In Amsterdam the same person may look for you in Dutch on a Monday morning and in English on a Thursday evening. Averaging those two visits into a single figure does not describe that person — it removes them from the report.
Most advice about bilingual measurement assumes the second language reaches a second audience: one market here, another over there, two sets of figures that never needed comparing. In this city that assumption quietly fails. The finance lead at an international headquarters, the operations manager at a scale-up where half the staff never learned Dutch, the partner at a tax firm guiding a foreign company through its first year here — all read both languages and pick one by habit, by device, or by whoever is standing behind them.
So the Dutch and English trees of your site are frequently not two markets. They are two doors into one room, and the figures for each describe a fraction of one visitor rather than the whole of two.
One buyer, two doors, one meaningless average
Consider what an average position across a whole property means when the property holds both trees. Half the rows come from Dutch queries, half from English ones, and the halves behave differently for reasons unrelated to page quality. English queries here are shorter and far more contested, because the competing results are not limited to companies operating in this country. Dutch queries are narrower, so the same page sits higher. Blend them and the number moves whenever the ratio shifts, in weeks when nothing changed.
The practical damage is that the blended figure hides which half converts. A hospitality group whose guests booked from abroad and arrived speaking English needs to know whether the English pages carry the enquiries. If the Dutch pages take most of the clicks and almost none of the bookings, that is a finding. Blended, it is invisible.
- The ratio moves on its own. Seasonal demand, a conference in town, a launch at a group office abroad — each shifts the mix without either tree changing.
- Competition is not symmetrical. An English query for the same intent has more and stronger results behind it, which caps the position your page can reach.
- Intent differs by language for one person. People research in English and sign in the language of the contract, so the halves sit at different points of one journey.
- The obvious remedy is expensive. Publishing everything twice doubles what must be maintained and crawled, and adds no demand that was not already there.
What the panel pulls in, and how fresh any of it is
The Search Console side of the analytics environment Semalt now runs is eight views rather than one report with filters: a dashboard covering every verified site at once — clicks, impressions, rate, trend lines and health flags — then a per-site overview and separate views for keywords, pages, devices, countries and traffic. Background workers keep the data synchronised, so a dashboard is current when opened.
The keyword view is where most people live: position history per term, click trends, and device and country splits attached to each term rather than reported separately. That detail matters — a term that looks stable in aggregate is often two stories on desktop and on mobile.
The record of what already happened
For anyone who needs to know which queries produced which clicks, and on what kind of device.
- Dashboard across all verified sites. Clicks, impressions, rate and trend in one place, with health flags where attention is needed.
- Keywords with their history. Position over time and click development, plus the device and country breakdown per term.
- Pages, devices, countries, traffic. Four cuts of one period, so a finding in one can be checked against another.
- Ranges with sensible defaults. Twenty-eight and ninety days preset, with the two-day delay in the source data accounted for.
Where to cut the data so the languages come apart
The instinct is to split by country: the Netherlands in one column, everywhere else in the other. For a genuine export business that works. Here it does not, because the English half of your audience is sitting at a desk in this country. A regional marketing team searches in English from an office fifteen minutes away and lands in the domestic column like everyone else.
The cut that separates the halves is the page tree. The pages view distinguishes what lives under the English path from what lives under the Dutch one, and the queries behind those impressions can then be read per tree. Country becomes a second question asked inside each half — of those who used the English pages, how many were abroad — not a substitute for the language question.
| How you cut it | What it answers well | What it hides |
|---|---|---|
| By country | Whether demand exists outside the Netherlands at all | The English-speaking audience working inside the country |
| By page tree | Which language version people actually landed on | Queries that had the alternative version served instead |
| By device | Where the reading happens and how the rate differs | Nothing about language on its own |
| By query | The exact wording, in whichever language it was typed | Terms identical in both languages, such as brand names |
| Tree crossed with country | Whether the English pages serve locals or foreign visitors | Little, but it needs two views read side by side |
The last row is where the useful readings come from, and where the country and device heatmaps earn their place: a dense grid is quicker to scan than five hundred sorted rows.
Entries and exits tell you more than a position average
A position average is a state. What you need is a movement, and the keyword dynamics view is built for that: it tracks which terms entered the top three, top ten or top thirty during a period, and which fell out of those bands. Read as arrivals and departures, a quarter becomes legible in about a minute.
