Google already speaks our pages. It does not send the click

The gap is not translation. It is the same prompt problem in another language.

The expectation was that Google translates our English page for a Spanish or German searcher, answers them on the results page, and keeps the click. The feature is real and it does appear in our search appearance report, at 192 impressions, 0 clicks and a position of 24.53. On the scale of the site that is a rounding error, and it is not the mechanism.

The non-English traffic that earns no click tells a different story. There are 5 such queries in the export, worth 2,954 impressions and 0 clicks. Every one of them asks for a ranked opinion, in Spanish or in German, and the demand for a forced ranking is stated in the query. They are the prompt-shaped comparison queries from the zero-click study, in another language.

Which means a translated page is not competing for those queries, and it was never going to be. The buyer who types a Spanish prompt into a search box is asking for an answer, and the answer is what travels. What a localised page can be cited for is the assistant's answer, and that is a different return from a different channel.

The demand behind that channel is real and specific. We already earn 491 citations across 15 German grounding queries, and 96.3% of them are on the two topics we localised. That is the case for having done it, and it is a case about citations, not about clicks.

Last verified 2026-10-23. The two localized pages went live on 22 October 2026 and this baseline was taken the day after, so there is no post-launch data yet. The plan is below.

What the translated-results feature does for this site

The search appearance report lists the features our pages are shown in. Two of them produce impressions.

Where the non-English queries are, and what they ask for

The non-English queries in the export, split by language and by whether they ask for a ranking.

Every non-English query in the export asks for a ranking, in one language or the other. The French count is zero, and that is a finding rather than an absence of data.

Every non-English query in the export, and what it asks for

A query was counted as non-English when it carries a function word from that language. The pattern is stated so it can be re-run, and the wording is the searcher's own.

What a translated result is not doing

There are three explanations for a non-English searcher who does not click, and the data points at one of them.

**Google translated the page and answered in place.** That is what the translated-results feature does, and it is doing very little here: 192 impressions, and a position of 24.53. It cannot account for 2,954 impressions of non-English traffic.

**Our page was in the language but the buyer wanted a different page.** Possible in principle, and it would show as clicks on the translated page. There are none, and the localised pages had not been published when the export was taken.

**The buyer wanted an answer, not a page.** This is the reading the query text supports. The Spanish and German queries do not name a format, a document or a guide. They name a set of products and ask for the ranking, and they ask for it in the language they would ask an assistant in.

Where the non-English demand actually shows up

The Bing AI grounding-query report of 4 August 2026 shows where a non-English buyer's question is answered, and with what. It is the clearest evidence we have that the demand is real.

Each language, with what we earn and what we have published

German, French and Spanish, one at a time. The differences between them are the reason we localised into two languages and not three, and the reason the third is the one to watch.

German had measured demand. French was a bet, and it is labelled as one.

Two localised pages went live on the same day, and they were not the same kind of decision. The difference matters and it is visible in the data.

The German decision had a number behind it before a word was written. The Bing AI report from August shows 15 German grounding queries earning us 491 citations, with the top two topics worth 473 of them. We knew which two comparison pages to write, and we knew what they were for, before we started.

The French decision did not have that. There are 3 French grounding queries in the same report, worth 89 citations, and the leading one is the landing-page comparison. The topic is ours and the demand is real, but the volume is a fraction of the German one, and Google's own export gives us no French query at all.

We shipped the French page anyway, and we would defend that, because a market-entry decision and a content decision are different things and the page is a cheap test of a question we could not otherwise answer. But it is a bet, and the page says so. The German page is a bet on a measured number, and the French page is a bet on a small measured number plus an untested assumption. Those should not be read as the same kind of decision, and they should not be measured the same way when the results come in.

What a German buyer is actually asking for

The grounding queries are the closest thing we have to a transcript of what a non-English buyer wants, and they are worth reading closely.

The leading German grounding query asks for a comparison of live-chat tools, by name, with the word Vergleich in it, which is a comparison. The second asks for a comparison of AI visibility tools with the words Sichtbarkeit und Tracking, which is brand visibility tracking. The third asks for the same comparison with a ranking forzado translated into German, which is the prompt-shaped pattern again.

