Why the same question gets a different answer in Zurich and Austin
Language, local sources, retailers and rules shape what an engine says. How to measure visibility per market without mistaking a translated keyword for a question a buyer asks.
A climber in Zurich and a climber in Austin ask an engine for the best hardshell for alpine routes. They get different shortlists, ordered differently, citing sites the other has never heard of. Nothing is broken. The engines are doing what search did before them: answering from the web that is closest to the person asking. If you sell in both places, you have two visibilities, and one number that averages them describes neither.
Four reasons the answers differ
Language. The German question is not a translation of the English one. It is phrased differently, it retrieves different pages, and the pages it retrieves were written for a different reader. “Beste Hardshelljacke zum Alpinklettern” lands on German test reports and forum threads; the English phrasing lands on American review sites. The engine did not choose to favour local brands; the corpus it retrieved from is local.
Local sources. Every market has its own reviewers, magazines, forums and comparison sites, and they carry their own preferences. An engine that cites a Swiss mountaineering magazine will name the brands that magazine tests. A brand that is unreviewed in a market is, to the engine, close to unknown there.
Retailers. Retailer pages are cited often for buying questions, and retailers differ by country. Which brands a market’s large outdoor retailers stock, and how they describe them, shapes the answers in that market. A brand sold direct-only in one country loses the retailer pages that carry it elsewhere.
Regulation and availability. Products are sold under different names, at different prices, or not at all. Claims allowed in one jurisdiction are illegal in another, so product pages differ. An engine answering in Switzerland may know a product does not ship there and leave it out, correctly.
There is a fifth, quieter reason: the engines themselves localise. The same product from the same company behaves differently by country and language setting, in ways none of them document in detail. We can observe the difference; we do not claim to explain it.
Why collection has to happen inside the market
It is tempting to collect everything from one place and switch the language. It does not produce the same answers. Engines use the location they see, the language of the question, and account and interface settings together, and a German question asked from a data centre in Virginia is not the answer a buyer in Munich sees. The result is a blend that belongs to nobody.
This is why Bluemoon collects each market from inside that market, in that market’s language. It costs more. It is also the only way to store the answer a buyer in that market would actually have received, which is the thing you want to measure.
Designing a multi-market measurement
Questions per market: native, not translated
Start from buyer intents, not keywords. For each intent, write the question the way a native speaker would ask it in that market, with a native speaker in the room. Then keep a map from intent to question per market, so that when you compare DE and US you are comparing the same intent, not the same string.
| Intent | DE phrasing | US phrasing |
|---|---|---|
| Best product for a use case | Welche Hardshelljacke ist die beste zum Alpinklettern? | What’s the best hardshell jacket for alpine climbing? |
| Comparison of two brands | Nordkamm oder Alpsteel für Hochtouren? | Nordkamm vs Alpsteel for mountaineering? |
| Value pick | Gute Hardshell unter 400 Franken? | Good hardshell jacket under $400? |
| Fit and sizing | Fällt die Nordkamm Gratwand Pro klein aus? | Does the Nordkamm Gratwand Pro run small? |
Minimum sample, per market and per engine
The rule from our note on what observed means applies on each side of a comparison separately. A rate needs its count, and a difference between two rates needs both counts to be large enough. If DE has 1,200 answers and US has 30, the US figure is the weak side, and the comparison is only as reliable as that side.
Same engines, same period, same collection cadence on both sides. If one market was added last month, say so and compare only the overlapping window. Where an engine is not measured in a market, show a dash for it in both, so the denominators stay comparable.
An illustrative reading
Here is a made-up brand, Nordkamm, tracked on the same intent set in DE and US over the same collection window. The numbers are invented; the point is the shape.
Illustrative example with invented figures for a fictional brand. Equal counts on both sides make the comparison fair; with unequal counts, the smaller side sets the limit.
How to read a difference
- 1Check both denominators
The smaller count sets the reliability of the comparison. If one side is thin, say so before anything else.
- 2Check the engines
Are the same six engines in both, over the same days? A missing engine on one side is a dash, not a zero.
- 3Compare sources before brands
List the top cited domains in each market. Most brand differences are source differences wearing a costume.
- 4Compare roles, not just presence
Named in 40% but recommended in 14% is a different problem from named in 17%.
- 5Look for market facts
Not sold there, a different product name, a price that lands in another bracket, a claim you cannot make locally.
- 6Then classify the gap
A page to publish in that language, a placement to pitch with a local reviewer or retailer, or a distribution fact to accept.
Six steps, in order. The first two ask whether the difference is real; the rest ask why.
In the Nordkamm example, the honest reading is not “US visibility is bad”. It is: with equal counts on both sides, the brand is named less than half as often in US answers and recommended rarely, and the first place to look is the list of US-cited sources, where a brand unreviewed by the sites the engines rely on would produce exactly this shape. The next step is a source table per market, not a marketing plan.
What this costs and why we pay it
Collecting from inside each market multiplies the work by the number of markets, and it forces the awkward conversation about which markets a plan includes and which show a dash. We think both are features. A dash for France is more useful than a French number made from a Virginia proxy, and a company that sells in eight countries deserves to know it has eight visibilities. The full method, including how markets are defined and collected, is on the research page.
Related
What “observed” means: how we count a mention
Every rate in Bluemoon is built from stored answers. Here are the definitions behind the count, the limits of a daily sample, and the short list of things we infer instead of observe.
Playbooks6 min read
How to read an AI answer like an analyst
A stored answer holds more than your brand name. Five steps that take you from the question to the roles, the sources and the gap, and out the other side with one brief.
Product6 min read
Why Bluemoon never invents a number
AI visibility tools are easy to fake: a confident percentage with nothing behind it. Here is the rule we build every screen around, and what it costs us.