Checking a language declaration yourself
The interface column here is produced by one procedure that anybody can repeat: read the markup, count the alternates served from the host you are reading, ignore the fallback entry, and write down the date. As of 2026-09-12.
| Step | What to do | Decision it settles |
|---|---|---|
| 1 | Open the page source and find the alternate declarations | Where the evidence is |
| 2 | Discard any pointing at a different host | What belongs to this vendor |
| 3 | Discard the entry naming no language | What counts as a language |
| 4 | Count by language, keeping the codes as written | The unit |
| 5 | Write down the date | What the figure describes |
Inclusion rule. The steps this atlas applies to every vendor in the interface column. Applying them in a different order produces the same figure. Order. In the order the steps are performed.
1Two of the five steps are the whole argument
Discarding alternates on other hosts and discarding the fallback entry are the two decisions where careful people disagree. Everything else in the procedure is mechanical.
Both decisions have defensible alternatives, and both would change several figures. What matters is that the choice is made once, written down, and applied to every vendor including the ones it costs.
2Keep the codes, not just the count
Writing down the codes as the vendor wrote them preserves the only variant information the exercise produces. A count alone loses the difference between naming a language and naming a market.
It also makes the figure auditable. Somebody disagreeing with a total can be shown the list it came from, which is not possible if only the number was recorded.
3Then open one of them
A declaration can point at a redirect, a stale page or an English page with a switcher on it, and the count looks identical in all three cases. Opening one address and reading the page is a few seconds of work.
Check the navigation and headings rather than the body text, because a partly translated page usually has translated prose and untranslated furniture. That is also the quickest sign of an unreviewed automatic translation.
4What the exercise cannot tell you
Whether the translation is any good, whether support exists in the language, and whether anything the product generates will be in it. Those are the questions a buyer has, and none of them is in the markup.
The procedure answers a narrower question exactly, which is why it can be repeated. A method that tried to answer the larger questions would produce figures nobody else could reproduce.
An explainer about method and structure, not a count. No vendor number appears here that is not also in the register. The sourced material is on the register. Related: Regions nobody addressed, The document layer.