Put the same question to an assistant twice — once in Dutch, once in English — and two different shortlists of sources come back. Neither of them is wrong, and neither of them is a ranking.
Everything a rank tracker knows rests on one assumption: there is an ordered list, and your page holds a place in it. Above a growing share of questions that list is now preceded by a written answer with a handful of sources attached, and those sources have no order. They are a set. You are in it or you are not, and the same question an hour later may return a different one.
In this city that comes with a second complication most published advice ignores, because most of it assumes a company has one language. Here the buyer has two, uses both, and switches without noticing. So the question is not only whether you are drawn on, but whether you are drawn on in the language the person happened to be thinking in.
A set of sources is not a list of positions
An employment and immigration practice on the Zuidas, the sort that walks a foreign employer through its first Dutch hires, has held good positions on a dozen procedural questions for years. Those positions have not moved. What changed is that some of the people who used to arrive there now read a summary drawing on three or four sources and open nothing. The practice did not lose a rank. It lost the step between the rank and the visit.
What a tracker records
A position in an ordered list, repeatable on demand, checkable by anyone with the same tool and the same location.
- Stable enough to chart weekly
- Wrong in ways you can identify
What the answer layer produces
An unordered set of sources, assembled fresh, varying between accounts and between one hour and the next.
- Presence or absence, with no middle
- No interface reports it back to you
Three properties make this layer behave unlike anything a tracker was built for.
- The output is not stable. The same wording produces different sources on a different day, from a different account, on a different device. There is no correct answer to check.
- There is no order to occupy. Being named second rather than fourth means very little. Being named at all is the whole outcome, which makes it close to binary.
- Nobody sends you a report. No console says how often a domain was used as a source. The absence is structural, not a gap somebody fills next quarter.
- The click, when it comes, is late. A reader who follows a source after a full summary is checking or comparing. Fewer visits, each further along.
The Dutch question and the English question have different sources
Take one intent and phrase it both ways: what a foreign employer must arrange before a first hire in the Netherlands. Asked in Dutch, the answer is assembled from Dutch-language material — trade publications, official pages, practices writing for a domestic readership. Asked in English, it pulls in international explainers, relocation portals and pages published from anywhere at all, by companies with no presence here and no ability to act on the question.
Both answers are defensible. They draw on different pools of writing, and the pools genuinely differ in what they hold. The awkward consequence is that a firm which is a natural source in one pool can be absent from the other while believing the topic is covered, because a page exists in both languages and both rank.
| Asked in | Sources tend to be | Your standing in that pool | What it changes |
|---|---|---|---|
| Dutch | Domestic publishers, official material, local practices | Strong if you write natively for that readership | You are a plausible source by default |
| English | International explainers, portals, global content sites | Weak if your English pages are translations | You compete with everyone, not with the street |
| English, with the location named | A narrower field, closer to the domestic one | Recoverable with specific local detail | The realistic target for most firms here |
| Either, with your brand named | Your own pages, plus whoever describes you | Usually solid, and usually already true | Little — this is the question you win anyway |
The third row is the practical opening. An English page carrying detail nobody outside this market could write — the sequence, the bodies involved, the timing, the local exceptions — is not competing with a global explainer at all. A translated page of generalities is, and it loses, because those were already available in English from a hundred other places.
Nothing here produces a number you can audit
A great deal of confident nonsense is being sold on this subject, so the limits belong early rather than in a footnote. The layer has no reporting interface. Nothing reports how often a domain was used as a source, in which answers, for which questions.
What can be done is narrower and still worth doing. A fixed panel of questions, asked on a schedule in both languages with the returned domains written down, produces a record that is genuinely yours. It is a sample rather than a census, and read consistently over six months it tells you about direction even where no single round tells you about standing.
Fixed wording, or it means nothing
Change the phrasing between rounds and you have measured your own phrasing rather than the market.