It also guards against a trap bilingual sites fall into. Suppose the overall average improves by four tenths of a position — progress, on the face of it. Underneath, eleven informational terms in one tree entered the top thirty, cheap movement in a thin field, while three commercial terms in the other dropped out of the top ten. The arithmetic improved and the visibility that pays got worse. Only entries and exits show it.
Departures from the top ten
The most expensive movement on the board, because click volume falls away sharply below that band.
- Check whether the losing page sits in the mirrored tree
- Check whether your other version now ranks instead
Arrivals in the top thirty
Encouraging, but rarely worth anything in traffic yet. Useful as an early signal that a topic is taking.
- Group them by theme rather than counting
- Note which tree produced them
Terms swapping trees
One query served first by the Dutch page and later by the English one. The clearest symptom of internal competition.
- Decide which version should own the query
- Give the other a different job
Single-impression terms
Long tails that surfaced once. They inflate keyword counts and say nothing about direction.
- Filter them out before reporting
- Revisit only if a cluster forms
What live results-page tracking adds that Search Console cannot
Search Console describes what happened to you. It says nothing about the queries where you do not appear, and nothing about who does. That is the gap rank tracking fills, and the SERP tracking part of the panel holds six views: a ranking score across sites, average position and keyword count, the strongest terms by search volume with their current position, the best-performing pages, a competitor view, and a global ranking across the portfolio with a twenty-eight-day trend curve.
The competitor view behaves differently in a bilingual city. It lists the domains contesting your keyword space with their domain authority and the number of terms you share. Run it on your Dutch terms and you get local firms and Dutch-language publishers. Run it on the English terms for the same intent and a different set appears — international vendors and large content sites with no presence here, above you for a query your buyer typed down the road.
The results page as it stands today
For the questions that begin with "who else is there" rather than "how did we do".
- Score, average and count in one row. A ranking score across sites, the average position and how many terms are followed.
- Strongest terms and best pages. Search volume set against current position, so effort goes where the ceiling is highest.
- Competitors with two numbers attached. Domain authority and shared keyword count, which together say whether a rival is really in your space.
- Portfolio-wide ranking. One global figure across every domain you hold, with a twenty-eight-day trend behind it.
Reading a fictional scale-up off the table
The figures below are invented for the argument — no client data is involved — but the shape turns up regularly. A company of sixty, offices here, half the staff not Dutch-speaking, selling to other businesses. Both trees carry the same eighteen pages. Twenty-eight days, read per tree.
| Reading | Dutch tree | English tree | What it suggests |
|---|---|---|---|
| Impressions | 41,000 | 18,000 | Dutch demand is larger in raw volume |
| Clicks | 1,150 | 760 | The gap narrows considerably |
| Click-through rate | 2.8 % | 4.2 % | English impressions are better matched |
| Average position | 11.4 | 16.9 | A harder field, as expected |
| Enquiries attributed | 9 | 14 | The smaller tree carries the commercial weight |
| Terms in both top-twenty lists | 13 of 20 | Heavy internal overlap | |
Blended, this company reports 59,000 impressions, 1,910 clicks, a rate of 3.2 % and an average position of 13.1. All four are true and none supports a decision. The per-tree reading supports three: the English pages are worth more per visit despite ranking lower, the Dutch pages win impressions they do not convert, and thirteen shared terms out of twenty means the versions keep meeting on one results page.
Site tags, shared access and the view across clients
Few of the organisations described above hold a single domain. A regional marketing team reports to a group office abroad and looks after country sites built at different times by different agencies. A studio in the canal belt runs a dozen clients and would rather hand over a login than assemble another monthly slide. Both need figures grouped without being merged.
The panel is multi-tenant for that reason: Google account groups can be linked, individual sites shared with another email address without handing over everything, and site tags act as a global filter, so a tag such as the English trees reshapes every view at once rather than being reapplied per report.
- Grouping without merging. A tag narrows the environment to a subset — what you want when two domains belong to one client but not to one report.
- Selective sharing. Access is granted per site to a specific address, which suits a team that should see its own country and nobody else's.
- One global ranking figure. Across the portfolio, useful because it is coarse: it belongs in the quarterly summary, not the weekly.
