Measured 22 September 2026, re-measured 24 and 27 September · Netherlands
Twenty questions — eighteen a client actually asks, plus two controls — put to three AI surfaces: Google’s AI Overview, Google’s AI Mode and ChatGPT. Sixty runs, fifty-two of which produced an answer. Twenty-six firms named.
Someone moving to the Netherlands has no local network and no way to judge a Dutch law firm. So they ask a machine, and they ask it long before they type any firm name: how to appeal an IND refusal, what happens when an employer loses recognised sponsor status, whether they need a lawyer at all.
We asked those questions from a clean browser profile that was not signed in to Google, and wrote down who the machines named. Every answer is stored verbatim, so every number on this page can be checked against the text it came from.
On 27 September we put the same twenty questions to the same three surfaces again. Thirty firms were named on the eighteen buyer questions. The three named most often were the same as before: Everaert Advocaten (11 mentions), Kroes Advocaten (10) and AILF Agayev Immigration Law Firm (10).
Below them the list kept moving. Compared with 24 September, 48% of the names ChatGPT gave rest on one of the two measurements and not the other; on Google AI Mode it was 48%, on AI Overview 37%. A firm named once or twice can be on the list one week and gone three days later.
One change of method: on 27 September Google’s answers were read in an ordinary browser window, still not signed in to Google, because Google had started blocking our automated one. ChatGPT was measured exactly as before, and it moved just as much. The rest of this page describes the measurement of 22 September.
Twenty-six firms were named at least once across the eighteen buyer questions, in 97 mentions. Seven of them were named on all three surfaces. The top three alone account for 35 of the 97.
| Firm | Answers naming it | Surfaces |
|---|---|---|
| Kroes Advocaten | 14 | all three |
| Everaert Advocaten | 12 | all three |
| AILF Agayev Immigration Law Firm | 9 | all three |
| Law & More | 7 | all three |
| Mynta Law | 7 | all three |
| Adam & Wolf Immigration Lawyers | 6 | AI Mode, ChatGPT |
| Koevoets Immigration Lawyers | 5 | AI Mode, ChatGPT |
| Orion Immigration Law | 5 | AI Mode, ChatGPT |
| Russell Advocaten | 4 | AI Overview, AI Mode |
For scale: the official register of the Dutch Bar lists 642 lawyers across 411 firms in the four immigration-related practice areas — immigration law, asylum, immigration criminal law and immigration detention. Twenty-six names came up. Being in the register and being in the answer are different things.
Of the 86 domains Google’s AI Overview cited across thirteen of these questions, only 29 also appeared in the organic results on the same page — 34%.
On three questions, not one cited domain was in the organic list: “Immigration lawyer or immigration consultant in the Netherlands”, “Who handles both the visa and the tax side for expats moving to the Netherlands”, and the Dutch-language “Wie is een goede advocaat vreemdelingenrecht in Amsterdam”.
If an SEO report says your rankings are healthy, it is describing a different system from the one that answers your client’s question.
| Question type | Google AI Overview | Google AI Mode | ChatGPT |
|---|---|---|---|
| “Who should I hire?” (5) | never appears at all | names firms in 4 | names firms in 5 |
| “I have this problem” (6) | appears every time, names firms in 4 | names firms in 3 | names firms in 1 |
| “How do I choose?” (4) | appears 3 times, names firms in 3 | names firms in 4 | names firms in 2 |
Google’s AI Overview does not appear at all on hiring questions — not a weak answer, no answer. It did the same on a London control question, so this is about the type of question, not about the Dutch market. ChatGPT is the mirror image: it hands out firm names on hiring questions and almost none on problem questions, where it answers with the IND and the Juridisch Loket instead.
A firm invisible on one surface can be prominent on another. Koevoets Immigration Lawyers and Orion Immigration Law were named five times each, and never once inside Google’s AI Overview. Measure one surface and you lose whole firms.
On five problem-type questions — an IND refusal, a decision taking too long, an employer losing recognised-sponsor status, a permit not renewed, a family visa denied — ChatGPT named no law firm at all. It answered with government bodies.
Those are the questions asked by someone with an urgent problem and a reason to pay. Right now nobody occupies them.
One of our two control questions was the Dutch-language “Wie is een goede advocaat vreemdelingenrecht in Amsterdam”. AI Mode named four firms in it and ChatGPT five, while the AI Overview did not appear at all — and four firms in this measurement are named only in the two control questions, never in an English-language buyer answer.
The same market, the same city, a different language, a largely different list. If your clients arrive in English, the Dutch-language result tells you very little about your own visibility — and the reverse holds too.
On 24 September we put the same twenty questions to the same surfaces again. Two days, nothing changed by us, nobody told to do anything.
On Google AI Mode — see below for why only that one — and counting the eighteen buyer questions, as every headline on this page does, 18 firms were named on 22 September and 20 on 24 September. Two left the list, four arrived: 27% of the names rest on one of the two measurements and not the other. And on 16 of the 18 questions — 89% — the set of firms that came back had changed.
An earlier version of this page said 41%. That figure counted the two control questions as well, which breaks the rule stated in the method section below: controls stay out of every headline. They moved far more than the buyer questions — the Dutch and the London answers changed almost all of their names — and adding them made the market look more restless than it is.
Why only AI Mode: on 22 September our own capture cut ChatGPT and AI Overview answers off at 1,500 characters. We found that on 24 September and raised the limit. Any difference we reported on those two surfaces would therefore measure our repair, not the market, so we do not report one. AI Mode was never truncated, which is why it can be compared.
