Which Language Should You Actually Measure?

The objection is already forming, so let us start with it: obviously you measure in the language of the market. You sell in Germany, you ask the question in German. It is the whole reason a per-language score exists rather than one global number, and it is the thing most brands get wrong by never doing it at all.

It is also, for technical B2B, about half of an answer. The missing half is not a detail. It changes which competitors you see, which gaps look urgent, and whether the report you produce describes your market or a slice of it.

What the rule gets right

Start with why the rule exists, because it is sound.

An assistant asked a question in German answers largely out of the German-language web. Not a translation of the English answer: a different answer, assembled from different pages, naming a different set of companies. A supplier that dominates the English-language conversation in a category can be entirely missing from the German one, because the German-language web says less about it. This is not a subtlety at the margins. It is routinely the largest gap in an export brand's whole picture, and it is invisible to anyone checking only in English.

So the rule is right about the mechanism. Language is not a wrapper around one underlying answer. It is a different corpus producing a different answer, and if you sell into 5 markets you have 5 independent positions rather than one position seen from 5 angles.

Where it breaks: the buying committee does not share a language

Here is the part the rule skips. In industrial and technical purchasing, the people who decide are not one person, and they do not research the same way.

The engineer specifying a component frequently works in English, whatever their nationality, because that is the language the material arrives in. Datasheets are in English. Standards documents are in English. Part numbering, tolerance tables, application notes, the forum thread where somebody solved the same problem: English. An engineer in Lyon or Bologna asking an assistant which suppliers can machine a particular alloy to a particular tolerance may well ask in English, and get the English answer, and never see the French or Italian one.

The procurement officer at the same company is doing something different. They are looking for suppliers who can deliver, invoice, and be reached, and that search is far more likely to run in the local language, because the practical constraints are local ones. Management, when it gets involved, is local too.

Which means a brand can be in a genuinely split position: named in the English answer that engineers see, absent from the local answer that procurement sees, or the reverse. Both are real. Measuring only one and calling it your visibility in that market is a reporting error, not a shortcut.

The test that settles it, per market

This is empirical, and you do not have to reason about it in the abstract. It is settled per market with the same question asked twice.

Take a real buying question, phrased the way a specifier would phrase it, with the technical constraint in it rather than a category label. Ask it in the local language. Ask the same question in English. Do not compare the wording of the two answers, which is a distraction. Compare the company names.

Three outcomes, and each one means something different for what you do next. If the two lists are close, the language question is not where your problem is, and you can measure in either and move on. If the English list includes you and the local one does not, your gap is in the local web: the German or Spanish or French material about you is thin, and that is a content problem in that language, not a product problem. If the local list includes you and the English one does not, you are a local champion who disappears the moment a specifier researches in the language the technical material is written in, which is a more dangerous position than it looks, because that specifier is often the one who writes the shortlist.

Two ways this gets misread

The first is treating a shared language as a shared market. Asking in Spanish does not tell you about Spain and Mexico and Colombia at once; asking in German does not settle Germany, Austria and Switzerland together. The answers diverge because the underlying web does, and a single Spanish-language result read as continental coverage is one of the easier ways to be confidently wrong.

The second is expecting the language split to track geography. It does not, and it should not be sold as if it did. What you are measuring is which corpus the assistant drew on when the question arrived in a given language. It is not where the person asking was sitting, and no tool in this category can tell you that from an assistant's answer. PSentry segments by language and market for exactly this reason, and it does not claim to geolocate anybody, because the answer does not contain that information.

What to do instead of picking one

Stop treating it as a choice and treat it as two axes of the same prompt set.

Split your prompts by the role that would ask them. Specification questions, with constraints and units and standards in them, get asked in English and in the local language, both. Sourcing questions, phrased the way a buyer looking for a supplier phrases them, get asked in the local language first. You will find the two sets behave differently, and that difference is more informative than either set alone.

Then hold the set still. The point of measuring is comparison over time, and a prompt set you keep rewriting produces movement that is entirely your own doing. Add prompts when you enter a market or launch a line. Do not reword the existing ones because a phrasing occurred to you.

Doing this properly is where the arithmetic turns against manual checking: a modest prompt set, across 4 assistants, in 5 languages, repeated often enough to mean anything, is not an afternoon's work. But the decision that matters here is not which tool runs it. It is refusing the tidy assumption that one market equals one language, when the people you are selling to have never behaved that way.

Questions this always raises

We only sell in one country. Does any of this apply?

Partly. If your buyers are technical, some of them are researching in English even at home, particularly for anything involving standards or specifications. Running your core prompts in English as well as your own language costs little and occasionally turns up a competitor set you did not know you were in.

Should we translate our prompt set or rewrite it?

Rewrite it. A literal translation carries the source language's phrasing, and buyers in that market do not phrase things that way. It also tends to carry your category label, which may not be the label they use.

If English wins everywhere, why bother with local languages at all?

Because it usually does not, and because the case where it does is itself a finding worth confirming rather than assuming. Markets where the local answer names a domestic supplier and the English one names you are common, and they are the ones where a competitor is quietly being handed the sourcing conversation.