So a German buyer in this category is asking the same questions an English buyer asks, in German, and expecting the same answer. That is the entire market-entry case, and it is why the pages we wrote are evidence-led adaptations of the English ones rather than translations of them: the demand is the same demand, and the answer has to be the same answer in the reader's language.

The one thing that changes is that a German reader arriving on a German page can act on it, because the pricing evidence, the capture dates and the refusal notes are in their language. An assistant citing an English page can produce a German answer, but a German reader landing on it cannot comfortably check it. That difference is small and it is the whole argument for localising the page rather than only the answer.

What we just recorded, and what we will read at 30, 60 and 90 days

A localization programme that cannot be measured is a bet that never gets graded.

The two pages went live on 22 October 2026. We fetched them on the day after and recorded what they serve: the German page answers with status 200 and carries 3 alternate links, and the French page answers with status 200 and carries 3. Each one points at its English original and at an x-default, and each one is its own canonical. That is the contract, and it is the same on both.

The baseline is therefore zero. The search export closes on 2026-10-19, so it contains no row for either page, and the Bing report is from August, before either page existed. There is nothing to compare against yet, and we are not going to imply otherwise.

The measurement plan is written down so it cannot be edited later to suit the result. At 30, 60 and 90 days we re-read three things: the impressions and clicks on the two localised URLs, the number of German and French grounding queries that cite them in the Bing AI report, and the click-through rate of the non-English comparison queries. If the citations do not move, the honest reading is that localised pages earn citations rather than clicks, and that is still a return.

Four things the data says about localisation

The findings, in the order they matter to someone deciding whether to localise.

What a localised page is actually for, in one search

One German grounding query, the page that answers it, and what the reader gets.

A German buyer types a query naming four live-chat tools and asking for a comparison. Bing's assistant has to answer it, and to answer it with a recommendation it needs something to be right about. Before the localisation programme, the thing it had was our English comparison page, which it could read and cite, and which earned 135 citations on that single query.

After it, the assistant has a German page whose pricing evidence, whose capture dates and whose refusal notes are in the reader's language, which means the answer it gives is the same answer in German. That is the whole mechanism. The citation goes to a page the reader can read, and the reader lands on a page that answers their question in the language they asked it.

None of that depends on the reader clicking through, and that is the point. If you localise for clicks, this is a bad bet, because the click in this channel is the exception. If you localise so that an assistant can cite you accurately in a language, the economics are the same economics as the English comparison pages, and the evidence says the demand is there.

The five questions this study keeps getting asked

Four readers, and what each of them should take

The same measurement reads differently depending on what you are doing with it.

**You are deciding whether to write a page in another language.** Localise where you can already see the demand, which is where the assistants are already citing you. For us that is German, and it is the two comparison topics we picked, and that is because the data pointed there rather than because the language looked important.

**You are deciding what to expect from it.** Expect the citation and the answer, not the click. A page in another language is a source for a machine, and the buyer who reaches it through that machine is the return.

**You are auditing your own non-English traffic.** There is a check worth doing, and it takes an afternoon: read the wording of your zero-click queries in other languages. If they read like questions, you have a snippet problem. If they read like instructions to a machine, you have an intent problem.

**You are a market-entry team with a language on a list.** The order of work matters. Find the queries where the assistants already cite you, and write the pages for those, rather than working down a country list.

What this does not tell you

Where this study sits in the rest of our work

Three studies that bear on the same question: what a page is actually for, and how you measure it.

How we checked

Three sources, each read for one question. The search console search-appearance report of 22 October 2026 gives what the translated-results and product-snippet features deliver. The search console query export of the same date gives the non-English queries, and a query was counted as non-English when it carries a function word from that language, a rule stated on the page so it can be re-run. The Bing AI grounding-query report of 4 August 2026 gives the citations we earn per language, and the topic split counts any German query naming a live-chat or AI-visibility comparison.

The two localized pages were fetched over HTTP on 23 October 2026 and their link elements parsed, so the hreflang counts and the canonical are read off the pages as served rather than described from the build. A link is counted as an alternate only when it carries an hreflang attribute, and the RSS link is excluded.

Every figure on this page is a field of the recorded data, and the exports are hashed in the data file so each number can be traced to the row it came from. The page is re-verified quarterly, and the baseline section is the part that changes when the measurement window opens.