Why a single total misleads here
Suppose a page gets four hundred visits in a month. If your audience were one language in one country, that figure would tell you whether the month went well.
Here it might be three hundred locals who always find you, sixty people from a neighbouring country who arrived through a search you did not know you ranked for, and forty who read your English version and could become customers if anyone followed up.
Those three groups have nothing in common except the counter. Averaged together they produce a figure that moves for reasons you cannot see and therefore cannot act on.
Split at the link, not afterwards
The split has to exist in the data before you can read it, and the cheapest place to create it is the link itself.
- Give each language version of a page its own destination rather than one page with a switcher inside it.
- Point one routed short link at all of them, so visitors still only ever see a single address.
- Add a separate link per channel on top of that: one for each profile, one for print, one for the newsletter.
- Leave the arrangement alone for a month before drawing any conclusion from it.
After that, the report answers two questions at once: which language people read, and which channel brought them. Neither answer is available from a total.
Language and country disagree
This is the part that surprises people the first time they see it.
The country figure comes from where the request originated. The language figure comes from which version the visitor was routed to. In a region with long-standing bilingual communities, cross-border commuting and a sizeable diaspora, those two rarely line up.
A concrete example: a company finds that a third of the visits to its Russian language page come from outside the country entirely, while its English page is read mostly by people inside it. Both facts change where the next advertisement goes, and neither is visible without the split.
Small numbers need longer windows
Splitting an audience three ways also splits the sample three ways, and small samples wander for no reason at all.
A version with twelve visits in a week has told you nothing. The same version with a hundred and fifty over a quarter has told you something.
So stretch the window rather than the interpretation. Compare quarters, and resist building a story around a version that had four visits on Tuesday and nine on Wednesday.
What the split actually changes
Three decisions come out of this, and all three are practical.
- Which language to write the next thing in first, which is usually not the one you assumed.
- Which version deserves advertising of its own rather than a translated copy of somebody else's campaign.
- Whether a version is worth keeping at all. A page nobody opens in any month costs upkeep and clutters your sitemap.
The third one is uncomfortable, and it is the reason the exercise pays for itself.
Keeping the comparison fair
One version usually receives more attention than the others, and its numbers are then used to prove it deserved that attention.
Before concluding that a language does not perform, check that its page is as current as the others, that it is linked from the same places, and that its address has actually been shared anywhere. A version published once and never mentioned again is not a failed audience.
Correct the obvious inequality, wait another quarter, then decide. Both the change and the measurement are cheap, which is the only reason this is worth repeating.