1,688 are empty." I'm not saying that.
Disclosure
Our site: modelnumberfinder.com. Both sitemaps are public and every figure here comes from a parse you can rerun with xml.etree and about ten lines.
The fix is a sitemap generated from the route table instead of a maintained list, and one page type where there are two. Both are boring. That's why they're still open.
Top comments (0)