Your product is in English, your support is in English, your standups are in English. Your Search Console report is not. Roughly half the queries bringing people to your site are German, written by people who will never read a word of your interface copy — and nobody on the team can tell whether those queries represent buyers or noise.
This is the standard situation for a company built in Berlin by people who arrived from somewhere else. It is not a translation problem and it is not a marketing problem. It is a measurement problem: the report is telling you something specific, in a language the people reading the report do not operate in, and so the information sits there unused for months.
The report already knows more than the team does
Search Console records what people typed, how often it produced an impression, how often that turned into a click, and roughly where you sat in the results. Four columns, nothing more. In a single-language company those four columns are read once a week and forgotten. In a company where the audience and the operators speak different languages, they are the only honest description of who is actually looking for you.
| Column | What it counts | What it does not tell you |
|---|---|---|
| Impressions | Times your page appeared for that query | Whether anyone looked at it |
| Clicks | Times someone chose your result | What happened afterwards |
| Click-through rate | Clicks divided by impressions | Whether the ratio is good for that position |
| Average position | A mean across every impression | The spread that produced the mean |
The fourth row is where most misreadings begin. An average position of nine can mean you sat at nine every single time, or that you sat at three for a handful of queries and at twenty for everything else. Those are two entirely different situations calling for two entirely different responses, and the column presents them identically.
There is a second, quieter problem with averages in a bilingual property. The mean is calculated across every impression regardless of query language, so a page performing well for its English audience and badly for the German queries it accidentally attracts reports as mediocre for both. Nothing in the interface flags this. The page looks like an ordinary underperformer, gets rewritten, and comes back with the same number — because the number was never describing one thing in the first place.
Split by query language before anything else
Most teams segment by page or by device. In a company whose site is English and whose market is partly German, both of those are the wrong first cut. The useful split is by the language of the query, because it separates two audiences with different intent, different competition and different conversion behaviour that otherwise average into one meaningless number.
English queries, English pages
The audience you designed for. Usually other people in the same situation as your team — international, recently arrived, working in English by default.
- Matches your existing copy
- Competition is global, not local
German queries, English pages
The audience finding you by accident. They arrive, see a language they were not expecting, and most of them leave within seconds.
- Impressions look healthy
- Click-through rate is quietly terrible
Once those two segments sit side by side, a question that previously had no answer becomes answerable: is the German demand large enough to justify German pages? Not as a matter of principle, but as a matter of counted impressions. In many cases the answer is a clear yes for two or three specific topics and a clear no for everything else — which is a far more useful conclusion than a decision to translate the whole site.
Doing this split by hand is tedious but not difficult: German queries are recognisable at a glance, and a couple of hundred rows can be sorted in twenty minutes the first time. What matters is that it is done once properly and then kept, rather than re-derived from memory whenever someone asks. Once the segments exist as a saved filter, every subsequent question inherits them for free, and the weekly review stops starting from zero.
One caution worth stating plainly: the split is by query language, not by page language. A German query can perfectly well land on an English page, and that combination is exactly the case worth isolating. Teams that instead segment by URL prefix end up measuring their own site structure rather than their audience, which feels rigorous and answers nothing.
Neither column means anything alone
Good position, poor click-through
You are visible and being skipped.
- In a bilingual setting this is usually a language mismatch. A German query returns an English title. The searcher reads the title, concludes the page is not for them, and moves on. The position is fine; the promise is wrong.
- Check the title and description first. They are what the searcher actually evaluates. Rewriting the page body changes nothing about this specific failure.
- Compare against similar queries. If English queries at the same position convert at three times the rate, the gap is the language and not the offer.
- Decide deliberately. Either serve those queries in German or stop appearing for them. Sitting between the two options is the only genuinely bad outcome.
Poor position, strong click-through
The few people who see you want exactly this.
- This is the most valuable pattern in the whole report. Demand is confirmed and the offer matches. Only the exposure is missing, which is the part that responds to work.
- It hides in low-volume rows. A query with sixty impressions and a very high rate will never appear in a report sorted by clicks, which is how most teams sort.
- Sort by rate, filtered to positions eight and below. One filter, applied once, surfaces a list that has usually been sitting there unread for a year.
- Treat each row as a page brief. The query is the topic, the rate is the evidence, and the position is the gap you are closing.
Both patterns are invisible in a summary view. They only appear when position and rate are read as a pair, which is why a dashboard showing total clicks over time — the default in most reporting setups — provides almost no operational information at all. It tells you whether things are broadly getting better. It never tells you what to do on Tuesday.
A third pattern deserves mention because it is specific to companies in this position: strong position, strong rate, no revenue. German queries that convert well into visits and never into enquiries usually indicate that the visitor found what they were looking for informationally and hit a wall at the point of contact — a form in English, a pricing page assuming a different market, a support promise that does not extend to their language. The report ends at the click, so this failure is invisible in it entirely and only surfaces when someone compares the two segments downstream.
Search Console and rank tracking answer different questions
| Search Console | Rank tracking | |
|---|---|---|
| Covers | Only queries that produced an impression | Any query you choose to watch |
| Timing | About two days behind | Sampled on a schedule |
| Position figure | An average over all impressions | A reading at a moment in time |
| Best used for | Discovering demand you did not know about | Watching queries you already care about |
For a bilingual company the first row matters most. Search Console can only report German queries you already appear for. Queries where you are absent entirely — because no page addresses them — leave no trace at all. That blind spot is precisely where the untapped German demand sits, and closing it requires a deliberately chosen watch list rather than a report of what already happened. Keeping both views in one workspace removes the reconciliation step that otherwise consumes an afternoon each month.