- Export sized to purpose. Table formats hold a working volume of rows; the rendered document is deliberately far smaller.
| Format | Row ceiling | What it is for |
|---|---|---|
| CSV | 10,000 | Working file for a spreadsheet or a further cut |
| JSON | 10,000 | Feeding another system or a warehouse |
| 250 | Rendered on the server for circulation and signature |
The report builder is configurable and takes your own logo and colours — less for the agency's pride than for the client who must recognise the document six months later. Time series, metric cards, sortable tables of fifty to two hundred rows per page, sparklines and heatmaps are the vocabulary on offer. The neighbouring technical threads run through the English articles on this site.
Three things the split figures let you decide
Reports that change nothing are a cost. Once the trees can be read apart, three decisions open up, and they are the whole return on the measurement work.
Which tree gets the next page
Not both by default. Where a topic has demand in one language and thin, contested demand in the other, one good page beats two adequate ones.
- Base it on impressions per tree, not on tidiness
- Treat an asymmetrical site as a normal outcome
Which terms enter the campaign
The keyword pool draws on Search Console, live results-page data and your own seed terms; every candidate is released, rejected or parked one at a time.
- Reject terms your other tree already owns
- Park terms with no page behind them yet
The third decision concerns the questions you ask rather than the pages you write. The Stream assistant in the My SEO area sits on real project data: a router model decides per question which blocks to load — Search Console, results-page data, campaign, custom — pulling between none and three by relevance. Answers stream token by token, twenty messages of context are kept, and lists of keywords or URLs go in as a batch. Ask per tree and it answers per tree; ask a blended question and the answer is blended.
Around this sits the campaign layer, where readings turn into placements and on-site changes. The entry tier runs at 149 USD per month per domain and keeps the cycle going unsupervised; the higher tier at 500 USD adds manual term selection, directed link placement with a target authority, and a review step before changes go live. First measurable movement typically shows after four to eight weeks. Billing is per domain, so a Dutch and an English site on two domains means paying twice — one more argument against mirroring for symmetry's sake. That split is set out in the campaign automation section of the panel.
Frequently asked questions
Should we simply publish everything in both languages?
Only where both languages have demand for that topic. A full mirror doubles what has to be written, maintained and crawled, and where one person reads both it often just puts two of your pages in front of one searcher. Mirror the pages carrying commercial intent and let the rest be asymmetrical.
How do we tell which language a visitor actually used?
Through the query itself, read per page tree. The country split will not do it, because most of the English-speaking audience here is physically in the Netherlands. Separate the trees in the pages view, then look at the terms that produced impressions for each. Brand searches and terms spelled identically in both languages stay ambiguous.
Why does our English average position look so much worse?
Usually because the field is larger. English queries for the same intent attract results from companies with no connection to this market, so the ceiling is lower. Judge the English tree against its own history and its own rate, not against Dutch numbers won in a smaller pool.
How often is it worth looking at any of this?
Keyword dynamics monthly, the competitor view quarterly, the dashboard whenever something feels off. Weekly position readings mostly produce noise, and with a two-day delay a seven-day window holds five usable days. For a new campaign, four to eight weeks is the earliest movement means anything.
Can a client see their own figures without seeing everything?
Yes — individual sites can be shared with a specific email address instead of granting access to the whole account. With site tags this is the usual arrangement for agencies and regional teams: each party sees its own domains, and whoever holds the portfolio keeps the view across all of them.
Does any of this replace analytics on the site itself?
No. Search Console and results-page tracking describe what happens before the click. What someone does after arriving stays a separate measurement, and a report claiming to cover both from search data alone is guessing at the second half.
The measurement problem here is smaller than it looks and more consequential than it seems. Nobody needs a new methodology; they need to stop averaging over a population of one. Cut by page tree, read arrivals and departures rather than states, check the competitor field per language, and accept an asymmetrical site because the demand behind it is asymmetrical.
To see the shape of your own split, connect the property and read the two trees apart in the Semalt dashboard. That count of shared strong terms settles the mirroring argument faster than a quarter of discussion.
Hulp Nodig met Uw SEO?
Onze Amsterdam SEO-experts kunnen u helpen deze strategieën te implementeren en uw rankings te verbeteren.
Gratis Consult Aanvragen