- Revise the list once a year
- Keep retired questions on the sheet
Both languages, every round
The pair is the point. One language on its own hides exactly the asymmetry you are looking for.
- Same intent, phrased natively in each
- Never a translation of the other
Record the whole set
Who else appears is the more useful half. A domain present in every answer is a content model worth studying.
- One column per round
- Your own row among the others
Read the trend, never the reading
One round is noise. Six rounds is a direction, and a direction is what a decision needs.
- Date every round
- Compare like with like only
What the panel offers instead of a counter
Given that the count does not exist, the useful question is what a tool can honestly provide. The generative research module Semalt built into its panel answers with estimation rather than accounting: six views modelling how a language model is likely to read your market, and where a domain sits inside that reading.
Generative market research
For teams that need a map of the answer layer rather than a scoreboard nobody can verify.
- A competitiveness score with a market circle. Rivals sorted into top tier, middle tier and niche, arranged around your domain rather than listed beneath it.
- Model-generated market context. How a domain is positioned, an estimate of its traffic, and the openings the model considers available.
- Query research with intent classification. Candidate questions grouped by what the person asking them actually wants.
- Pages marked as levers. The addresses worth expanding, or worth linking to from elsewhere on the site.
- Competitor strengths against your gaps. What rivals cover thoroughly, and the subjects nobody in the set has treated properly.
- A global visibility figure. One number for standing in the generative landscape, coarse by design and useful only as a trend line.
None of that counts citations, and the section does not claim to. It is a structured second opinion, produced in minutes, where the alternative is an afternoon of asking questions by hand.
The circle is a map; the score is an opinion with a decimal point
The market circle earns its place by answering a question people ask badly. Nobody needs their exact standing among forty domains; they need to know which handful of names occupy the same conversation, which sit above it, and which are small enough to ignore. Three tiers answers that. A ranked list of forty invites an argument about position eleven.
For a bilingual company the tiering behind that circle repays being run twice, because the tiers differ by language. A branding studio in Noord may sit in the middle tier of a Dutch field made up of other Dutch studios, and in the niche tier of an English field that includes agencies from London and Berlin chasing the same phrasing. Both readings are accurate. Only one is visible if you look once.
Intent classification, and why the split differs by language
The query research view produces candidate questions and sorts them by intent, and that sorting is the valuable part. A question asked to understand something and one asked to hire someone look alike in a keyword list, demand entirely different pages, and are treated differently above it: the first is resolved inside the summary, the second usually sent onward.
Understanding questions
How something works, what a term means, what the sequence is. Answered in the summary, rarely producing a visit.
- Write them for standing, not for traffic
- Judge by whether you appear, not by clicks
Selection questions
Who does this, at what price, with what track record. A summary cannot settle these, so the sources get opened.
- These belong on pages built to convert
- Mirror them in both languages deliberately
Procedural questions
Deadlines, forms, thresholds, which body decides. Asked in Dutch far more often, even by people who work in English.
- Check whether your English tree covers them
- Specific local detail wins here
Comparison questions
One thing against another. Abundant, easy to write for, and dominated by large content sites in the English pool.
- Enter only with first-hand material
- Otherwise a poor use of a writing week
Run the exercise in both languages and the proportions shift. Procedural material clusters in Dutch; comparison and selection material spreads more evenly, with heavier competition on the English side. That asymmetry is a content plan rather than a problem — it says which tree deserves the next long article and which deserves a shorter, sharper page.
The pages the model flags, and what to do about them
Two of the six views produce work rather than understanding: the one marking pages as levers for expansion or internal linking, and the one setting competitor strengths against your gaps. Before acting on a flagged page, ask which tree it sits in and whether the other has a counterpart. Four situations follow, and they lead different ways.
| Situation | What it usually means | The move |
|---|---|---|
| Flagged, only the Dutch version exists | The topic has substance and the English pool has never heard from you | Write the English version natively, with local specifics added |
| Flagged, both versions exist | The page is close to being a source and thin in one place | Deepen the stronger one first; let the other point at it |
| Flagged, and it is a translated stub | The model found the topic, not the page | Rewrite or remove — another translation adds nothing |
| Gap named, nobody covers it | Either a genuine opening or a subject with no demand behind it | Check it against real query data before writing a word |
That last row saves the most time. A gap identified by a model is a gap in what has been written, not in what people ask about; some subjects are uncovered because nobody wants them. Crossing the suggestion against the search analytics views next to it takes a minute and prevents a fortnight of work with no audience.