A practical consequence for the watch list: it should contain German queries you do not yet rank for at all. That feels wrong at first, because the list stays empty for weeks. But an empty row is information — it says the gap is still open — whereas a list containing only queries you already appear for can never tell you anything you did not already know.
The queries that never become clicks
A growing share of searches now end inside a generated answer. The user reads a synthesised paragraph, gets what they needed and never visits a result. In Search Console this shows up as impressions without clicks, and it is indistinguishable from a page that simply performed badly — which means the standard interpretation is wrong for a growing slice of the report.
This matters twice over in a bilingual setting, because generated answers are produced independently per language. Being cited in English answers tells you nothing about whether you are cited in German ones, and the two often diverge sharply — an English-language source with strong technical documentation can be well represented in one and entirely absent from the other. Checking both requires asking the same question twice, once in each language, and recording the answer. Tracking that in a single place alongside the conventional numbers is what keeps the comparison honest over time.
The practical instruction is short. Once a month, ask three or four questions that matter to your business in both languages, note whether a generated answer appears, and note whether you are mentioned in it. Ten minutes of work, recorded consistently, produces a trend line within a quarter — and that trend line is the only evidence available on this question, since no report exposes it directly.
Three audiences, three formats
Weekly, detailed, filtered
Wants rows it can act on. A table view of fifty to two hundred rows per page is enough, because filtering happens anyway.
- Split by query language
- Sorted by rate, not clicks
Quarterly, short, comparative
Wants four numbers against the same quarter last year. A printed export holding up to two hundred and fifty rows is more than sufficient.
- No query-level detail
- Year-on-year framing
On demand, complete, raw
Wants everything. Machine-readable exports carry up to ten thousand rows and load straight into whatever you already use.
- Unfiltered
- Joined against your own data
Occasional, evidential, sourced
Wants attribution that survives scrutiny. What counts here is a documented trend, not this month's figure.
- Traceable over time
- Tied to enquiries, not impressions
The most common mistake in a small team is using one export for all four. A ten-thousand-row table in a board pack looks thorough and goes unread; a summarised view in the weekly meeting generates questions that the detailed view would have answered immediately. Separating them costs an hour once.
Asking the report a question directly
Most of the time spent with this data is not analysis. It is navigation — finding the view, setting the filter, remembering where last month's comparison lives. An assistant that accepts plain-language questions removes that step: each question pulls between zero and three data blocks depending on what it actually needs, and the last twenty messages stay available as context.
- Ask in the language you think in. The question can be in English even when the underlying queries are German. That alone removes the largest practical barrier for an international team.
- Four filters narrow the scope. Enough to isolate a language segment, a period and a property without building a saved view for every combination.
- Twenty messages of history. Enough for a follow-up chain; if you routinely hit the limit, the original question was too broad and should be split.
- Zero blocks is a valid answer. Some questions are answered from context alone, and not loading data is faster than loading it unnecessarily.
For a weekly review this replaces most of the clicking. The question becomes "which German queries gained impressions but lost clicks last month" rather than a sequence of filter selections — and because the phrasing is recorded, the same review can be repeated identically four weeks later. Running that review inside the same workspace that holds the tracking data means the follow-up question does not require switching tools.
What the first thirty days look like
| Week | Focus | Output |
|---|---|---|
| One | Connect Google sources, confirm coverage | A single sign-in replacing scattered access |
| Two | Split every query by language | Two segments with real volumes attached |
| Three | Sort by rate below position eight | A shortlist of confirmed unmet demand |
| Four | Set the watch list and the report cadence | A repeatable weekly review under fifteen minutes |
Week two is the one that changes how the company thinks. Until the language split exists, German demand is an anecdote someone mentions occasionally. After it exists, it is a number with a size, and the decision about whether to serve it becomes an ordinary prioritisation question rather than an argument about strategy.
Which queries are realistically reachable is a separate question from which ones you currently appear for, and it is answered by keyword research rather than by the report. Where the two disagree — strong demand, no page, no impressions — is where the next quarter's work sits, and turning those gaps into pages is what content strategy covers. Reading the report in one connected environment keeps all three views on the same screen instead of in three separate tools.
Open your query data by language
Questions from teams working across two languages
Why do German queries show impressions but almost no clicks?
Because the title in the results is in English and the searcher decides before clicking. The page may be perfectly relevant; the listing does not say so in a language they read. Either serve those queries in German or accept that the impressions will not convert — but do not treat it as a content quality problem, because it is not one.
Should we translate the whole site?
Almost never as a first step. Split the queries by language, find the two or three topics carrying real German volume, and build those pages properly. A full translation produces a large number of thin pages that compete with each other, which is measurably worse than a small number of well-made ones.
Our average position improved but clicks fell. What happened?
Usually one of two things. Either you started appearing for a broader set of queries at mediocre positions, which pulls the average in one direction while adding impressions that never click; or an answer box now satisfies the query before anyone reaches a result. Check whether impressions rose at the same time — if they did, the first explanation is the likely one.
How long before a change shows up in the report?
Four to eight weeks for the first movement, with two days of lag on top of that. Reviewing weekly is fine for spotting failures; judging whether something worked needs a longer window. Teams that reassess after ten days consistently conclude that nothing works.
Can one person handle this alongside another role?
Yes, once the setup exists. The recurring cost is roughly fifteen minutes a week for the review plus a longer session each quarter. The expensive part is the initial configuration — connecting sources, defining segments, deciding formats — and that is a one-off of a few hours.
Does any of this differ for a company with several domains?
The reading does not change; the administration does. Each property needs its own connection, and comparisons across properties are only meaningful if the periods and filters match. Labels that apply consistently across every view are what make that comparison possible without rebuilding it by hand each time.