Is GEO just SEO? What the Dutch market data shows

Largely yes. Google states in its own documentation that there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisation necessary. In an independent replication, C-SEO Bench (2025), classic SEO was 7.7 times more effective on retail queries than the best-performing rewriting technique. What genuinely differs is not the craft but the questions — and how you find out which of them you are missing.

This piece is written in English because almost everything published about generative engine optimisation is. The measurements in it are Dutch, and as far as we can tell they are not available anywhere in English.

What the evidence actually says

Start with the vendor. Google's page AI features and your website sets two conditions for a page to be shown as a supporting link in AI Overviews or AI Mode: it must be indexed, and it must be eligible for a snippet. It then says, in as many words, that there are no additional technical requirements and no other special optimisations necessary. That is the company that operates by far the largest AI search surface describing its own system.

Then the research. The best-known GEO paper, presented at KDD in 2024, reported visibility gains of up to 40% from adding citations, quotations and statistics. That 40% appears in nearly every sales deck in this market. It was measured as share of the answer text within a fixed source set — not as organic discoverability.

In 2025 that work was repeated on a stricter design: over 1,900 questions and more than 16,000 documents across six domains, two language models and nine techniques. Three of 54 technique-by-domain combinations produced a statistically significant improvement. For question answering, none did, and on the second model none did either — three out of 108. In the retail test, with GPT-4o-mini as the answer model, simply moving the document higher in the model's context, which is what classic SEO does, produced a rank improvement of 2.77 against 0.36 for the best rewriting technique — a method the authors added themselves; the best of the original GEO techniques reached 0.11. The authors write that GEO methods should be seen as a complement to traditional SEO, not a replacement.

There is counter-evidence and it deserves a fair hearing. A March 2026 paper from a Japanese team reports a 17.3% citation improvement from structural changes across six engines, with p<0.001 over 200 articles and 377 real queries. We weigh it more lightly for three reasons: it is a single paper without a limitations section, the evaluation runs on a benchmark corpus rather than organic discovery, and a critical survey of 45 GEO studies places this class of work in the lower tiers of its evidence ladder — fixed context window, synthetic ranker, or a language model as judge. That same survey concludes that no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability or on downstream behaviour.

What is different, and it is not the writing

Here is where the Dutch data comes in, and where the answer stops being "it's all the same".

We measure the buying questions of our clients against ChatGPT, Gemini, Perplexity, Google AI Overviews and Claude, several times a week, with a frozen question set. Between 3 July and 7 September 2026 that produced 34,815 error-free answers containing 320,681 source citations. Two findings from that set matter for this question.

A company's own domain accounts for 2.2% of everything AI assistants cite.

The other 97.8% comes from 18,996 different domains, of which 8,155 appeared exactly once. The single most-cited domain reaches 1.42%.

That first number reframes the whole debate. Improving your own pages is worth doing, but it addresses roughly one fiftieth of the material an answer is assembled from. The rest sits on comparison sites, industry overviews, institutional pages and video. In classic SEO your own page is the unit of competition; here it is one of nineteen thousand.

The second finding is about where you are visible at all. On identical questions, in identical weeks, with 1,600 measurements each, our clients were cited in 26.4% of Perplexity measurements, 22.6% of Gemini's, and 2.2% of ChatGPT's. That is a factor of twelve between two systems answering the same question in the same week. Anyone reporting one blended "AI visibility score" is averaging across a gap that large.

The Dutch market is loud in marketing and quiet in search

One more measurement, because it explains why this market feels bigger than it is. These are the Dutch monthly search volumes for the category's own terms, from our own keyword data:

Dutch search term	Monthly searches	Cost per click
aeo	1,300	€4.81
generative engine optimization	590	€13.90
geo optimalisatie	590	€9.50
geo seo	590	€6.64
geo vs seo	170	€3.74
answer engine optimization	140	€14.76
aeo tools	30	€15.78

What is not small is what advertisers pay for them: around €13.90 cost-per-click on generative engine optimization, €14.76 on answer engine optimization, €15.78 on aeo tools. A category with a few hundred monthly searches and fifteen-euro clicks is a category where supply has arrived well ahead of demand.

For a reader outside the Netherlands the useful generalisation is this: treat the size of the conversation about GEO as no evidence at all for the size of the market for it, and ask any vendor for the search data in your own language before you believe a market-size claim.

So what should you actually do

If you are already findable in Google and considering a separate GEO content track, do not start by rewriting. Start by establishing which questions you are absent from and which pages are retrieved in your place. That is measurement work, not writing work, and it determines whether anything needs rewriting at all.

If you are not findable in Google, the order is even clearer. There is no measured route by which AI visibility appears without discoverability, for an ordinary business in an ordinary market. Anyone selling you an AI track while your basics are broken is selling you the second step.

And whichever of the two you are: check that the crawlers can reach you before you touch a single sentence. Vercel and MERJ analysed real server logs in November 2024 — 569 million GPTBot requests, 370 million from Claude — and found no evidence that these crawlers execute JavaScript. Googlebot and AppleBot do. If your text only appears after JavaScript has run, ChatGPT and Claude see an empty shell, and the quality of your writing is beside the point.

Where we stand

We are a managed service, not a tool. We measure 100 buying questions per client across four AI systems with a frozen question set, three measurement rounds a week, reported per system and never merged into one figure. What we find in the gaps we write ourselves and publish on your site after you approve it.

The Dutch-language version of this argument, with the same sources, is at Moet ik voor GEO SEO andere content maken?. Pricing is on the prijzenpagina, and how the measurement works on AI-zichtbaarheid laten meten.