Two things follow, and the second one matters more. A single snapshot describes a day, not a position — ours included. And anyone who tells you they moved a firm up owes you this number first: without knowing how much the list moves on its own, an improvement and a coincidence look exactly alike.
We crawled 24 of these firms page by page and counted pages dedicated to a single client question: an IND refusal, the 30% ruling, family reunification.
| Named (13) | Not named (11) | |
|---|---|---|
| Topic pages, median | 4 | 0 |
| Firms with no topic page at all | 0 | 6 |
| Pages in the sitemap, median | 72 | 49 |
The exception is worth as much as the rule. Adam & Wolf Immigration Lawyers was named in six answers with a ten-page website and not one topic page. Whatever is being rewarded, it is not simply the page count.
And we are not going to tell you that writing pages causes the mentions. The named firms also have larger sites — a median of 72 pages in the sitemap against 49 — so size explains part of it, and on 24 firms the whole thing is a correlation. We would rather say that than sell it as a mechanism.
Of the seven firms whose Google card reads “Immigration attorney”, five were named. Of the sixty-two carrying a generic “Law firm” or “Lawyer” card, eleven were. The gap holds on each surface taken separately.
Seven firms is a small group — one firm moves that share by fourteen points — and the obvious objection is that firms which pick the specialist category are firms that specialise anyway. So we treat it as the best available candidate for a test, not as advice. It takes five minutes to change, and it is exactly the kind of single change whose effect can be measured three weeks later against firms that changed nothing.
What was measured. Eighteen buyer questions written before the run, in four groups — who to hire, I have a problem, how to choose, adjacent tax questions — plus two controls, one in Dutch and one about London. Twenty questions, three surfaces, sixty question-answer pairs. 22 September 2026, from Amsterdam, in a browser profile not signed in to Google. Headline counts on this page use the eighteen buyer questions; the two controls are reported separately, because the Dutch answers name firms that the English ones never do, and would distort every total they entered.
What “named” means. The firm’s name appears in the text of the answer. It is checked in both directions: every recorded mention is confirmed against the stored answer, and every firm-like name in the answers is checked against our list, so a firm we failed to recognise surfaces as a gap rather than as a silence. Ordinary words are excluded — an early version of this measurement made the city of Rotterdam the most-named “firm” in the market, because a portal’s domain looked like a name.
Names we did not recognise are chased down, not dropped. Fifteen appeared in the answers that were not on our list. Two turned out to be real firms and have been added — Russell Advocaten, named in four answers, and WS Advocaten in one. Nine were not firms at all: a portal, the bar's own specialist association, the IND, and phrases like “Budget Attorneys” that a pricing table made look like a name. We then went further and loaded the whole bar register into the matcher, because a list of 154 association members is not the market. Each time, the stored answers were counted again from the saved text.
Widening the list exposed four mistakes of our own, and they are worth stating because they changed this page after it was published. Adding Russell Advocaten taught the matcher the bare word “Russell”, and the London control promptly credited it with a mention of Charles Russell Speechlys — a different firm in a different country. A firm called Justice Solutions Law taught it “solutions”, which then matched “work visa solutions”. A firm called De Rechter Advocaten taught it “de rechter”, Dutch for “the court”. And one firm on this very page, AILF Agayev, had been credited with six answers on the strength of the phrase “law firm” appearing in them — which dropped it from first place to third.
The rule that now catches all four is one a reader can check: a name is written like a name. A word or phrase counts only where the answer capitalises it, or writes it as a domain. “Kroes” and “lawandmore.eu” pass; “a practice in Amsterdam” and “appeal to the court” do not.
The citation comparison is a separate, earlier snapshot. The 86 AI Overview citations were captured by hand on 21 September across thirteen of these questions; the organic results they are compared against came from a search API on the same set. On a few questions that API returned only seven to nine organic results rather than ten.
One day is one day. AI answers move. This is a photograph, not a verdict, and a firm absent here may be present next month. That is exactly why we re-measure rather than conclude.
How this page counted, and how I count now. This page also counted a firm whose name appeared only as a source link under Google’s answer. From October I count a firm only where the answer writes its name. The October index uses that rule, so some numbers there are lower than here.
We do not claim to know why. This page reports where firms are absent. It does not explain the cause, and anyone who tells you they know the cause of an AI mention is guessing.
Who is not on the list. A firm not named here was not named in any of these fifty-two answers. That is not a judgement about the firm, its work or its clients.
Corrections welcome. If your firm is listed wrongly, write to danil@sysenko.com and the page will be corrected.
Then it was not named in any of these fifty-two answers. That is all it means — and it is still worth knowing which questions it was absent from, because absence on “who should I hire” is a different thing from absence on “my permit was refused”, where the assistants name almost nobody at all.
I will run the same twenty questions for your firm and send you what came back: the questions, the firms named on each one, and the text of the answers. No charge, no call, no slide deck — one email, and you can check every line of it yourself.
If after reading that you want the longer version — where exactly you are absent, who comes up instead on the questions your clients ask, and up to three changes I would make first — that is a quick check, €150, paid after you have read it. If you then make one change, I can measure whether it moved: the same questions before and about four weeks after, against a threshold written down in advance, also €150.
What I will not do is promise that anything moves. These systems are three years old and they change monthly; nobody can honestly promise a position in them. What I can promise is that your measurement will be the same one described on this page, and that if nothing changed after a change was made, the next report will say so in the first line.