The linking suggestions need one warning on a mirrored site. Links crossing between the trees scatter the signal instead of concentrating it, and they read badly to someone moved into another language mid-sentence. Keep linking inside each tree and let the switch happen where the reader expects it. The same principle shapes how we plan content work for clients running both.
A monthly rhythm, and the figure that stays out of the deck
The work is small if it is regular and impossible if it is occasional. A rhythm that survives a busy quarter looks roughly like this, and fits inside an hour a month for one domain.
The sampling round
A repeatable procedure producing your own evidence, not somebody else's estimate.
- Ten questions, fixed wording. Five in Dutch, five in English, covering the intents that actually produce enquiries.
- Every domain returned, written down. A column per round, with your own row among the others — the correct amount of attention to give it.
- Panel readings taken the same day. Circle, gaps and visibility figure, noted with date and range so the next round is comparable.
- One sentence of interpretation. What changed and what you think caused it. In six months that column is worth more than the figures beside it.
The recording work is shortened by the Stream assistant in the campaign area, which is bound to the project's own data: a router decides per question which blocks to load, between none and three of them by relevance, and lists of keywords or addresses go in as a batch. Asking which flagged pages have no counterpart in the other tree beats assembling that list by hand.
Exports behave as elsewhere in the panel: a working file for the sheet rather than a rendered document, since this material wants sorting and filtering. The neighbouring pieces in this series sit in the English section of this blog.
Frequently asked questions
Can we find out how often we are used as a source?
Not exactly, and not from any official interface. You can sample: ask a fixed set of questions on a schedule and record which domains appear. That gives a defensible trend over several rounds. Anyone offering an exact count is presenting an estimate without the label.
Why do the Dutch and English answers name different companies?
Because they are assembled from different bodies of writing. The Dutch pool is smaller, local and mostly written by people operating here. The English pool is global, so a publisher with no connection to this market can outweigh a firm ten minutes away.
Does translating our Dutch articles get us into the English answers?
Rarely on its own. A translation carries the same generality into a pool where it is already abundant. What travels is detail only somebody working here could supply — the local procedure, the bodies, the timing, the exceptions. Add that and the page earns a reason to be drawn on.
Is this worth attention before our ordinary rankings are in order?
Generally no. Both layers draw on the same material, and a well structured, well linked and genuinely informative site tends to become a plausible source as a side effect. Fix the ordinary things first, then sample to see where you stand.
How quickly would changes show up?
Slowly and unevenly. Ordinary campaign work shows first measurable movement after roughly four to eight weeks. This layer lags further behind and moves in steps, because it depends on material being read again rather than on a position being recalculated. Six monthly rounds is the shortest honest window.
The short version is uncomfortable and worth saying plainly. This layer matters, it cannot be counted, and the gap between those two facts is where most of the current marketing lives. You can still work on it seriously — by writing material worth drawing on, sampling your own standing on a schedule, and treating every model-produced score as an opinion with a date attached.
For a bilingual company the first step is smaller than any of that: run the pair of questions and see whether both halves of your site exist in the answers at all. To put the estimate beside your own search figures, connect the domain and open the market view in the Semalt dashboard. What it gives you is an opinion — but a dated, repeatable one, which is more than the alternative.
Hulp Nodig met Uw SEO?
Onze Amsterdam SEO-experts kunnen u helpen deze strategieën te implementeren en uw rankings te verbeteren.
Gratis Consult Aanvragen